scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-06 08:16:43 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 78

Merge branch 'litellm_internal_staging' of https://github.com/BerriAI/litellm into litellm_fix_guardrail_put_inmemory_sync

Browse source 
# Conflicts:
#	litellm/proxy/guardrails/guardrail_registry.py
This commit is contained in:
mateo-berri 2026-09-02 17:35:06 -07:00
commit 0fb83995dd
1014 changed files with 77743 additions and 11230 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

8
.circleci/config.yml
View file

@ -112,10 +112,10 @@ commands:
node --version
npm --version
install_rust:
description: "Install pinned rustup (1.28.2) and Rust toolchain (1.97.1) with checksum verification. Adds ~/.cargo/bin to PATH. Run this before any `uv sync` or `uv build` of the workspace: the root package builds litellm-rust through maturin, and on an image without cargo maturin fetches an unpinned rustup and a floating toolchain by itself."
description: "Install pinned rustup (1.28.2) and Rust toolchain (1.98.0) with checksum verification. Adds ~/.cargo/bin to PATH. Run this before any `uv sync` or `uv build` of the workspace: the root package builds litellm-rust through maturin, and on an image without cargo maturin fetches an unpinned rustup and a floating toolchain by itself."
steps:
- run:
name: Install Rust (rustup 1.28.2, toolchain 1.97.1)
name: Install Rust (rustup 1.28.2, toolchain 1.98.0)
command: |
case "$(uname -m)" in
x86_64)
@ -135,7 +135,7 @@ commands:
"https://static.rust-lang.org/rustup/archive/1.28.2/${RUSTUP_TRIPLE}/rustup-init"
echo "${RUSTUP_SHA256} /tmp/rustup-init" | sha256sum -c -
chmod +x /tmp/rustup-init
/tmp/rustup-init -y --no-modify-path --profile minimal --default-toolchain 1.97.1
/tmp/rustup-init -y --no-modify-path --profile minimal --default-toolchain 1.98.0
rm -f /tmp/rustup-init
echo 'export PATH="$HOME/.cargo/bin:$PATH"' >> "$BASH_ENV"
export PATH="$HOME/.cargo/bin:$PATH"
@ -300,7 +300,7 @@ jobs:
if ($rustupActual -ne $rustupExpected) {
throw "rustup installer hash mismatch: expected $rustupExpected got $rustupActual"
}
& $rustupInit -y --profile minimal --default-toolchain stable
& $rustupInit -y --profile minimal --default-toolchain 1.98.0
if ($LASTEXITCODE -ne 0) {
exit $LASTEXITCODE
}

23
.github/actions/cache-cargo-build/action.yml vendored
View file

@ -4,17 +4,16 @@ description: >-
so only the first job on a given Cargo.lock compiles the bridge from scratch.
litellm builds through maturin, which compiles litellm-rust/crates/python-bridge
in release mode before it can produce a wheel. `uv sync` therefore pays a full
build in every job that installs the workspace: measured at 2m40s per unit shard
on 2026-08-21, more than the whole unit tier spends running tests. Nothing caught
it, because the uv cache holds wheels uv downloads rather than wheels it builds,
and a path dependency whose source moves every commit could never hit that cache
anyway. Cargo rebuilds only what changed when its target directory survives, so a
warm job pays for the bridge crate alone.
in the dev profile for editable installs. `uv sync` therefore pays a full build
in every job that installs the workspace. Nothing caught it, because the uv cache
holds wheels uv downloads rather than wheels it builds, and a path dependency
whose source moves every commit could never hit that cache anyway. Cargo rebuilds
only what changed when its target directory survives, so a warm job pays for the
bridge crate alone.
The key namespace is separate from test-rust.yml's. Both cache the same directory,
but that workflow fills it with debug and clippy artifacts, which a release build
cannot reuse, and a shared key would let whichever ran first deny the other a save.
The key namespace is separate from test-rust.yml's check and release caches. They
cache the same directory for different workloads, and a shared key would let
whichever ran first deny the others a save.
runs:
using: composite
@ -26,6 +25,6 @@ runs:
~/.cargo/registry
~/.cargo/git
litellm-rust/target
key: ${{ runner.os }}-cargo-release-${{ hashFiles('litellm-rust/Cargo.lock') }}
key: ${{ runner.os }}-maturin-dev-${{ hashFiles('litellm-rust/Cargo.lock') }}
restore-keys: |
${{ runner.os }}-cargo-release-
${{ runner.os }}-maturin-dev-

230
.github/scripts/close_duplicate_issues.py vendored
View file

@ -1,230 +0,0 @@
#!/usr/bin/env python3
"""
Detect and close duplicate GitHub issues using title similarity.
Modes:
--scan Compare all open issues against each other (batch)
--issue-number N Check a single issue against older open issues
Requires the `gh` CLI to be authenticated.
"""
import argparse
import difflib
import json
import re
import subprocess
import sys
def normalize_title(title: str) -> str:
"""Strip common prefixes, lowercase, and collapse whitespace."""
title = re.sub(
r"^\[?(bug|feature request|enhancement|question|docs)[:\]]?\s*",
"",
title,
flags=re.IGNORECASE,
)
return " ".join(title.lower().split())
def gh(*args: str) -> str:
"""Run a gh CLI command and return stdout."""
result = subprocess.run(
["gh", *args],
capture_output=True,
text=True,
check=True,
)
return result.stdout
def fetch_open_issues(repo: str | None) -> list[dict]:
"""Fetch all open issues (excluding PRs) via gh api --paginate."""
if repo:
endpoint = (
f"repos/{repo}/issues?state=open&per_page=100&sort=created&direction=asc"
)
else:
endpoint = "repos/{owner}/{repo}/issues?state=open&per_page=100&sort=created&direction=asc"
cmd = ["api", "--paginate", endpoint]
raw = gh(*cmd)
# gh --paginate concatenates JSON arrays, so we may get multiple arrays
issues = []
for line in raw.strip().splitlines():
line = line.strip()
if not line:
continue
parsed = json.loads(line)
if isinstance(parsed, list):
issues.extend(parsed)
else:
issues.append(parsed)
# Filter out pull requests (they also appear in the issues endpoint)
return [i for i in issues if "pull_request" not in i]
def close_as_duplicate(
issue_number: int, duplicate_of: int, repo: str | None, dry_run: bool
) -> None:
"""Close an issue as duplicate of another, adding a comment and label."""
repo_args = ["--repo", repo] if repo else []
if dry_run:
print(
f" [DRY RUN] Would close #{issue_number} as duplicate of #{duplicate_of}"
)
return
# Add comment
comment_body = (
f"Closing as duplicate of #{duplicate_of}.\n\n"
"If you believe this is not a duplicate, please reopen and add context "
"explaining how this differs."
)
gh("issue", "comment", str(issue_number), "--body", comment_body, *repo_args)
# Add label
gh("issue", "edit", str(issue_number), "--add-label", "duplicate", *repo_args)
# Close with not_planned reason
gh(
"api",
f"repos/{repo or '{owner}/{repo}'}/issues/{issue_number}",
"-X",
"PATCH",
"-f",
"state=closed",
"-f",
"state_reason=not_planned",
)
print(f" Closed #{issue_number} as duplicate of #{duplicate_of}")
def find_duplicate(
issue: dict, candidates: list[dict], threshold: float
) -> dict | None:
"""Return the first candidate whose normalized title is above threshold."""
norm = normalize_title(issue["title"])
for candidate in candidates:
if candidate["number"] == issue["number"]:
continue
cand_norm = normalize_title(candidate["title"])
ratio = difflib.SequenceMatcher(None, norm, cand_norm).ratio()
if ratio >= threshold:
return candidate
return None
def scan_all(
issues: list[dict], threshold: float, repo: str | None, dry_run: bool
) -> int:
"""Compare every issue against all older issues. Returns count of duplicates found."""
# Sort oldest first
issues.sort(key=lambda i: i["number"])
closed_count = 0
for idx, issue in enumerate(issues):
older = issues[:idx]
if not older:
continue
dup = find_duplicate(issue, older, threshold)
if dup:
ratio = difflib.SequenceMatcher(
None,
normalize_title(issue["title"]),
normalize_title(dup["title"]),
).ratio()
print(
f"#{issue['number']}: \"{issue['title']}\"\n"
f" -> duplicate of #{dup['number']}: \"{dup['title']}\" "
f"({ratio:.0%} similar)"
)
close_as_duplicate(issue["number"], dup["number"], repo, dry_run)
closed_count += 1
return closed_count
def check_single(
issue_number: int,
issues: list[dict],
threshold: float,
repo: str | None,
dry_run: bool,
) -> bool:
"""Check a single issue against all older open issues. Returns True if duplicate found."""
target = None
for i in issues:
if i["number"] == issue_number:
target = i
break
if target is None:
print(f"Issue #{issue_number} not found among open issues.")
return False
older = [i for i in issues if i["number"] < issue_number]
dup = find_duplicate(target, older, threshold)
if dup:
ratio = difflib.SequenceMatcher(
None,
normalize_title(target["title"]),
normalize_title(dup["title"]),
).ratio()
print(
f"#{target['number']}: \"{target['title']}\"\n"
f" -> duplicate of #{dup['number']}: \"{dup['title']}\" "
f"({ratio:.0%} similar)"
)
close_as_duplicate(issue_number, dup["number"], repo, dry_run)
return True
print(f"#{issue_number}: no duplicate found above threshold {threshold}")
return False
def main() -> None:
parser = argparse.ArgumentParser(
description="Detect and close duplicate GitHub issues"
)
mode = parser.add_mutually_exclusive_group(required=True)
mode.add_argument("--scan", action="store_true", help="Scan all open issues")
mode.add_argument("--issue-number", type=int, help="Check a single issue number")
parser.add_argument(
"--threshold", type=float, default=0.85, help="Similarity threshold (0-1)"
)
parser.add_argument(
"--close",
action="store_true",
help="Actually close duplicates (default is dry-run)",
)
parser.add_argument(
"--repo", type=str, help="Repository (owner/repo). Auto-detected if omitted."
)
args = parser.parse_args()
dry_run = not args.close
if dry_run:
print("=== DRY RUN MODE (pass --close to actually close issues) ===\n")
print("Fetching open issues...")
issues = fetch_open_issues(args.repo)
print(f"Found {len(issues)} open issues.\n")
if args.scan:
count = scan_all(issues, args.threshold, args.repo, dry_run)
print(f"\nTotal duplicates {'found' if dry_run else 'closed'}: {count}")
else:
found = check_single(
args.issue_number, issues, args.threshold, args.repo, dry_run
)
sys.exit(0 if found else 0) # Always exit 0; finding no dup is not an error
if __name__ == "__main__":
main()

68
.github/scripts/smoke_test_native_wheel.py vendored Normal file
View file

@ -0,0 +1,68 @@
from __future__ import annotations
import subprocess
import sys
import tempfile
import zipfile
from pathlib import Path
from typing import Final
CHILD_SCRIPT: Final = """
from importlib.util import module_from_spec, spec_from_file_location
from pathlib import Path
import sys
native_path = Path(sys.argv[1])
spec = spec_from_file_location("litellm.rust_bridge._native", native_path)
if spec is None or spec.loader is None:
raise RuntimeError("cannot create native extension import specification")
module = module_from_spec(spec)
spec.loader.exec_module(module)
before = module.gil_stats()
if not isinstance(before.get("releases"), int):
raise AssertionError(f"unexpected gil_stats result: {before!r}")
try:
module._panic_for_test()
except BaseException as error:
if type(error).__name__ != "PanicException":
raise AssertionError(f"expected PanicException, got {type(error).__name__}") from error
else:
raise AssertionError("Rust panic returned without raising")
after = module.gil_stats()
if not isinstance(after.get("releases"), int):
raise AssertionError(f"native module unusable after panic: {after!r}")
"""
def main() -> int:
if len(sys.argv) != 2:
sys.stderr.write(f"usage: {Path(sys.argv[0]).name} WHEEL\n")
return 2
wheel: Final = Path(sys.argv[1])
with tempfile.TemporaryDirectory() as temporary_directory, zipfile.ZipFile(wheel) as archive:
native_members: Final = tuple(
member
for member in archive.infolist()
if member.filename.startswith("litellm/rust_bridge/_native.") and member.filename.endswith(".so")
)
if len(native_members) != 1:
sys.stderr.write(f"expected one native extension, found {len(native_members)}\n")
return 1
native_path: Final = Path(temporary_directory) / Path(native_members[0].filename).name
native_path.write_bytes(archive.read(native_members[0]))
result: Final = subprocess.run((sys.executable, "-c", CHILD_SCRIPT, str(native_path)), check=False)
if result.returncode != 0:
sys.stderr.write(f"native wheel smoke test exited with status {result.returncode}\n")
return 1
return 0
if __name__ == "__main__":
sys.exit(main())

282
.github/scripts/verify_linux_native_wheel.py vendored Normal file
View file

@ -0,0 +1,282 @@
from __future__ import annotations
import importlib.util
import os
import re
import subprocess
import sys
import zipfile
from collections.abc import Callable, Mapping, Sequence
from itertools import product
from pathlib import Path, PurePosixPath
from types import MappingProxyType, ModuleType
from typing import Final, Protocol
EXPECTED_PYTHON_TAG: Final = "cp310"
EXPECTED_ABI_TAG: Final = "abi3"
EXPECTED_PLATFORM_TAG: Final = "linux_x86_64"
class CommandRunner(Protocol):
def __call__(
self,
command: tuple[str, ...],
*,
check: bool,
capture_output: bool,
text: bool,
) -> subprocess.CompletedProcess[str]: ...
def _run_command(
command: tuple[str, ...],
*,
check: bool,
capture_output: bool,
text: bool,
) -> subprocess.CompletedProcess[str]:
return subprocess.run(command, check=check, capture_output=capture_output, text=text)
def _dist_info_directory(member: zipfile.ZipInfo) -> str | None:
parts: Final = PurePosixPath(member.filename).parts
if not parts or not parts[0].endswith(".dist-info"):
return None
return parts[0]
def _wheel_metadata_tags(archive: zipfile.ZipFile, members: tuple[zipfile.ZipInfo, ...]) -> tuple[str, ...]:
if len(members) != 1:
return ()
lines: Final = archive.read(members[0]).splitlines()
return tuple(line.removeprefix(b"Tag:").strip().decode("ascii") for line in lines if line.startswith(b"Tag:"))
def _load_native_module(native_path: Path) -> ModuleType | None:
module_spec: Final = importlib.util.spec_from_file_location("litellm.rust_bridge._native", native_path)
if module_spec is None or module_spec.loader is None:
return None
try:
native_module: Final = importlib.util.module_from_spec(module_spec)
module_spec.loader.exec_module(native_module)
except Exception as error: # noqa: BLE001 # native module initialization can raise arbitrary exceptions
sys.stderr.write(f"native module load failed: {error}\n")
return None
return native_module
def main(
argv: Sequence[str] | None = None,
environment: Mapping[str, str] | None = None,
load_native_module: Callable[[Path], ModuleType | None] = _load_native_module,
run_command: CommandRunner = _run_command,
) -> int:
arguments: Final = tuple(sys.argv if argv is None else argv)
resolved_environment: Final = os.environ if environment is None else environment
if len(arguments) != 2:
sys.stderr.write(f"usage: {Path(arguments[0]).name} WHEEL\n")
return 2
wheel: Final = Path(arguments[1])
wheel_tags: Final = wheel.stem.rsplit("-", maxsplit=3)
if len(wheel_tags) != 4:
sys.stderr.write(f"cannot parse wheel tags from {wheel.name}\n")
return 1
wheel_identity: Final = wheel_tags[0].split("-")
if len(wheel_identity) != 2 or wheel_identity[0] != "litellm" or not wheel_identity[1]:
sys.stderr.write(f"unexpected wheel identity: {wheel_tags[0]}\n")
return 1
expected_dist_info_directory: Final = f"{wheel_tags[0]}.dist-info"
expected_dist_info_directories: Final = frozenset((expected_dist_info_directory,))
python_tag: Final = wheel_tags[1]
abi_tag: Final = wheel_tags[2]
platform_tag: Final = wheel_tags[3]
expanded_filename_tags: Final = frozenset(
"-".join(tag) for tag in product(python_tag.split("."), abi_tag.split("."), platform_tag.split("."))
)
with zipfile.ZipFile(wheel) as archive:
wheel_members: Final = archive.infolist()
dist_info_directories: Final = frozenset(
directory for member in wheel_members if (directory := _dist_info_directory(member)) is not None
)
required_dist_info_files: Final = ("METADATA", "RECORD", "WHEEL")
dist_info_file_counts: Final = MappingProxyType(
{
filename: sum(
member.filename == f"{expected_dist_info_directory}/{filename}" for member in wheel_members
)
for filename in required_dist_info_files
}
)
wheel_metadata_members: Final = tuple(
member for member in wheel_members if member.filename == f"{expected_dist_info_directory}/WHEEL"
)
wheel_metadata_tags: Final = _wheel_metadata_tags(archive, wheel_metadata_members)
native_members: Final = tuple(
member
for member in wheel_members
if member.filename.startswith("litellm/rust_bridge/_native.") and member.filename.endswith(".so")
)
if len(native_members) != 1:
sys.stderr.write(f"expected one native extension, found {len(native_members)}\n")
return 1
unexpected_members: Final = tuple(
member.filename
for member in wheel_members
if member.filename.endswith((".pdb", ".dwp", ".rlib", ".rmeta", "Cargo.toml", "Cargo.lock"))
or any(part.endswith(".dSYM") for part in PurePosixPath(member.filename).parts)
)
native_member: Final = native_members[0]
uncompressed_wheel_size: Final = sum(member.file_size for member in wheel_members)
native_path: Final = wheel.parent / "native" / Path(native_member.filename).name
native_path.parent.mkdir(parents=True, exist_ok=True)
native_path.write_bytes(archive.read(native_member))
wheel_metadata_tags_match: Final = (
len(wheel_metadata_tags) == len(expanded_filename_tags)
and frozenset(wheel_metadata_tags) == expanded_filename_tags
)
commit_sha: Final = resolved_environment.get(
"RELEASE_WHEEL_COMMIT_SHA", resolved_environment.get("GITHUB_SHA", "unknown")
)
rustc_version: Final = run_command(
("rustc", "--version"),
check=True,
capture_output=True,
text=True,
).stdout.strip()
pyproject: Final = (Path(__file__).parents[2] / "pyproject.toml").read_text()
maturin_match: Final = re.search(r'"maturin==([^";]+)', pyproject)
if maturin_match is None:
sys.stderr.write("build-system does not pin an exact Maturin version\n")
return 1
maturin_version: Final = maturin_match.group(1)
native_percentage: Final = native_member.file_size / uncompressed_wheel_size * 100
size_report: Final = "\n".join(
(
"## Native wheel build report",
"",
"| Build | Value |",
"| --- | --- |",
f"| Commit | `{commit_sha}` |",
f"| Platform | `{platform_tag}` |",
f"| Python ABI | `{python_tag}-{abi_tag}` |",
f"| Rust compiler | `{rustc_version}` |",
f"| Maturin | `{maturin_version}` |",
"| Cargo profile | `release` |",
"",
"| Artifact | Size |",
"| --- | ---: |",
f"| Compressed wheel | {wheel.stat().st_size / 1_000_000:.2f} MB |",
f"| Uncompressed wheel | {uncompressed_wheel_size / 1_000_000:.2f} MB |",
f"| Native extension | {native_member.file_size / 1_000_000:.2f} MB |",
f"| Native share | {native_percentage:.2f}% |",
"",
)
)
summary_path: Final = resolved_environment.get("GITHUB_STEP_SUMMARY")
if summary_path is None:
sys.stdout.write(size_report)
else:
Path(summary_path).write_text(size_report)
sections: Final = run_command(
("readelf", "--sections", "--wide", str(native_path)),
check=True,
capture_output=True,
text=True,
).stdout
debug_sections: Final = tuple(section for section in (".debug_", ".zdebug_") if section in sections)
debug_sections_absent: Final = not debug_sections
static_symbol_table_absent: Final = ".symtab" not in sections
dynamic_symbols: Final = run_command(
("readelf", "--dyn-syms", "--wide", str(native_path)),
check=True,
capture_output=True,
text=True,
).stdout
extension_entry_point_present: Final = "PyInit__native" in dynamic_symbols
native_module: Final = load_native_module(native_path)
native_module_loads: Final = native_module is not None
panic_test_hook_absent: Final = native_module is not None and not hasattr(native_module, "_panic_for_test")
native_size_limit: Final = 20_000_000
native_size_within_limit: Final = native_member.file_size <= native_size_limit
validations: Final = (
(f"Python tag is {EXPECTED_PYTHON_TAG}", python_tag == EXPECTED_PYTHON_TAG),
(f"ABI tag is {EXPECTED_ABI_TAG}", abi_tag == EXPECTED_ABI_TAG),
(f"Platform tag is {EXPECTED_PLATFORM_TAG}", platform_tag == EXPECTED_PLATFORM_TAG),
("Wheel dist-info directory matches the filename", dist_info_directories == expected_dist_info_directories),
(
"Required dist-info files are present exactly once",
all(count == 1 for count in dist_info_file_counts.values()),
),
("Wheel metadata tags match the filename", wheel_metadata_tags_match),
("Debug sections are absent", debug_sections_absent),
("Static symbol table is absent", static_symbol_table_absent),
("Python extension entry point is present", extension_entry_point_present),
("Native module loads", native_module_loads),
("Production module omits the panic test hook", panic_test_hook_absent),
("Native extension does not exceed 20 MB", native_size_within_limit),
("Wheel contents are valid", not unexpected_members),
)
verified_report: Final = size_report + "\n".join(
("", "| Validation | Expected | Result |", "| --- | --- | :---: |")
+ tuple(f"| {label} | Yes | {'O' if passed else 'X'} |" for label, passed in validations)
+ ("",)
)
if summary_path is not None:
Path(summary_path).write_text(verified_report)
invalid_dist_info_files: Final = any(count != 1 for count in dist_info_file_counts.values())
validation_errors: Final = tuple(
message
for failed, message in (
(bool(debug_sections), f"{native_member.filename} contains debug sections: {', '.join(debug_sections)}"),
(not static_symbol_table_absent, f"{native_member.filename} contains a static symbol table"),
(not extension_entry_point_present, "native extension does not export PyInit__native"),
(
python_tag != EXPECTED_PYTHON_TAG,
f"unexpected Python tag: expected {EXPECTED_PYTHON_TAG}, found {python_tag}",
),
(abi_tag != EXPECTED_ABI_TAG, f"unexpected ABI tag: expected {EXPECTED_ABI_TAG}, found {abi_tag}"),
(
platform_tag != EXPECTED_PLATFORM_TAG,
f"unexpected platform tag: expected {EXPECTED_PLATFORM_TAG}, found {platform_tag}",
),
(
dist_info_directories != expected_dist_info_directories,
f"unexpected dist-info directories: expected {expected_dist_info_directory}, "
f"found {', '.join(sorted(dist_info_directories))}",
),
(invalid_dist_info_files, f"required dist-info file counts are invalid: {dist_info_file_counts}"),
(
not invalid_dist_info_files and not wheel_metadata_tags_match,
f"WHEEL tags do not match filename: expected {', '.join(sorted(expanded_filename_tags))}, "
f"found {', '.join(sorted(wheel_metadata_tags))}",
),
(
native_module is not None and not panic_test_hook_absent,
"production native module exposes _panic_for_test",
),
(
not native_size_within_limit,
f"native extension exceeds 20 MB: {native_member.file_size / 1_000_000:.2f} MB",
),
(bool(unexpected_members), f"wheel contains unexpected build artifacts: {', '.join(unexpected_members)}"),
)
if failed
)
sys.stderr.write("".join(f"{message}\n" for message in validation_errors))
return 0 if all(passed for _, passed in validations) else 1
if __name__ == "__main__":
sys.exit(main())

69
.github/workflows/auto-close-duplicates.yml vendored Normal file
View file

@ -0,0 +1,69 @@
name: Auto-close duplicate issues
on:
schedule:
- cron: "0 9 * * *"
workflow_dispatch:
inputs:
dry_run:
description: Log which issues would close without closing anything
type: boolean
default: true
grace_period_days:
description: Days a duplicate notice must go unanswered before the close
type: number
default: 3
pull_request:
paths:
- .github/workflows/auto-close-duplicates.yml
- scripts/auto-close-duplicates.ts
- scripts/auto-close-duplicates.test.ts
permissions: {}
jobs:
test:
if: github.event_name == 'pull_request'
runs-on: ubuntu-latest
timeout-minutes: 5
permissions:
contents: read
steps:
- name: Checkout repository
uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0
with:
persist-credentials: false
- name: Setup Bun
uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
with:
bun-version: "1.4.0"
- name: Test the sweep
run: bun test scripts/auto-close-duplicates.test.ts
sweep:
if: github.event_name != 'pull_request' && github.repository == 'BerriAI/litellm'
runs-on: ubuntu-latest
timeout-minutes: 10
permissions:
contents: read
issues: write
steps:
- name: Checkout repository
uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0
with:
persist-credentials: false
- name: Setup Bun
uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
with:
# Exact version, never latest: the next step holds an issues: write token
bun-version: "1.4.0"
- name: Close unanswered duplicates, reopen ones the reporter answered
run: bun run scripts/auto-close-duplicates.ts
env:
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
DRY_RUN: ${{ inputs.dry_run == true }}
GRACE_PERIOD_DAYS: ${{ inputs.grace_period_days }}

40
.github/workflows/check_duplicate_issues.yml vendored
View file

@ -1,12 +1,19 @@
name: Check Duplicate Issues
# Flagging only. "Auto-close duplicate issues" closes a flagged issue 3 days later,
# and only when its title is identical to an older open issue and nobody replied.
# The HTML marker below is the handshake between the two, so keep it in the template.
on:
issues:
types: [opened, edited]
permissions: {}
jobs:
check-duplicate:
runs-on: ubuntu-latest
timeout-minutes: 5
permissions:
issues: write
contents: read
@ -19,35 +26,12 @@ jobs:
threshold: 0.6
reaction: eyes
comment: |
**⚠️ Potential duplicate detected**
<!-- litellm:potential-duplicate candidates={{#issues}}{{number}},{{/issues}} -->
**Potential duplicate detected**
This issue appears similar to existing issue(s):
This looks similar to:
{{#issues}}
- [#{{number}}]({{html_url}}) - {{title}} ({{accuracy}}% similar)
- #{{number}} - {{title}}
{{/issues}}
Please review the linked issue(s) to see if they address your concern. If this is not a duplicate, please provide additional context to help us understand the difference.
- name: Checkout close script
if: github.event.action == 'opened'
uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0
with:
sparse-checkout: .github/scripts
persist-credentials: false
- name: Set up Python
if: github.event.action == 'opened'
uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
with:
python-version: "3.12"
- name: Auto-close if high-confidence duplicate
if: github.event.action == 'opened'
env:
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
run: |
python3 .github/scripts/close_duplicate_issues.py \
--issue-number ${{ github.event.issue.number }} \
--repo ${{ github.repository }} \
--threshold 0.85 \
--close
If this is a duplicate, add a thumbs-up reaction to the existing issue and follow along there. When the title is identical to an older open issue, this issue closes automatically in 3 days unless someone responds. If it is not a duplicate, comment here or add a thumbs-down reaction to this comment and it stays open.

6
.github/workflows/image-scan.yml vendored
View file

@ -80,7 +80,7 @@ jobs:
LITELLM_IMAGE: litellm-image-scan:${{ github.sha }}
run: |
python -m pip install "pytest==9.0.3"
python -m pytest tests/proxy_migration_tests/test_offline_image_migration.py -v
python -m pytest tests/proxy_migration_tests/test_offline_image_migration.py tests/proxy_migration_tests/test_image_bedrock_realtime_extra.py -v
# Scans the whole shipped artifact: OS/apk plus every language package
# baked into the image, including ones no lockfile declares (e.g. prisma's
@ -124,7 +124,7 @@ jobs:
LITELLM_IMAGE: litellm-runtime-scan:${{ github.sha }}
run: |
python -m pip install "pytest==9.0.3"
python -m pytest tests/proxy_migration_tests/test_offline_image_migration.py -v
python -m pytest tests/proxy_migration_tests/test_offline_image_migration.py tests/proxy_migration_tests/test_image_bedrock_realtime_extra.py -v
migrations-image:
name: migrations-image
@ -185,7 +185,7 @@ jobs:
LITELLM_COMPONENT_PORT: "4000"
run: |
python -m pip install "pytest==9.0.3"
python -m pytest tests/proxy_migration_tests/test_component_image_serves_offline.py -v
python -m pytest tests/proxy_migration_tests/test_component_image_serves_offline.py tests/proxy_migration_tests/test_image_bedrock_realtime_extra.py -v
ui-image:
name: ui-image

130
.github/workflows/report-rust-release-wheel.yml vendored Normal file
View file

@ -0,0 +1,130 @@
name: Report LiteLLM Rust release wheel
on: # zizmor: ignore[dangerous-triggers] reporter executes no PR code and consumes no PR artifacts or outputs
workflow_run:
workflows:
- LiteLLM Rust
types:
- completed
permissions: {}
concurrency:
group: ${{ github.workflow }}-${{ github.event.workflow_run.pull_requests[0].number || github.event.workflow_run.id }}
cancel-in-progress: false
jobs:
report-release-wheel:
name: report release wheel
if: >-
github.event.workflow_run.event == 'pull_request' &&
github.event.workflow_run.path == '.github/workflows/test-rust.yml' &&
github.event.workflow_run.head_repository.full_name == github.repository &&
github.event.workflow_run.pull_requests[0].number != null
runs-on: ubuntu-latest
timeout-minutes: 5
permissions:
issues: write # PR comments use the issues API
pull-requests: read # Current-head validation rejects stale workflow runs
steps:
- name: Link release wheel report on PR
uses: actions/github-script@f28e40c7f34bde8b3046d885e986cb6290c5673b # v7.1.0
env:
COMMENT_MARKER: "<!-- litellm-release-wheel-size -->"
with:
script: |
const marker = process.env.COMMENT_MARKER;
const workflowRun = context.payload.workflow_run;
const allowedConclusions = new Set([
"action_required",
"cancelled",
"failure",
"neutral",
"skipped",
"stale",
"startup_failure",
"success",
"timed_out",
]);
if (
!allowedConclusions.has(workflowRun.conclusion) ||
workflowRun.event !== "pull_request" ||
workflowRun.path !== ".github/workflows/test-rust.yml" ||
workflowRun.head_repository?.full_name !==
`${context.repo.owner}/${context.repo.repo}` ||
workflowRun.pull_requests?.length !== 1
) {
throw new Error("unexpected source workflow");
}
const pullRequest = workflowRun.pull_requests[0];
const pullRequestNumber = pullRequest.number;
const headSha = workflowRun.head_sha;
const runId = workflowRun.id;
if (
!Number.isSafeInteger(pullRequestNumber) ||
pullRequestNumber <= 0 ||
!Number.isSafeInteger(runId) ||
runId <= 0 ||
!/^[0-9a-f]{40}$/.test(headSha) ||
pullRequest.head?.sha !== headSha
) {
throw new Error("invalid source workflow metadata");
}
const runUrl =
`${context.serverUrl}/${context.repo.owner}/${context.repo.repo}` +
`/actions/runs/${runId}`;
const result =
workflowRun.conclusion === "success"
? "successfully"
: `with \`${workflowRun.conclusion}\``;
const body = [
marker,
"## LiteLLM Rust workflow",
"",
`Workflow completed ${result} for \`${headSha}\``,
"",
`[View workflow run](${runUrl})`,
].join("\n");
const comments = await github.paginate(github.rest.issues.listComments, {
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: pullRequestNumber,
per_page: 100,
});
const existing = comments.find(
(comment) =>
comment.user?.login === "github-actions[bot]" &&
comment.body?.startsWith(marker),
);
const currentPullRequest = (
await github.rest.pulls.get({
owner: context.repo.owner,
repo: context.repo.repo,
pull_number: pullRequestNumber,
})
).data;
if (
currentPullRequest.state !== "open" ||
currentPullRequest.head.repo?.full_name !==
`${context.repo.owner}/${context.repo.repo}` ||
currentPullRequest.head.sha !== headSha
) {
core.info("source workflow no longer matches the current pull request head");
return;
}
if (existing) {
await github.rest.issues.updateComment({
owner: context.repo.owner,
repo: context.repo.repo,
comment_id: existing.id,
body,
});
} else {
await github.rest.issues.createComment({
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: pullRequestNumber,
body,
});
}

77
.github/workflows/test-redis-compat.yml vendored Normal file
View file

@ -0,0 +1,77 @@
name: "Unit Tests: Redis Client Version Compatibility"
on:
pull_request:
branches:
- main
- litellm_internal_staging
- litellm_oss_staging
- "litellm_**"
paths:
- "litellm/_redis.py"
- "litellm/_redis_credential_provider.py"
- "tests/test_litellm/test_redis.py"
- "tests/test_litellm/caching/test_redis_connection_pool.py"
- ".github/workflows/test-redis-compat.yml"
- "pyproject.toml"
- "uv.lock"
permissions:
contents: read
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
cancel-in-progress: true
jobs:
redis-compat:
name: "redis-py ${{ matrix.redis-version }}"
runs-on: ubuntu-latest
timeout-minutes: 15
strategy:
fail-fast: false
matrix:
# 5.3.1 is the version pinned in uv.lock (redisvl caps it below 6); the
# newer legs prove the inspect.signature introspection in litellm/_redis.py
# keeps extracting kwargs on the redis-py releases people actually run now.
# Only the exact release 6.0.0 is skipped: rq (pulled by the proxy extra)
# specifies `redis != 6`, which excludes 6.0.0 alone, so 6.4.0 stands in
# for the 6.x line.
redis-version: ["5.3.1", "6.4.0", "7.4.1", "8.0.1"]
steps:
- uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0
with:
persist-credentials: false
- name: Set up Python
uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
with:
python-version: "3.12"
- name: Set up uv
uses: ./.github/actions/setup-uv-with-retries
with:
version: "0.10.9"
- name: Install dependencies
run: |
.github/scripts/uv_sync_with_retries.sh --frozen --group ci --group proxy-dev --extra google --extra proxy --extra semantic-router
- name: Pin redis-py to the matrix version
env:
REDIS_VERSION: ${{ matrix.redis-version }}
run: |
uv pip install "redis==${REDIS_VERSION:?}"
uv run --no-sync python -c "import redis; assert redis.__version__ == '${REDIS_VERSION:?}', redis.__version__; print('redis-py', redis.__version__)"
- name: Run redis unit tests
run: |
uv run --no-sync pytest \
tests/test_litellm/test_redis.py \
tests/test_litellm/caching/test_redis_connection_pool.py \
--tb=short -vv \
--reruns 2 \
--reruns-delay 1 \
--durations=20

65
.github/workflows/test-rust.yml vendored
View file

@ -4,6 +4,12 @@ on:
push:
paths:
- "litellm-rust/**"
- ".cargo/**"
- "pyproject.toml"
- "rust-toolchain.toml"
- ".github/scripts/smoke_test_native_wheel.py"
- ".github/scripts/verify_linux_native_wheel.py"
- "tests/test_litellm/rust_bridge/native_route_wheel_test.py"
- ".github/workflows/test-rust.yml"
pull_request:
branches:
@ -13,6 +19,12 @@ on:
- "litellm_**"
paths:
- "litellm-rust/**"
- ".cargo/**"
- "pyproject.toml"
- "rust-toolchain.toml"
- ".github/scripts/smoke_test_native_wheel.py"
- ".github/scripts/verify_linux_native_wheel.py"
- "tests/test_litellm/rust_bridge/native_route_wheel_test.py"
- ".github/workflows/test-rust.yml"
permissions:
@ -40,9 +52,7 @@ jobs:
persist-credentials: false
- name: Set up Rust
run: |
rustup toolchain install stable --profile minimal --component clippy,rustfmt
rustup default stable
run: rustup toolchain install
- name: Cache Cargo registry and target
uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
@ -51,7 +61,7 @@ jobs:
~/.cargo/registry
~/.cargo/git
litellm-rust/target
key: ${{ runner.os }}-cargo-${{ hashFiles('litellm-rust/Cargo.lock') }}
key: ${{ runner.os }}-cargo-${{ hashFiles('rust-toolchain.toml', 'litellm-rust/Cargo.lock') }}
restore-keys: |
${{ runner.os }}-cargo-
@ -69,3 +79,50 @@ jobs:
- name: Run core tests with Bedrock auth
run: cargo test -p litellm-core --features bedrock-auth --locked
release-wheel:
name: release wheel
runs-on: ubuntu-latest
timeout-minutes: 20
permissions:
contents: read
env:
CARGO_TERM_COLOR: always
steps:
- name: Checkout repository
uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0
with:
persist-credentials: false
- name: Set up Python
uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
with:
python-version: "3.12"
- name: Set up uv
uses: ./.github/actions/setup-uv-with-retries
with:
version: "0.10.9"
- name: Set up Rust
run: rustup toolchain install
- name: Build release wheel
run: uv build --wheel --out-dir dist
- name: Build panic contract wheel
run: >-
uv build --wheel --out-dir panic-dist
--config-setting "maturin.build-args=--features panic-test,extension-module"
- name: Smoke-test native panic unwinding
run: python .github/scripts/smoke_test_native_wheel.py panic-dist/*.whl
- name: Verify stripped native extension
env:
RELEASE_WHEEL_COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.sha }}
run: python .github/scripts/verify_linux_native_wheel.py dist/*.whl
- name: Test native route wheel
run: python tests/test_litellm/rust_bridge/native_route_wheel_test.py dist/*.whl

2
.github/workflows/test-unit.yml vendored
View file

@ -96,6 +96,7 @@ jobs:
- shard: misc
artifact-name: misc
test-path: >-
tests/sdk_function_trace
tests/test_litellm/batches
tests/test_litellm/secret_managers
tests/test_litellm/a2a_protocol
@ -103,6 +104,7 @@ jobs:
tests/test_litellm/completion_extras
tests/test_litellm/compression
tests/test_litellm/containers
tests/test_litellm/endpoints
tests/test_litellm/experimental_mcp_client
tests/test_litellm/models
tests/test_litellm/repositories

2
CLAUDE.md
View file

@ -23,6 +23,8 @@ When adding new features, add meaningful tests. Don't add tests that don't check
Same thing for bug fixes. The tests should make it so that this specific bug can never happen again without failing tests (i.e., regression)
Never test structure of code only function of it
`tests/test_litellm/` mirrors `litellm/` in a parallel path (see `tests/test_litellm/readme.md`). Name tests `test_<filename>.py`, but always match the existing test file in the directory you touch — many provider dirs use longer descriptive names (e.g. `test_anthropic_chat_transformation.py`) to avoid ambiguity across sibling folders. For bug fixes, extend the existing mapped test file rather than creating a new one. Only create a new test file for a new feature (provider, endpoint, or transformation module) that has no mapped test yet, following that directory's naming convention (or `test_<filename>.py` if you're the first test there). One focused regression test beats many shallow ones
End-to-end tests belong in `tests/e2e/` and must follow the harness conventions documented in that directory's `CLAUDE.md`

23
Dockerfile
View file

@ -1,10 +1,10 @@
# syntax=docker/dockerfile:1.7
# Base image for building
ARG LITELLM_BUILD_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:a31344ab2cb8618db84f535eec56f76f6178b142cb92cb2e48676cc2dcebea72
ARG LITELLM_BUILD_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:e624c5d5e42382ce7165ddafcbbf8e6769a24cbd02ea6114b880b05ae5ba2a8d
# Runtime image
ARG LITELLM_RUNTIME_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:a31344ab2cb8618db84f535eec56f76f6178b142cb92cb2e48676cc2dcebea72
ARG LITELLM_RUNTIME_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:e624c5d5e42382ce7165ddafcbbf8e6769a24cbd02ea6114b880b05ae5ba2a8d
ARG UV_IMAGE=ghcr.io/astral-sh/uv:0.11.7@sha256:240fb85ab0f263ef12f492d8476aa3a2e4e1e333f7d67fbdd923d00a506a516a
# Pinned by digest like the other base images; bump explicitly on Node upgrades.
ARG UI_BUILD_IMAGE=node:24.19-alpine3.24@sha256:d32cdf619f63fe0471182d08996dd516c6275bb5fd31ae06e55a570bd9e1ad43
@ -40,8 +40,8 @@ COPY --from=uvbin /uvx /usr/local/bin/uvx
RUN apk add --no-cache \
bash \
gcc \
python3 \
python3-dev \
python-3.13 \
python-3.13-dev \
rust \
openssl \
openssl-dev \
@ -51,6 +51,7 @@ RUN apk add --no-cache \
ENV UV_PROJECT_ENVIRONMENT=/app/.venv \
UV_LINK_MODE=copy \
UV_PYTHON_DOWNLOADS=0 \
PATH="/app/.venv/bin:${PATH}"
# Copy dependency metadata first for layer caching
@ -65,7 +66,8 @@ RUN uv sync --frozen --no-install-project --no-install-workspace --no-default-gr
--extra extra_proxy \
--extra semantic-router \
--extra saml \
--python python3
--extra bedrock-realtime \
--python python3.13
# Copy full source tree
COPY . .
@ -86,7 +88,8 @@ RUN uv sync --frozen --no-default-groups --no-editable \
--extra extra_proxy \
--extra semantic-router \
--extra saml \
--python python3
--extra bedrock-realtime \
--python python3.13
RUN HOME=/opt/prisma XDG_CACHE_HOME=/opt/prisma/.cache PRISMA_BINARY_CACHE_DIR=/opt/prisma/binaries \
npm_config_cache=/root/.npm \
@ -100,8 +103,14 @@ FROM $LITELLM_RUNTIME_IMAGE AS runtime
USER root
# The base image only configures Chainguard's authenticated apk repo, which
# requires an enterprise subscription. Add the public Wolfi repo so `apk add`
# also works for anyone installing extra packages into a running container.
# https://github.com/BerriAI/litellm/issues/33518
RUN echo "https://packages.wolfi.dev/os" >> /etc/apk/repositories
# node (without npm) is required by the prisma CLI at runtime
RUN apk add --no-cache bash openssl tzdata nodejs python3 libsndfile
RUN apk add --no-cache bash openssl tzdata nodejs python-3.13 libsndfile
WORKDIR /app
ENV PATH="/app/.venv/bin:${PATH}" \

2
README.md
View file

@ -354,6 +354,8 @@ curl -X POST 'http://0.0.0.0:4000/v1/chat/completions' \
| [Petals (`petals`)](https://docs.litellm.ai/docs/providers/petals) | ✅ | ✅ | ✅ | | | | | | | |
| [Pinstripes (`pinstripes`)](https://docs.litellm.ai/docs/providers/pinstripes) | ✅ | ✅ | ✅ | | | | | | | |
| [Predibase (`predibase`)](https://docs.litellm.ai/docs/providers/predibase) | ✅ | ✅ | ✅ | | | | | | | |
| [Qwen AI Platform (`qwen_ai_platform`)](https://docs.litellm.ai/docs/providers/qwencloud) | ✅ | ✅ | ✅ | ✅ | ✅ | | | | | ✅ |
| [QwenCloud (`qwencloud`)](https://docs.litellm.ai/docs/providers/qwencloud) | ✅ | ✅ | ✅ | ✅ | ✅ | | | | | ✅ |
| [Recraft (`recraft`)](https://docs.litellm.ai/docs/providers/recraft) | | | | | ✅ | | | | | |
| [Replicate (`replicate`)](https://docs.litellm.ai/docs/providers/replicate) | ✅ | ✅ | ✅ | | | | | | | |
| [Sagemaker Chat (`sagemaker_chat`)](https://docs.litellm.ai/docs/providers/aws_sagemaker) | ✅ | ✅ | ✅ | | | | | | | |

14
backend/Dockerfile
View file

@ -1,5 +1,5 @@
ARG LITELLM_BUILD_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:a31344ab2cb8618db84f535eec56f76f6178b142cb92cb2e48676cc2dcebea72
ARG LITELLM_RUNTIME_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:a31344ab2cb8618db84f535eec56f76f6178b142cb92cb2e48676cc2dcebea72
ARG LITELLM_BUILD_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:e624c5d5e42382ce7165ddafcbbf8e6769a24cbd02ea6114b880b05ae5ba2a8d
ARG LITELLM_RUNTIME_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:e624c5d5e42382ce7165ddafcbbf8e6769a24cbd02ea6114b880b05ae5ba2a8d
ARG UV_IMAGE=ghcr.io/astral-sh/uv:0.11.7@sha256:240fb85ab0f263ef12f492d8476aa3a2e4e1e333f7d67fbdd923d00a506a516a
FROM $UV_IMAGE AS uvbin
@ -16,7 +16,7 @@ COPY --from=uvbin /uv /uvx /usr/local/bin/
# instead of nodeenv downloading one whose dynamic deps may not be in Wolfi
# (e.g. Node 26.2.0 needs libatomic). Retry for transient apk.cgr.dev flakes.
RUN for i in 1 2 3; do \
apk add --no-cache bash gcc python3 python3-dev openssl openssl-dev libsndfile nodejs npm && break; \
apk add --no-cache bash gcc python-3.13 python-3.13-dev openssl openssl-dev libsndfile nodejs npm && break; \
[ $i = 3 ] && { echo "apk add failed after 3 retries" >&2; exit 1; }; \
sleep 5; \
done
@ -46,7 +46,8 @@ RUN --mount=type=cache,target=/root/.cache/uv \
--extra proxy-runtime \
--extra extra_proxy \
--extra semantic-router \
--python python3
--extra saml \
--python python3.13
# Stage 2 — copy source and install the project + workspace members.
COPY . .
@ -57,7 +58,8 @@ RUN --mount=type=cache,target=/root/.cache/uv \
--extra proxy-runtime \
--extra extra_proxy \
--extra semantic-router \
--python python3
--extra saml \
--python python3.13
RUN HOME=/opt/prisma XDG_CACHE_HOME=/opt/prisma/.cache PRISMA_BINARY_CACHE_DIR=/opt/prisma/binaries \
npm_config_cache=/root/.npm \
@ -71,7 +73,7 @@ FROM $LITELLM_RUNTIME_IMAGE AS runtime
USER root
RUN for i in 1 2 3; do \
apk add --no-cache bash openssl tzdata python3 libsndfile libatomic && break; \
apk add --no-cache bash openssl tzdata python-3.13 libsndfile libatomic && break; \
[ $i = 3 ] && { echo "apk add failed after 3 retries" >&2; exit 1; }; \
sleep 5; \
done

38
basedpyright-code-budget.json
View file

@ -1,9 +1,9 @@
{
"reportAny": {
"limit": 17270
"limit": 14074
},
"reportArgumentType": {
"limit": 2538
"limit": 2215
},
"reportAssignmentType": {
"limit": 319
@ -18,13 +18,13 @@
"limit": 40
},
"reportDeprecated": {
"limit": 212
"limit": 211
},
"reportDuplicateImport": {
"limit": 19
},
"reportExplicitAny": {
"limit": 5485
"limit": 4125
},
"reportFunctionMemberAccess": {
"limit": 7
@ -42,7 +42,7 @@
"limit": 12
},
"reportIndexIssue": {
"limit": 35
"limit": 25
},
"reportInvalidTypeForm": {
"limit": 34
@ -54,10 +54,10 @@
"limit": 0
},
"reportMissingParameterType": {
"limit": 5658
"limit": 5601
},
"reportMissingTypeArgument": {
"limit": 15425
"limit": 15290
},
"reportMissingTypeStubs": {
"limit": 40
@ -72,7 +72,7 @@
"limit": 0
},
"reportOptionalMemberAccess": {
"limit": 1055
"limit": 0
},
"reportOptionalOperand": {
"limit": 0
@ -90,40 +90,40 @@
"limit": 8
},
"reportReturnType": {
"limit": 213
"limit": 181
},
"reportTypedDictNotRequiredAccess": {
"limit": 25
"limit": 24
},
"reportUndefinedVariable": {
"limit": 0
},
"reportUnknownArgumentType": {
"limit": 44526
"limit": 44364
},
"reportUnknownLambdaType": {
"limit": 109
},
"reportUnknownMemberType": {
"limit": 38721
"limit": 38332
},
"reportUnknownParameterType": {
"limit": 19778
"limit": 19625
},
"reportUnknownVariableType": {
"limit": 30290
"limit": 29861
},
"reportUnnecessaryCast": {
"limit": 117
"limit": 111
},
"reportUnnecessaryComparison": {
"limit": 697
"limit": 692
},
"reportUnnecessaryContains": {
"limit": 5
},
"reportUnnecessaryIsInstance": {
"limit": 829
"limit": 826
},
"reportUntypedBaseClass": {
"limit": 0
@ -138,9 +138,9 @@
"limit": 138
},
"reportUnusedImport": {
"limit": 543
"limit": 542
},
"reportUnusedVariable": {
"limit": 139
"limit": 137
}
}

17
docker/Dockerfile.database
View file

@ -1,10 +1,10 @@
# syntax=docker/dockerfile:1.7
# Base image for building
ARG LITELLM_BUILD_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:a31344ab2cb8618db84f535eec56f76f6178b142cb92cb2e48676cc2dcebea72
ARG LITELLM_BUILD_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:e624c5d5e42382ce7165ddafcbbf8e6769a24cbd02ea6114b880b05ae5ba2a8d
# Runtime image
ARG LITELLM_RUNTIME_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:a31344ab2cb8618db84f535eec56f76f6178b142cb92cb2e48676cc2dcebea72
ARG LITELLM_RUNTIME_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:e624c5d5e42382ce7165ddafcbbf8e6769a24cbd02ea6114b880b05ae5ba2a8d
ARG UV_IMAGE=ghcr.io/astral-sh/uv:0.11.7@sha256:240fb85ab0f263ef12f492d8476aa3a2e4e1e333f7d67fbdd923d00a506a516a
# Pinned by digest like the other base images; bump explicitly on Node upgrades.
ARG UI_BUILD_IMAGE=node:24.19-alpine3.24@sha256:d32cdf619f63fe0471182d08996dd516c6275bb5fd31ae06e55a570bd9e1ad43
@ -39,8 +39,8 @@ COPY --from=uvbin /uvx /usr/local/bin/uvx
RUN apk add --no-cache \
bash \
gcc \
python3 \
python3-dev \
python-3.13 \
python-3.13-dev \
openssl \
openssl-dev \
nodejs \
@ -49,6 +49,7 @@ RUN apk add --no-cache \
ENV UV_PROJECT_ENVIRONMENT=/app/.venv \
UV_LINK_MODE=copy \
UV_PYTHON_DOWNLOADS=0 \
PATH="/app/.venv/bin:${PATH}"
# Copy dependency metadata first for layer caching
@ -63,7 +64,8 @@ RUN uv sync --frozen --no-install-project --no-install-workspace --no-default-gr
--extra extra_proxy \
--extra semantic-router \
--extra saml \
--python python3
--extra bedrock-realtime \
--python python3.13
# Copy full source tree
COPY . .
@ -84,7 +86,8 @@ RUN uv sync --frozen --no-default-groups --no-editable \
--extra extra_proxy \
--extra semantic-router \
--extra saml \
--python python3
--extra bedrock-realtime \
--python python3.13
RUN HOME=/opt/prisma XDG_CACHE_HOME=/opt/prisma/.cache PRISMA_BINARY_CACHE_DIR=/opt/prisma/binaries \
npm_config_cache=/root/.npm \
@ -98,7 +101,7 @@ FROM $LITELLM_RUNTIME_IMAGE AS runtime
USER root
# node (without npm) is required by the prisma CLI at runtime
RUN apk add --no-cache bash openssl tzdata nodejs python3 libsndfile
RUN apk add --no-cache bash openssl tzdata nodejs python-3.13 libsndfile
WORKDIR /app
ENV PATH="/app/.venv/bin:${PATH}" \

20
docker/Dockerfile.non_root
View file

@ -1,8 +1,8 @@
# syntax=docker/dockerfile:1.7
# Base images
ARG LITELLM_BUILD_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:a31344ab2cb8618db84f535eec56f76f6178b142cb92cb2e48676cc2dcebea72
ARG LITELLM_RUNTIME_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:a31344ab2cb8618db84f535eec56f76f6178b142cb92cb2e48676cc2dcebea72
ARG LITELLM_BUILD_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:e624c5d5e42382ce7165ddafcbbf8e6769a24cbd02ea6114b880b05ae5ba2a8d
ARG LITELLM_RUNTIME_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:e624c5d5e42382ce7165ddafcbbf8e6769a24cbd02ea6114b880b05ae5ba2a8d
ARG PROXY_EXTRAS_SOURCE=published
ARG UV_IMAGE=ghcr.io/astral-sh/uv:0.11.7@sha256:240fb85ab0f263ef12f492d8476aa3a2e4e1e333f7d67fbdd923d00a506a516a
# Pinned by digest like the other base images; bump explicitly on Node upgrades.
@ -37,8 +37,8 @@ COPY --from=uvbin /uvx /usr/local/bin/uvx
RUN for i in 1 2 3; do \
apk add --no-cache \
python3 \
python3-dev \
python-3.13 \
python-3.13-dev \
gcc \
rust \
bash \
@ -52,6 +52,7 @@ RUN for i in 1 2 3; do \
ENV UV_PROJECT_ENVIRONMENT=/app/.venv \
UV_LINK_MODE=copy \
UV_PYTHON_DOWNLOADS=0 \
PATH="/app/.venv/bin:${PATH}" \
LITELLM_NON_ROOT=true \
XDG_CACHE_HOME=/app/.cache
@ -69,7 +70,8 @@ RUN --mount=type=cache,target=/app/.cache/uv,id=litellm-uv-cache \
--extra extra_proxy \
--extra semantic-router \
--extra saml \
--python python3
--extra bedrock-realtime \
--python python3.13
# Copy full source tree
COPY . .
@ -96,7 +98,8 @@ RUN --mount=type=cache,target=/app/.cache/uv,id=litellm-uv-cache \
--extra extra_proxy \
--extra semantic-router \
--extra saml \
--python python3 \
--extra bedrock-realtime \
--python python3.13 \
--no-sources-package litellm-proxy-extras; \
else \
uv sync --frozen --no-default-groups --no-editable \
@ -105,7 +108,8 @@ RUN --mount=type=cache,target=/app/.cache/uv,id=litellm-uv-cache \
--extra extra_proxy \
--extra semantic-router \
--extra saml \
--python python3; \
--extra bedrock-realtime \
--python python3.13; \
fi
RUN HOME=/opt/prisma XDG_CACHE_HOME=/opt/prisma/.cache PRISMA_BINARY_CACHE_DIR=/opt/prisma/binaries \
@ -124,7 +128,7 @@ RUN for i in 1 2 3; do \
apk upgrade --no-cache && break || sleep 5; \
done && \
for i in 1 2 3; do \
apk add --no-cache python3 bash openssl tzdata libsndfile nodejs && break || sleep 5; \
apk add --no-cache python-3.13 bash openssl tzdata libsndfile nodejs && break || sleep 5; \
done
# Copy only what runtime needs. The application is installed inside the venv;

22
enterprise/litellm_enterprise/proxy/audit_logging_endpoints.py
View file

@ -7,7 +7,7 @@ GET - /audit/{id} - Get audit log by id
GET - /audit - Get all audit logs
"""
from typing import TYPE_CHECKING, Final, Optional
from typing import TYPE_CHECKING, Final
#### AUDIT LOGGING ####
from fastapi import APIRouter, Depends, HTTPException, Query
@ -58,33 +58,33 @@ async def get_audit_logs(
page: int = Query(1, ge=1),
page_size: int = Query(10, ge=1, le=100),
# Filter parameters
changed_by: Optional[str] = Query(
changed_by: str | None = Query(
None, description="Filter by user or system that performed the action"
),
changed_by_api_key: Optional[str] = Query(
changed_by_api_key: str | None = Query(
None, description="Filter by API key hash that performed the action"
),
action: Optional[str] = Query(
action: str | None = Query(
None, description="Filter by action type (create, update, delete)"
),
table_name: Optional[str] = Query(
table_name: str | None = Query(
None, description="Filter by table name that was modified"
),
object_id: Optional[str] = Query(
object_id: str | None = Query(
None, description="Filter by ID of the object that was modified"
),
start_date: Optional[str] = Query(None, description="Filter logs after this date"),
end_date: Optional[str] = Query(None, description="Filter logs before this date"),
object_team_id: Optional[str] = Query(
start_date: str | None = Query(None, description="Filter logs after this date"),
end_date: str | None = Query(None, description="Filter logs before this date"),
object_team_id: str | None = Query(
None,
description="Filter by team_id present in before_value or updated_values JSON (PostgreSQL only)",
),
object_key_hash: Optional[str] = Query(
object_key_hash: str | None = Query(
None,
description="Filter by token (key hash) present in before_value or updated_values JSON (PostgreSQL only)",
),
# Sorting parameters
sort_by: Optional[str] = Query(
sort_by: str | None = Query(
None,
description="Column to sort by (e.g. 'updated_at', 'action', 'table_name')",
),

26
enterprise/litellm_enterprise/proxy/common_utils/check_batch_cost.py
View file

@ -5,7 +5,7 @@ Polls LiteLLM_ManagedObjectTable to check if the batch job is complete, and if t
from dataclasses import replace as dataclasses_replace
from datetime import datetime, timedelta, timezone
from types import MappingProxyType
from typing import TYPE_CHECKING, Any, Dict, Final, List, Literal, Optional, Tuple, cast
from typing import TYPE_CHECKING, Final, List, Literal, Optional, Tuple, cast
from litellm._logging import verbose_proxy_logger
from litellm._uuid import uuid
@ -87,7 +87,7 @@ class CheckBatchCost:
return
self.batch_processed_support_confirmed = True
async def _get_user_info(self, batch_id: str, user_id: Optional[str]) -> Dict[str, Any]:
async def _get_user_info(self, batch_id: str, user_id: Optional[str]) -> dict[str, str | None]:
"""
Look up user email and key alias by user_id for enriching the S3 callback metadata.
Returns a dict with user_api_key_user_email and user_api_key_alias (both may be None).
@ -97,8 +97,10 @@ class CheckBatchCost:
if not user_id:
return {}
try:
user_row = await self.prisma_client.db.litellm_usertable.find_unique(
where={"user_id": user_id}
user_row: prisma_models.LiteLLM_UserTable | None = (
await self.prisma_client.db.litellm_usertable.find_unique(
where={"user_id": user_id}
)
)
if user_row is None:
return {}
@ -115,8 +117,10 @@ class CheckBatchCost:
if not api_key:
return None
try:
key_row = await self.prisma_client.db.litellm_verificationtoken.find_unique(
where={"token": api_key}
key_row: prisma_models.LiteLLM_VerificationToken | None = (
await self.prisma_client.db.litellm_verificationtoken.find_unique(
where={"token": api_key}
)
)
return getattr(key_row, "key_alias", None) if key_row is not None else None
except Exception as e:
@ -128,8 +132,10 @@ class CheckBatchCost:
if not team_id:
return None
try:
team_row = await self.prisma_client.db.litellm_teamtable.find_unique(
where={"team_id": team_id}
team_row: prisma_models.LiteLLM_TeamTable | None = (
await self.prisma_client.db.litellm_teamtable.find_unique(
where={"team_id": team_id}
)
)
return getattr(team_row, "team_alias", None) if team_row is not None else None
except Exception as e:
@ -138,7 +144,7 @@ class CheckBatchCost:
async def _build_creator_attribution_metadata(
self, job: "LiteLLM_ManagedObjectTable", batch_id: str
) -> Dict[str, Any]:
) -> dict[str, object]:
"""
Rebuild the spend-tracking metadata for the key, team, and tags that created the
batch so the batch-cost spend log is attributed the same way a non-batch request
@ -152,7 +158,7 @@ class CheckBatchCost:
team_id = getattr(job, "team_id", None)
request_tags = getattr(job, "request_tags", None)
metadata: Dict[str, Any] = {
metadata: dict[str, object] = {
"user_api_key_user_id": job.created_by,
"user_api_key": api_key,
"user_api_key_team_id": team_id,

23
enterprise/litellm_enterprise/proxy/hooks/managed_files.py
View file

@ -182,6 +182,10 @@ class _ManagedObjectTableActions(Protocol):
async def update_many(self, where: Mapping[str, object], data: Mapping[str, object]) -> int: ...
class _SchedulerWithJobLookup(Protocol):
def get_job(self, job_id: str) -> object: ...
class _CursorPageArgs(TypedDict, total=False):
cursor: Mapping[str, str]
skip: int
@ -853,7 +857,7 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints):
file_ids.append(file_id)
return file_ids
def get_file_ids_from_responses_input(self, input: Union[str, List[Dict[str, Any]]]) -> List[str]:
def get_file_ids_from_responses_input(self, input: Union[str, List[Dict[str, object]]]) -> List[str]:
"""
Gets file ids from responses API input.
@ -878,7 +882,7 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints):
# Check for direct input_file type
if item.get("type") == "input_file":
file_id = item.get("file_id")
if file_id:
if isinstance(file_id, str) and file_id:
file_ids.append(file_id)
# Check for input_file in content array
@ -887,7 +891,7 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints):
for content_item in content:
if isinstance(content_item, dict) and content_item.get("type") == "input_file":
file_id = content_item.get("file_id")
if file_id:
if isinstance(file_id, str) and file_id:
file_ids.append(file_id)
return file_ids
@ -1227,7 +1231,7 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints):
# Handle both output_file_id and error_file_id
for file_attr in ["output_file_id", "error_file_id"]:
file_id_value = getattr(response, file_attr, None)
file_id_value: str | None = getattr(response, file_attr, None)
if file_id_value and model_id:
decoded_output_file_id = _is_base64_encoded_unified_file_id(file_id_value)
if decoded_output_file_id and "llm_output_file_id," in decoded_output_file_id:
@ -1496,7 +1500,7 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints):
import litellm.proxy.proxy_server as proxy_server_module
# Check if the scheduler has the batch cost checking job registered
scheduler = getattr(proxy_server_module, "scheduler", None)
scheduler: Final[_SchedulerWithJobLookup | None] = getattr(proxy_server_module, "scheduler", None)
if scheduler is None:
return False
@ -1542,7 +1546,7 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints):
)
MAX_MATCHES_TO_RETURN = 10
batches = await self.prisma_client.db.litellm_managedobjecttable.find_many(
batches = await _managed_object_table(self.prisma_client).find_many(
where={
"file_purpose": "batch",
"batch_processed": False,
@ -1552,11 +1556,14 @@ class _PROXY_LiteLLMManagedFiles(CustomLogger, BaseFileEndpoints):
order={"created_at": "desc"},
)
referencing_batches = []
referencing_batches: Final[list[dict[str, object]]] = []
for batch in batches:
try:
# Parse the batch file_object to check for file references
batch_data = json.loads(batch.file_object) if isinstance(batch.file_object, str) else batch.file_object
decoded_file_object = _decode_json_blob(batch.file_object)
batch_data: Mapping[str, object] = (
decoded_file_object if isinstance(decoded_file_object, Mapping) else {}
)
# Extract file IDs from batch
# Batches typically reference the unified file ID in input_file_id

4
enterprise/litellm_enterprise/proxy/management_endpoints/internal_user_endpoints.py
View file

@ -39,9 +39,9 @@ async def available_enterprise_users(
if not premium_user:
# check if SSO is enabled - show 5 user limit
from litellm.proxy.auth.auth_utils import _has_user_setup_sso
from litellm.proxy.auth.auth_utils import has_user_setup_sso
if _has_user_setup_sso():
if has_user_setup_sso():
premium_user_data = EnterpriseLicenseData(
max_users=5,
)

4
enterprise/pyproject.toml
View file

@ -1,6 +1,6 @@
[project]
name = "litellm-enterprise"
version = "0.1.62"
version = "0.1.63"
description = "Package for LiteLLM Enterprise features"
readme = "README.md"
requires-python = ">=3.9"
@ -26,7 +26,7 @@ required-version = ">=0.10.9"
module-root = ""
[tool.commitizen]
version = "0.1.62"
version = "0.1.63"
version_files = [
"pyproject.toml:^version",
"../pyproject.toml:litellm-enterprise==",

12
gateway/Dockerfile
View file

@ -1,5 +1,5 @@
ARG LITELLM_BUILD_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:a31344ab2cb8618db84f535eec56f76f6178b142cb92cb2e48676cc2dcebea72
ARG LITELLM_RUNTIME_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:a31344ab2cb8618db84f535eec56f76f6178b142cb92cb2e48676cc2dcebea72
ARG LITELLM_BUILD_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:e624c5d5e42382ce7165ddafcbbf8e6769a24cbd02ea6114b880b05ae5ba2a8d
ARG LITELLM_RUNTIME_IMAGE=cgr.dev/chainguard/wolfi-base@sha256:e624c5d5e42382ce7165ddafcbbf8e6769a24cbd02ea6114b880b05ae5ba2a8d
ARG UV_IMAGE=ghcr.io/astral-sh/uv:0.11.7@sha256:240fb85ab0f263ef12f492d8476aa3a2e4e1e333f7d67fbdd923d00a506a516a
FROM $UV_IMAGE AS uvbin
@ -16,7 +16,7 @@ COPY --from=uvbin /uv /uvx /usr/local/bin/
# instead of nodeenv downloading one whose dynamic deps may not be in Wolfi
# (e.g. Node 26.2.0 needs libatomic). Retry for transient apk.cgr.dev flakes.
RUN for i in 1 2 3; do \
apk add --no-cache bash gcc python3 python3-dev openssl openssl-dev libsndfile nodejs npm && break; \
apk add --no-cache bash gcc python-3.13 python-3.13-dev openssl openssl-dev libsndfile nodejs npm && break; \
[ $i = 3 ] && { echo "apk add failed after 3 retries" >&2; exit 1; }; \
sleep 5; \
done
@ -47,7 +47,7 @@ RUN --mount=type=cache,target=/root/.cache/uv \
--extra extra_proxy \
--extra semantic-router \
--extra bedrock-realtime \
--python python3
--python python3.13
# Stage 2 — copy source and install the project + workspace members.
COPY . .
@ -59,7 +59,7 @@ RUN --mount=type=cache,target=/root/.cache/uv \
--extra extra_proxy \
--extra semantic-router \
--extra bedrock-realtime \
--python python3
--python python3.13
RUN HOME=/opt/prisma XDG_CACHE_HOME=/opt/prisma/.cache PRISMA_BINARY_CACHE_DIR=/opt/prisma/binaries \
npm_config_cache=/root/.npm \
@ -73,7 +73,7 @@ FROM $LITELLM_RUNTIME_IMAGE AS runtime
USER root
RUN for i in 1 2 3; do \
apk add --no-cache bash openssl tzdata python3 libsndfile libatomic && break; \
apk add --no-cache bash openssl tzdata python-3.13 libsndfile libatomic && break; \
[ $i = 3 ] && { echo "apk add failed after 3 retries" >&2; exit 1; }; \
sleep 5; \
done

1
gateway/routes/allowlist.py
View file

@ -86,6 +86,7 @@ GATEWAY_PATH_PREFIXES: tuple[str, ...] = (
"/comprehendmedical",
"/cohere/",
"/gemini/",
"/gigachat/",
"/google/",
"/vertex_ai/",
"/vertex-ai/",

2
helm/litellm-helm/Chart.yaml
View file

@ -18,7 +18,7 @@ type: application
# This is the chart version. This version number should be incremented each time you make changes
# to the chart and its templates, including the app version.
# Versions are expected to follow Semantic Versioning (https://semver.org/)
version: 1.1.2
version: 1.1.3
# This is the version number of the application being deployed. This version number should be
# incremented each time you make changes to the application. Versions are not expected to

4
helm/litellm-helm/README.md
View file

@ -26,7 +26,7 @@ If `db.useStackgresOperator` is used (not yet implemented):
| `replicaCount` | The number of LiteLLM Proxy pods to be deployed | `1` |
| `masterkeySecretName` | The name of the Kubernetes Secret that contains the Master API Key for LiteLLM. If not specified, use the generated secret name. | N/A |
| `masterkeySecretKey` | The key within the Kubernetes Secret that contains the Master API Key for LiteLLM. If not specified, use `masterkey` as the key. | N/A |
| `masterkey` | The Master API Key for LiteLLM. If not specified, a random key in the `sk-...` format is generated. | N/A |
| `masterkey` | The Master API Key for LiteLLM. If not specified, a random key in the `sk-...` format is generated on first install and reused on upgrades. | N/A |
| `environmentSecrets` | An optional array of Secret object names. The keys and values in these secrets will be presented to the LiteLLM proxy pod as environment variables. See below for an example Secret object. | `[]` |
| `environmentConfigMaps` | An optional array of ConfigMap object names. The keys and values in these configmaps will be presented to the LiteLLM proxy pod as environment variables. See below for an example Secret object. | `[]` |
| `image.repository` | LiteLLM Proxy image repository | `ghcr.io/berriai/litellm` |
@ -212,6 +212,8 @@ service, the **Proxy Endpoint** should be set to `http://<RELEASE>-litellm:4000`
The **Proxy Key** is the value specified for `masterkey` or, if a `masterkey`
was not provided to the helm command line, the `masterkey` is a randomly
generated string in the `sk-...` format stored in the `<RELEASE>-litellm-masterkey` Kubernetes Secret.
The key is generated once on the first install; later `helm upgrade` runs reuse the
value already in that Secret, so upgrading never rotates the master key.
```bash
kubectl -n litellm get secret <RELEASE>-litellm-masterkey -o jsonpath="{.data.masterkey}"

6
helm/litellm-helm/templates/secret-masterkey.yaml
View file

@ -1,9 +1,11 @@
{{- if not .Values.masterkeySecretName }}
{{ $masterkey := (.Values.masterkey | default (printf "sk-%s" (randAlphaNum 18))) }}
{{- $secretName := printf "%s-masterkey" (include "litellm.fullname" .) }}
{{- $existing := lookup "v1" "Secret" .Release.Namespace $secretName }}
{{- $masterkey := .Values.masterkey | default (dig "data" "masterkey" "" $existing | b64dec) | default (printf "sk-%s" (randAlphaNum 18)) }}
apiVersion: v1
kind: Secret
metadata:
name: {{ include "litellm.fullname" . }}-masterkey
name: {{ $secretName }}
data:
masterkey: {{ $masterkey | b64enc }}
type: Opaque

42
helm/litellm-helm/tests/hpa_tests.yaml
View file

@ -1,4 +1,4 @@
suite: "hpa with behavior"
suite: "hpa"
templates:
- hpa.yaml
tests:
@ -23,14 +23,44 @@ tests:
- equal: { path: spec.behavior.scaleUp.stabilizationWindowSeconds, value: 60 }
- equal: { path: spec.behavior.scaleDown.stabilizationWindowSeconds, value: 90 }
---
suite: "hpa without behavior"
templates:
- hpa.yaml
tests:
- it: "does not render behavior when not set"
set:
autoscaling.enabled: true
asserts:
- isKind: { of: HorizontalPodAutoscaler }
- isNull: { path: spec.behavior }
- it: "scales on cpu at the documented 60 percent by default"
set:
autoscaling.enabled: true
asserts:
- isKind: { of: HorizontalPodAutoscaler }
- equal: { path: "spec.metrics[0].resource.name", value: cpu }
- equal: { path: "spec.metrics[0].resource.target.type", value: Utilization }
- equal: { path: "spec.metrics[0].resource.target.averageUtilization", value: 60 }
- it: "does not scale on memory by default"
set:
autoscaling.enabled: true
asserts:
- lengthEqual: { path: spec.metrics, count: 1 }
- it: "honours an explicit cpu target override"
set:
autoscaling.enabled: true
autoscaling.targetCPUUtilizationPercentage: 75
asserts:
- equal: { path: "spec.metrics[0].resource.target.averageUtilization", value: 75 }
- it: "renders a memory metric only when a memory target is set"
set:
autoscaling.enabled: true
autoscaling.targetMemoryUtilizationPercentage: 80
asserts:
- lengthEqual: { path: spec.metrics, count: 2 }
- equal: { path: "spec.metrics[1].resource.name", value: memory }
- equal: { path: "spec.metrics[1].resource.target.averageUtilization", value: 80 }
- it: "renders no hpa when autoscaling is disabled"
asserts:
- hasDocuments: { count: 0 }

47
helm/litellm-helm/tests/masterkey-secret_tests.yaml
View file

@ -15,6 +15,53 @@ tests:
# Note: The masterkey is generated as "sk-<18-random-chars>" in plain text,
# but stored as base64 encoded in Kubernetes secret (requirement).
# "sk-" base64 encodes to "c2st", so we check for "^c2st" pattern.
- it: should reuse the master key already stored in the cluster instead of generating a new one on upgrade
template: secret-masterkey.yaml
set:
masterkeySecretName: ""
kubernetesProvider:
scheme:
"v1/Secret":
gvr:
version: "v1"
resource: "secrets"
namespaced: true
objects:
- kind: Secret
apiVersion: v1
metadata:
name: RELEASE-NAME-litellm-masterkey
namespace: NAMESPACE
data:
masterkey: c2stZXhpc3Rpbmcta2V5
asserts:
- equal:
path: data.masterkey
value: c2stZXhpc3Rpbmcta2V5
- it: should let an explicit masterkey value override the one already stored in the cluster
template: secret-masterkey.yaml
set:
masterkeySecretName: ""
masterkey: sk-explicit
kubernetesProvider:
scheme:
"v1/Secret":
gvr:
version: "v1"
resource: "secrets"
namespaced: true
objects:
- kind: Secret
apiVersion: v1
metadata:
name: RELEASE-NAME-litellm-masterkey
namespace: NAMESPACE
data:
masterkey: c2stZXhpc3Rpbmcta2V5
asserts:
- equal:
path: data.masterkey
value: c2stZXhwbGljaXQ=
- it: should not create a secret if masterkeySecretName is set
template: secret-masterkey.yaml
set:

11
helm/litellm-helm/values.yaml
View file

@ -200,7 +200,16 @@ autoscaling:
enabled: false
minReplicas: 1
maxReplicas: 100
targetCPUUtilizationPercentage: 80
# 60 is the documented recommendation. See "Recommended Machine Specifications"
# in https://docs.litellm.ai/docs/proxy/prod. A new replica clears the startupProbe
# above only after up to failureThreshold x periodSeconds = 300 seconds, so a target
# high enough to trip near saturation adds capacity minutes after it was needed.
targetCPUUtilizationPercentage: 60
# Deliberately left unset rather than given a value. The prisma query engine's
# resident memory is a high-water mark that ratchets to the pod's worst-ever write
# and is never returned, so a memory target reads the largest write a pod ever did
# rather than what it is doing now, and replicas ratchet up without scaling back in.
# Memory is a floor to provision under 'resources', not a signal to scale on.
# targetMemoryUtilizationPercentage: 80
# behavior: {}

4
helm/litellm/templates/backend/deployment.yaml
View file

@ -7,6 +7,10 @@ metadata:
{{- include "litellm.commonLabels" . | nindent 4 }}
app.kubernetes.io/component: backend
spec:
{{- with .Values.backend.strategy }}
strategy:
{{- toYaml . | nindent 4 }}
{{- end }}
selector:
matchLabels:
{{- include "litellm.backend.selectorLabels" . | nindent 6 }}

4
helm/litellm/templates/gateway/deployment.yaml
View file

@ -7,6 +7,10 @@ metadata:
{{- include "litellm.commonLabels" . | nindent 4 }}
app.kubernetes.io/component: gateway
spec:
{{- with .Values.gateway.strategy }}
strategy:
{{- toYaml . | nindent 4 }}
{{- end }}
selector:
matchLabels:
{{- include "litellm.gateway.selectorLabels" . | nindent 6 }}

12
helm/litellm/templates/migrations-job.yaml
View file

@ -7,6 +7,8 @@
#
# Running this pre-upgrade closes the window where new application pods would
# otherwise serve traffic against the previous release's unmigrated schema.
# Argo CD users can swap the Helm hook for a PreSync hook through
# `migrationJob.hooks`, which re-runs the Job on every sync.
apiVersion: batch/v1
kind: Job
metadata:
@ -14,10 +16,18 @@ metadata:
labels:
{{- include "litellm.commonLabels" . | nindent 4 }}
app.kubernetes.io/component: migrations
{{- if or .Values.migrationJob.hooks.helm.enabled .Values.migrationJob.hooks.argocd.enabled }}
annotations:
{{- if .Values.migrationJob.hooks.helm.enabled }}
helm.sh/hook: pre-install,pre-upgrade
helm.sh/hook-delete-policy: before-hook-creation
helm.sh/hook-weight: "0"
helm.sh/hook-weight: {{ .Values.migrationJob.hooks.helm.weight | default "0" | quote }}
{{- end }}
{{- if .Values.migrationJob.hooks.argocd.enabled }}
argocd.argoproj.io/hook: PreSync
argocd.argoproj.io/hook-delete-policy: BeforeHookCreation
{{- end }}
{{- end }}
spec:
backoffLimit: {{ .Values.migrationJob.backoffLimit }}
ttlSecondsAfterFinished: {{ .Values.migrationJob.ttlSecondsAfterFinished }}

4
helm/litellm/templates/ui/deployment.yaml
View file

@ -7,6 +7,10 @@ metadata:
{{- include "litellm.commonLabels" . | nindent 4 }}
app.kubernetes.io/component: ui
spec:
{{- with .Values.ui.strategy }}
strategy:
{{- toYaml . | nindent 4 }}
{{- end }}
selector:
matchLabels:
{{- include "litellm.ui.selectorLabels" . | nindent 6 }}

63
helm/litellm/tests/migration_job_hooks_tests.yaml Normal file
View file

@ -0,0 +1,63 @@
suite: test migrations Job hook annotations
templates:
- migrations-job.yaml
values:
- ./values/required.yaml
tests:
- it: runs as a Helm pre-install / pre-upgrade hook by default
asserts:
- equal:
path: metadata.annotations["helm.sh/hook"]
value: pre-install,pre-upgrade
- equal:
path: metadata.annotations["helm.sh/hook-delete-policy"]
value: before-hook-creation
- equal:
path: metadata.annotations["helm.sh/hook-weight"]
value: "0"
- notExists:
path: metadata.annotations["argocd.argoproj.io/hook"]
- it: adds the Argo CD PreSync hook when asked
set:
migrationJob.hooks.argocd.enabled: true
asserts:
- equal:
path: metadata.annotations["argocd.argoproj.io/hook"]
value: PreSync
- equal:
path: metadata.annotations["argocd.argoproj.io/hook-delete-policy"]
value: BeforeHookCreation
- it: drops the Helm hook so Argo CD owns the Job
set:
migrationJob.hooks.argocd.enabled: true
migrationJob.hooks.helm.enabled: false
asserts:
- equal:
path: metadata.annotations["argocd.argoproj.io/hook"]
value: PreSync
- notExists:
path: metadata.annotations["helm.sh/hook"]
- notExists:
path: metadata.annotations["helm.sh/hook-delete-policy"]
- notExists:
path: metadata.annotations["helm.sh/hook-weight"]
- it: renders an ordinary Job when both hooks are disabled
set:
migrationJob.hooks.helm.enabled: false
asserts:
- notExists:
path: metadata.annotations
- equal:
path: kind
value: Job
- it: honours a custom Helm hook weight
set:
migrationJob.hooks.helm.weight: "-5"
asserts:
- equal:
path: metadata.annotations["helm.sh/hook-weight"]
value: "-5"

66
helm/litellm/tests/rollout_strategy_tests.yaml Normal file
View file

@ -0,0 +1,66 @@
suite: test rolling update strategy on the component deployments
templates:
- gateway/deployment.yaml
- gateway/configmap.yaml
- backend/deployment.yaml
- ui/deployment.yaml
values:
- ./values/required.yaml
tests:
- it: leaves the strategy to Kubernetes defaults when unset
asserts:
- notExists:
path: spec.strategy
- it: renders the configured strategy on each deployment
set:
gateway.strategy:
type: RollingUpdate
rollingUpdate:
maxUnavailable: 0
maxSurge: 1
backend.strategy:
type: RollingUpdate
rollingUpdate:
maxUnavailable: "25%"
maxSurge: 2
ui.strategy:
type: Recreate
asserts:
- equal:
path: spec.strategy
value:
type: RollingUpdate
rollingUpdate:
maxUnavailable: 0
maxSurge: 1
template: gateway/deployment.yaml
- equal:
path: spec.strategy
value:
type: RollingUpdate
rollingUpdate:
maxUnavailable: 25%
maxSurge: 2
template: backend/deployment.yaml
- equal:
path: spec.strategy
value:
type: Recreate
template: ui/deployment.yaml
- it: keeps a component on the cluster default when only another one sets a strategy
set:
gateway.strategy:
type: Recreate
asserts:
- equal:
path: spec.strategy.type
value: Recreate
template: gateway/deployment.yaml
- notExists:
path: spec.strategy
template: backend/deployment.yaml
- notExists:
path: spec.strategy
template: ui/deployment.yaml

29
helm/litellm/values.yaml
View file

@ -75,6 +75,22 @@ serviceAccounts:
# generate` — the migration engine doesn't need the generated client.
migrationJob:
enabled: true
# Which controller is responsible for running the Job.
#
# `helm.enabled` renders the Helm pre-install / pre-upgrade hook, so the Job
# runs whenever `helm upgrade` sees a change to apply. `argocd.enabled`
# renders an Argo CD PreSync hook instead, which runs the Job on every sync
# even when the rendered manifests are unchanged: the way to re-run
# migrations on demand from a GitOps pipeline. Turning the Helm hook off
# while the Argo CD hook is on leaves the Job out of Helm's own upgrade
# path, which is what Argo CD users want since Argo, not Helm, applies the
# manifests.
hooks:
helm:
enabled: true
weight: "0"
argocd:
enabled: false
backoffLimit: 4
ttlSecondsAfterFinished: 120
# Wall-clock budget for the whole Job, shared across every `backoffLimit`
@ -257,6 +273,15 @@ gateway:
initialDelaySeconds: 5
periodSeconds: 10
timeoutSeconds: 10
# Rolling update tuning for the gateway Deployment. Empty by default, so
# Kubernetes applies its own RollingUpdate defaults (25% maxSurge /
# 25% maxUnavailable). Example, for a surge-only rollout behind a load
# balancer that must never lose capacity:
# type: RollingUpdate
# rollingUpdate:
# maxUnavailable: 0
# maxSurge: 1
strategy: {}
# Optional startupProbe. Empty by default, so existing installs are unchanged
# and liveness/readiness apply from container start. Set it to gate
# liveness/readiness until a slow cold start finishes — a high failureThreshold
@ -369,6 +394,8 @@ backend:
initialDelaySeconds: 5
periodSeconds: 10
timeoutSeconds: 10
# Same shape as gateway.strategy.
strategy: {}
# Optional startupProbe; same shape as gateway.startupProbe. Empty by default.
startupProbe: {}
hpa:
@ -433,6 +460,8 @@ ui:
httpGet: { path: /, port: http }
initialDelaySeconds: 2
periodSeconds: 10
# Same shape as gateway.strategy.
strategy: {}
# Optional startupProbe; same shape as gateway.startupProbe. Empty by default.
startupProbe: {}
hpa:

21
litellm-proxy-extras/litellm_proxy_extras/migrations/20260831000000_shadow_eval_typed_targets/migration.sql Normal file
View file

@ -0,0 +1,21 @@
DO $$
BEGIN
IF EXISTS (
SELECT 1 FROM information_schema.columns
WHERE table_name = 'LiteLLM_ShadowEvalJob' AND column_name = 'api_key_id'
) THEN
ALTER TABLE "LiteLLM_ShadowEvalJob" RENAME COLUMN "api_key_id" TO "target_id";
END IF;
END $$;
ALTER TABLE "LiteLLM_ShadowEvalJob" ADD COLUMN IF NOT EXISTS "target_type" TEXT NOT NULL DEFAULT 'key';
DROP INDEX IF EXISTS "LiteLLM_ShadowEvalJob_one_active_per_key_direction";
CREATE UNIQUE INDEX IF NOT EXISTS "LiteLLM_ShadowEvalJob_one_active_per_target_direction"
ON "LiteLLM_ShadowEvalJob"("target_type", "target_id", "direction") WHERE "stopped_at" IS NULL;
DROP INDEX IF EXISTS "LiteLLM_ShadowEvalJob_api_key_id_idx";
CREATE INDEX IF NOT EXISTS "LiteLLM_ShadowEvalJob_target_type_target_id_idx"
ON "LiteLLM_ShadowEvalJob"("target_type", "target_id");

3
litellm-proxy-extras/litellm_proxy_extras/migrations/20260901000000_shadow_eval_multi_router/migration.sql Normal file
View file

@ -0,0 +1,3 @@
ALTER TABLE "LiteLLM_ShadowEvalJob" ADD COLUMN IF NOT EXISTS "router_names" TEXT[] NOT NULL DEFAULT ARRAY[]::TEXT[];
ALTER TABLE "LiteLLM_ShadowEvalAttempt" ADD COLUMN IF NOT EXISTS "router_name" TEXT;

30
litellm-proxy-extras/litellm_proxy_extras/prisma_toolchain.py
View file

@ -14,9 +14,22 @@ then fails on a Node binary that was never written. Deleting a cache directory
that exists without a Node binary is what turns a killed bootstrap back into a
recoverable one.
Both budgets are overridable so an operator can widen them without a release:
``LITELLM_PRISMA_BOOTSTRAP_TIMEOUT`` for the toolchain install and
``LITELLM_PRISMA_COMMAND_TIMEOUT`` for every individual Prisma command.
``prisma migrate deploy`` is the other command whose runtime is not a
constant: it grows with the number of pending migrations, so a fresh database
that has to replay every migration this package ships overruns a per-command
budget sized for the short bookkeeping commands, on a laptop as much as on a
slow CI runner. The Python ``prisma`` wrapper spawns Node and the schema engine
as separate children, so killing the wrapper on timeout leaves them running:
the retry then contends with that orphan for Prisma's advisory lock and cannot
finish any sooner. Migrate deploy therefore runs under its own budget.
All three budgets are overridable so an operator can widen them without a
release: ``LITELLM_PRISMA_BOOTSTRAP_TIMEOUT`` for the toolchain install,
``LITELLM_PRISMA_MIGRATE_DEPLOY_TIMEOUT`` for ``prisma migrate deploy`` and
``LITELLM_PRISMA_COMMAND_TIMEOUT`` for every other Prisma command. The
per-command budget used to bound migrate deploy as well, so a deployment that
raised it above the deploy default keeps that larger budget for deploy unless
the deploy override says otherwise.
"""
import math
@ -36,10 +49,12 @@ except ImportError:
PRISMA_COMMAND_TIMEOUT_ENV_VAR = "LITELLM_PRISMA_COMMAND_TIMEOUT"
PRISMA_BOOTSTRAP_TIMEOUT_ENV_VAR = "LITELLM_PRISMA_BOOTSTRAP_TIMEOUT"
PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR = "LITELLM_PRISMA_MIGRATE_DEPLOY_TIMEOUT"
NODEENV_CACHE_DIR_ENV_VAR = "PRISMA_NODEENV_CACHE_DIR"
DEFAULT_PRISMA_COMMAND_TIMEOUT = 60.0
DEFAULT_PRISMA_BOOTSTRAP_TIMEOUT = 600.0
DEFAULT_PRISMA_MIGRATE_DEPLOY_TIMEOUT = 600.0
BOOTSTRAP_ARG = "--version"
@ -88,6 +103,15 @@ def prisma_bootstrap_timeout() -> float:
)
def prisma_migrate_deploy_timeout() -> float:
"""Seconds one ``prisma migrate deploy`` may run for, however many migrations are pending."""
if os.getenv(PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR) is not None:
return _timeout_from_env(
PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR, DEFAULT_PRISMA_MIGRATE_DEPLOY_TIMEOUT
)
return max(DEFAULT_PRISMA_MIGRATE_DEPLOY_TIMEOUT, prisma_command_timeout())
def nodeenv_cache_dir() -> Optional[Path]:
"""Where Prisma installs its private Node runtime, or None if unknowable."""
override = os.getenv(NODEENV_CACHE_DIR_ENV_VAR)

11
litellm-proxy-extras/litellm_proxy_extras/schema.prisma
View file

@ -1529,14 +1529,16 @@ model LiteLLM_AutoRouterSession {
model LiteLLM_ShadowEvalJob {
id String @id @default(cuid())
group_id String // legs of one job share this; the API's job id
api_key_id String // hashed virtual key whose traffic this leg shadows
router_name String // the auto-router under evaluation, in either direction
target_type String @default("key") // key | team | user
target_id String // hashed virtual key, team_id, or user_id whose traffic this leg shadows
router_name String // first (often only) auto-router under evaluation; router_names is the full set
router_names String[] @default([]) // all routers this job runs as shadow arms; empty on legacy rows, whose set is (router_name)
direction String @default("forward") // forward | reverse
baseline_model String? // reverse only: the fixed model the router is judged against
judge_model String
shadow_percentage Float
max_turns Int // sample-count ceiling: the whole budget on pre-max_budget jobs, the error-loop valve otherwise
max_budget Float? // per-key USD cap on the eval's own shadow + judge spend; null on jobs from before spend budgets
max_budget Float? // per-target USD cap on the eval's own shadow + judge spend; null on jobs from before spend budgets
created_at DateTime @default(now())
created_by String?
ends_at DateTime
@ -1544,7 +1546,7 @@ model LiteLLM_ShadowEvalJob {
stopped_by String? // operator who stopped it early; null when it ended on its own
@@index([group_id])
@@index([api_key_id])
@@index([target_type, target_id])
@@index([created_at])
}
@ -1554,6 +1556,7 @@ model LiteLLM_ShadowEvalAttempt {
job_id String
request_id String // the judged real request
outcome String // real | shadow | tie | error
router_name String? // the arm this verdict scores; NULL on legacy rows, meaning the job's own router
tier String? // router's tier for the prompt, when classified
real_model String?
shadow_model String?

130
litellm-proxy-extras/litellm_proxy_extras/utils.py
View file

@ -15,8 +15,11 @@ from litellm_proxy_extras.replica_identity import (
apply_replica_identity_full,
)
from litellm_proxy_extras.prisma_toolchain import (
PRISMA_COMMAND_TIMEOUT_ENV_VAR,
PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR,
ensure_prisma_toolchain,
prisma_command_timeout,
prisma_migrate_deploy_timeout,
)
@ -40,6 +43,8 @@ def _get_prisma_env() -> dict:
_MIGRATION_TS_RE = re.compile(r"^(\d{14})_")
_MIGRATION_DEADLOCK_MARKER = "deadlock detected"
_SPEND_LOGS_ALTER_RE = re.compile(r'^ALTER\s+TABLE\s+"LiteLLM_SpendLogs"\s', re.IGNORECASE)
_SPEND_LOGS_ARTIFACT_DROP_RE = re.compile(
r'^DROP\s+TABLE\s+"LiteLLM_SpendLogs_[^"]*"', re.IGNORECASE
@ -262,6 +267,50 @@ class ProxyExtrasDBManager:
env=prisma_env,
)
@staticmethod
def _roll_back_migration_best_effort(migration_name: str) -> None:
"""Mark a migration rolled back, tolerating a concurrent resolver
having already done it."""
try:
ProxyExtrasDBManager._roll_back_migration(migration_name)
except (subprocess.CalledProcessError, subprocess.TimeoutExpired):
pass
@staticmethod
def _failed_migration_logs(migration_name: str) -> Optional[str]:
"""Return failed migration logs, or None if the ledger is unavailable."""
database_url = os.getenv("DATABASE_URL")
if not database_url:
return None
try:
import psycopg
except ImportError:
return None
cleaned_url = ProxyExtrasDBManager._strip_prisma_query_params(database_url)
ledger_table = psycopg.sql.SQL("{}.{}").format(
psycopg.sql.Identifier(
ProxyExtrasDBManager._prisma_schema_param(database_url) or "public"
),
psycopg.sql.Identifier("_prisma_migrations"),
)
try:
with psycopg.connect(
cleaned_url, connect_timeout=10, autocommit=True
) as conn:
row = conn.execute(
psycopg.sql.SQL(
"SELECT logs FROM {} "
"WHERE migration_name = %s AND finished_at IS NULL "
"AND rolled_back_at IS NULL"
).format(ledger_table),
(migration_name,),
).fetchone()
except (psycopg.OperationalError, psycopg.DatabaseError):
return None
return (row[0] or "") if row else ""
@staticmethod
def _resolve_specific_migration(migration_name: str):
"""Mark a specific migration as applied"""
@ -512,6 +561,13 @@ class ProxyExtrasDBManager:
try:
import psycopg
except ImportError:
logger.warning(
"psycopg is not installed; skipping the LiteLLM_SpendLogs "
"partition check. If this table is partitioned (see "
"db_scripts/partition_spend_logs.sql), schema reconciliation "
"will try to rewrite its primary key and fail. Install the "
"litellm[extra_proxy] extra, which now includes psycopg."
)
return False
cleaned_url = ProxyExtrasDBManager._strip_prisma_query_params(database_url)
@ -651,7 +707,8 @@ class ProxyExtrasDBManager:
v2 migration resolver (opt-in via --use_v2_migration_resolver).
Runs `prisma migrate deploy` and handles standard recovery paths
(P3005 baseline, P3009/P3018 idempotent errors). Critically, it does
(P3005 baseline, P3009/P3018 idempotent errors, deadlocks against a
concurrent migrate deploy). Critically, it does
NOT call `_resolve_all_migrations` — the diff-and-force recovery that
caused schema thrashing when two LiteLLM versions contended for the
same DB during rolling deploys.
@ -691,12 +748,13 @@ class ProxyExtrasDBManager:
original_dir = os.getcwd()
os.chdir(migrations_dir)
deploy_timeout = prisma_migrate_deploy_timeout()
try:
for attempt in range(4):
try:
result = subprocess.run(
[_get_prisma_command(), "migrate", "deploy"],
timeout=prisma_command_timeout(),
timeout=deploy_timeout,
check=True,
capture_output=True,
text=True,
@ -706,8 +764,12 @@ class ProxyExtrasDBManager:
return True
except subprocess.TimeoutExpired:
logger.info(
f"prisma migrate deploy attempt {attempt + 1} timed out, retrying"
logger.warning(
"prisma migrate deploy attempt %s timed out after %ss, retrying. "
"Raise %s if this database needs longer to apply its pending migrations.",
attempt + 1,
deploy_timeout,
PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR,
)
time.sleep(random.randrange(5, 15))
continue
@ -757,6 +819,20 @@ class ProxyExtrasDBManager:
f"Detail: {resolve_err}"
) from resolve_err
continue
if migration_match:
migration_name = migration_match.group(1)
ledger_logs = ProxyExtrasDBManager._failed_migration_logs(migration_name)
if ledger_logs is not None and (
ledger_logs == "" or _MIGRATION_DEADLOCK_MARKER in ledger_logs
):
logger.info(
"Migration %s failed in a concurrent migrate deploy "
"deadlock race, rolling its ledger row back and retrying",
migration_name,
)
ProxyExtrasDBManager._roll_back_migration_best_effort(migration_name)
time.sleep(random.randrange(5, 15))
continue
raise RuntimeError(
"Database migration failed and cannot be auto-recovered. "
f"Manual intervention required.\n\nPrisma error:\n{stderr}"
@ -802,11 +878,42 @@ class ProxyExtrasDBManager:
) from resolve_err
continue
if migration_match and _MIGRATION_DEADLOCK_MARKER in stderr:
logger.info(
"Migration %s deadlocked against a concurrent "
"migrate deploy, rolling its ledger row back "
"and retrying",
migration_match.group(1),
)
ProxyExtrasDBManager._roll_back_migration_best_effort(
migration_match.group(1)
)
time.sleep(random.randrange(5, 15))
continue
raise RuntimeError(
"Database migration failed and cannot be auto-recovered. "
f"Manual intervention required.\n\nPrisma error:\n{stderr}"
) from e
if _MIGRATION_DEADLOCK_MARKER in stderr:
logger.info(
"prisma migrate deploy attempt %s deadlocked against "
"a concurrent migrate deploy, retrying",
attempt + 1,
)
time.sleep(random.randrange(5, 15))
continue
if "P1002" in stderr and "advisory lock" in stderr:
logger.info(
"prisma migrate deploy attempt %s timed out waiting for "
"the advisory lock a concurrent migrate deploy holds, retrying",
attempt + 1,
)
time.sleep(random.randrange(5, 15))
continue
raise RuntimeError(
"Database migration failed and cannot be auto-recovered. "
f"Manual intervention required.\n\nPrisma error:\n{stderr}"
@ -814,9 +921,10 @@ class ProxyExtrasDBManager:
raise RuntimeError(
"Database migration failed after 4 attempts (retry loop "
"exhausted by timeouts or repeated idempotent-recovery "
"continues). Check database connectivity, load, and "
"_prisma_migrations ledger state."
"exhausted by timeouts, deadlock retries, or repeated "
"idempotent-recovery continues). Check database connectivity, "
"load, and _prisma_migrations ledger state, and raise "
f"{PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR} if the attempts timed out."
)
finally:
os.chdir(original_dir)
@ -901,7 +1009,7 @@ class ProxyExtrasDBManager:
# Set migrations directory for Prisma
result = subprocess.run(
[_get_prisma_command(), "migrate", "deploy"],
timeout=prisma_command_timeout(),
timeout=prisma_migrate_deploy_timeout(),
check=True,
capture_output=True,
text=True,
@ -1119,7 +1227,11 @@ class ProxyExtrasDBManager:
)
return True
except subprocess.TimeoutExpired:
logger.info(f"Attempt {attempt + 1} timed out")
logger.warning(
"Attempt %s timed out. Raise %s if this database needs longer to apply its schema.",
attempt + 1,
PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR if use_migrate else PRISMA_COMMAND_TIMEOUT_ENV_VAR,
)
time.sleep(random.randrange(5, 15))
except subprocess.CalledProcessError as e:
attempts_left = 3 - attempt

4
litellm-proxy-extras/pyproject.toml
View file

@ -1,6 +1,6 @@
[project]
name = "litellm-proxy-extras"
version = "0.4.91"
version = "0.4.92"
description = "Additional files for the LiteLLM Proxy. Reduces the size of the main litellm package."
readme = "README.md"
requires-python = ">=3.9"
@ -26,7 +26,7 @@ required-version = ">=0.10.9"
module-root = ""
[tool.commitizen]
version = "0.4.91"
version = "0.4.92"
version_files = [
"pyproject.toml:^version",
"../pyproject.toml:litellm-proxy-extras==",

220
litellm-proxy-extras/tests/test_setup_database_fail_fast.py
View file

@ -240,3 +240,223 @@ def test_v2_does_not_call_resolve_all_migrations(monkeypatch, tmp_path):
ok = ProxyExtrasDBManager.setup_database(use_migrate=True, use_v2_resolver=True)
assert ok is True
assert resolve_called["n"] == 0, "v2 must not invoke the diff-and-force recovery"
_DEADLOCK_P3018_STDERR = (
"Error: P3018\n"
"Migration name: 20260415120000_health_check_latest_per_model_index\n"
"Database error code: 40P01\n"
"deadlock detected"
)
def _stub_v2_env(monkeypatch, tmp_path):
monkeypatch.setenv("DATABASE_URL", "postgresql://u:p@localhost:9/x")
monkeypatch.setattr(
ProxyExtrasDBManager, "_warn_if_db_ahead_of_head", lambda _: None
)
monkeypatch.setattr(ProxyExtrasDBManager, "_get_prisma_dir", lambda: str(tmp_path))
(tmp_path / "schema.prisma").write_text("// stub")
monkeypatch.setattr("time.sleep", lambda _: None)
def _succeed_after(failures: int, stderr: str):
calls = {"n": 0}
class _OkResult:
stdout = "Applied migration.\n"
stderr = ""
def _run(*args, **kwargs):
if "deploy" not in args[0]:
return _OkResult()
calls["n"] += 1
if calls["n"] <= failures:
raise subprocess.CalledProcessError(
returncode=1, cmd=args[0], stderr=stderr, output=""
)
return _OkResult()
return _run
def test_v2_p3018_deadlock_rolls_back_and_retries(monkeypatch, tmp_path):
"""v2: losing the migrate deploy deadlock race against a concurrent
instance rolls the ledger row back and retries instead of dying."""
_stub_v2_env(monkeypatch, tmp_path)
rolled_back = []
monkeypatch.setattr(
ProxyExtrasDBManager,
"_roll_back_migration",
lambda name: rolled_back.append(name),
)
monkeypatch.setattr(
ProxyExtrasDBManager,
"_resolve_specific_migration",
lambda name: pytest.fail("a deadlocked migration must never be marked applied"),
)
monkeypatch.setattr("subprocess.run", _succeed_after(1, _DEADLOCK_P3018_STDERR))
ok = ProxyExtrasDBManager.setup_database(use_migrate=True, use_v2_resolver=True)
assert ok is True
assert rolled_back == ["20260415120000_health_check_latest_per_model_index"]
def test_v2_p3018_persistent_deadlock_exhausts_attempts(monkeypatch, tmp_path):
"""v2: a deadlock on every attempt still fails after the retry budget."""
_stub_v2_env(monkeypatch, tmp_path)
monkeypatch.setattr(ProxyExtrasDBManager, "_roll_back_migration", lambda name: None)
with patch(
"subprocess.run",
side_effect=_fake_migrate_deploy_failure(1, _DEADLOCK_P3018_STDERR),
):
with pytest.raises(RuntimeError, match="after 4 attempts"):
ProxyExtrasDBManager.setup_database(use_migrate=True, use_v2_resolver=True)
def test_v2_p3009_deadlocked_ledger_row_rolls_back_and_retries(monkeypatch, tmp_path):
"""v2: the surviving instance sees the victim's failed ledger row as P3009.
When that row's logs show a deadlock, roll it back and retry."""
_stub_v2_env(monkeypatch, tmp_path)
stderr = (
"Error: P3009\n"
"migrate found failed migrations in the target database\n"
"The `20260415120000_health_check_latest_per_model_index` migration "
"started at 2026-09-01 18:46:13 UTC failed"
)
monkeypatch.setattr(
ProxyExtrasDBManager,
"_failed_migration_logs",
lambda name: "ERROR: deadlock detected\nDETAIL: Process 72 waits for ShareLock",
)
rolled_back = []
monkeypatch.setattr(
ProxyExtrasDBManager,
"_roll_back_migration",
lambda name: rolled_back.append(name),
)
monkeypatch.setattr(
ProxyExtrasDBManager,
"_resolve_specific_migration",
lambda name: pytest.fail("a deadlocked migration must never be marked applied"),
)
monkeypatch.setattr("subprocess.run", _succeed_after(1, stderr))
ok = ProxyExtrasDBManager.setup_database(use_migrate=True, use_v2_resolver=True)
assert ok is True
assert rolled_back == ["20260415120000_health_check_latest_per_model_index"]
def test_v2_p3009_empty_ledger_logs_rolls_back_and_retries(monkeypatch, tmp_path):
"""v2: empty failed ledger logs mean a concurrent deploy moved it on."""
_stub_v2_env(monkeypatch, tmp_path)
stderr = (
"Error: P3009\n"
"migrate found failed migrations in the target database\n"
"The `20260415120000_health_check_latest_per_model_index` migration "
"started at 2026-09-01 18:46:13 UTC failed"
)
monkeypatch.setattr(ProxyExtrasDBManager, "_failed_migration_logs", lambda name: "")
rolled_back = []
monkeypatch.setattr(
ProxyExtrasDBManager,
"_roll_back_migration",
lambda name: rolled_back.append(name),
)
monkeypatch.setattr(
ProxyExtrasDBManager,
"_resolve_specific_migration",
lambda name: pytest.fail("a deadlocked migration must never be marked applied"),
)
monkeypatch.setattr("subprocess.run", _succeed_after(1, stderr))
ok = ProxyExtrasDBManager.setup_database(use_migrate=True, use_v2_resolver=True)
assert ok is True
assert rolled_back == ["20260415120000_health_check_latest_per_model_index"]
def test_v2_p3009_unreadable_ledger_still_raises(monkeypatch, tmp_path):
"""v2: an unreadable ledger cannot establish that P3009 was a deadlock."""
_stub_v2_env(monkeypatch, tmp_path)
stderr = (
"Error: P3009\n"
"migrate found failed migrations in the target database\n"
"The `20260415120000_health_check_latest_per_model_index` migration "
"started at 2026-09-01 18:46:13 UTC failed"
)
monkeypatch.setattr(ProxyExtrasDBManager, "_failed_migration_logs", lambda name: None)
monkeypatch.setattr(
ProxyExtrasDBManager,
"_roll_back_migration",
lambda name: pytest.fail("an unreadable ledger must not trigger a retry"),
)
monkeypatch.setattr("subprocess.run", _succeed_after(1, stderr))
with pytest.raises(RuntimeError, match="cannot be auto-recovered"):
ProxyExtrasDBManager.setup_database(use_migrate=True, use_v2_resolver=True)
def test_v2_p3009_non_deadlock_ledger_row_still_raises(monkeypatch, tmp_path):
"""v2: a failed ledger row whose logs show a real SQL error stays fatal."""
_stub_v2_env(monkeypatch, tmp_path)
stderr = (
"Error: P3009\n"
"migrate found failed migrations in the target database\n"
"The `20260101000000_genuinely_broken` migration started at "
"2026-09-01 18:46:13 UTC failed"
)
monkeypatch.setattr(
ProxyExtrasDBManager,
"_failed_migration_logs",
lambda name: 'ERROR: syntax error at or near "BRKN"',
)
with patch("subprocess.run", side_effect=_fake_migrate_deploy_failure(1, stderr)):
with pytest.raises(RuntimeError, match="cannot be auto-recovered"):
ProxyExtrasDBManager.setup_database(use_migrate=True, use_v2_resolver=True)
def test_v2_bare_deadlock_stderr_retries(monkeypatch, tmp_path):
"""v2: a deadlock reported without a Prisma error code (the advisory-lock
waiter as victim) is retried, not fatal."""
_stub_v2_env(monkeypatch, tmp_path)
monkeypatch.setattr(
"subprocess.run", _succeed_after(1, "Database error: deadlock detected")
)
ok = ProxyExtrasDBManager.setup_database(use_migrate=True, use_v2_resolver=True)
assert ok is True
_P1002_ADVISORY_LOCK_STDERR = (
"Error: P1002\n\n"
"The database server at `127.0.0.1`:`45743` was reached but timed out.\n\n"
"Context: Timed out trying to acquire a postgres advisory lock "
"(SELECT pg_advisory_lock(72707369)). Elapsed: 10000ms."
)
def test_v2_advisory_lock_timeout_retries(monkeypatch, tmp_path):
"""v2: the advisory-lock waiter that times out while a peer's retry holds
the lock retries instead of dying."""
_stub_v2_env(monkeypatch, tmp_path)
monkeypatch.setattr("subprocess.run", _succeed_after(2, _P1002_ADVISORY_LOCK_STDERR))
ok = ProxyExtrasDBManager.setup_database(use_migrate=True, use_v2_resolver=True)
assert ok is True
def test_v2_p1002_without_advisory_lock_context_still_raises(monkeypatch, tmp_path):
"""v2: a plain P1002 (database unreachable) stays fatal."""
_stub_v2_env(monkeypatch, tmp_path)
stderr = "Error: P1002\n\nThe database server at `db`:`5432` was reached but timed out."
monkeypatch.setattr("subprocess.run", _succeed_after(1, stderr))
with pytest.raises(RuntimeError, match="cannot be auto-recovered"):
ProxyExtrasDBManager.setup_database(use_migrate=True, use_v2_resolver=True)

9
litellm-rust/AGENTS.md
View file

@ -1,6 +1,6 @@
# AGENTS.md
litellm-rust has exactly THREE crates. A crate is a LAYER, not a route. Routes (ocr, realtime, chat) and providers (mistral, openai) are MODULES inside the layers.
litellm-rust has four crates. A crate is a layer or shared foundation, not a route. Routes (ocr, realtime, chat) and providers (mistral, openai) are modules inside the layers.
## Crates
@ -8,9 +8,10 @@ litellm-rust has exactly THREE crates. A crate is a LAYER, not a route. Routes (
|-------|------|
| litellm-core | The LiteLLM SDK in Rust. One public entrypoint per top-level call (`messages::messages()`), owning types, transforms, provider resolution, auth, and the provider HTTP call. Call it, get a typed response. |
| litellm-ai-gateway | The axum server (behind the `server` feature) plus the WebSocket hosts. Translates HTTP/WS to core entrypoints; owns no provider logic and no handlers. |
| litellm-python-bridge | PyO3 cdylib exposing Rust to the litellm Python SDK — marshals Python objects and calls core entrypoints. |
| litellm-python-interop | Domain-neutral PyO3 foundation for GIL handling and typed Python/Serde conversion. |
| litellm-python-bridge | PyO3 cdylib exposing LiteLLM Rust APIs to the Python SDK. Owns API registration, domain wiring, and Python exception mapping. |
Dependency direction (acyclic): litellm-core ← litellm-ai-gateway ← litellm-python-bridge.
Dependency direction is acyclic: `litellm-python-bridge` depends on the domain layers and `litellm-python-interop`; the interop foundation depends on no LiteLLM domain crate.
## Where a route lives
@ -28,7 +29,7 @@ core/src/messages/
Handlers never live in `ai-gateway`. `ocr`, `audio_transcription`, and `realtime` are still hosted there from before this rule; they move to `core` as they are touched.
Adding a crate: default to a MODULE. New crate ONLY on a real trigger — separate artifact (binary/cdylib), proc-macro, shared foundation, or publishable standalone. A new provider or route is none of these.
Adding a crate: default to a module. A new crate requires a real trigger: separate artifact (binary/cdylib), proc-macro, shared foundation, or publishable standalone. A new provider or route is none of these.
Adding a crate fails crates/core/tests/workspace_crate_allowlist.rs until you update its allowlist and this file — intentional.

9
litellm-rust/CLAUDE.md
View file

@ -21,12 +21,13 @@ variants of it. The test for a good abstraction is that adding the next provider
is a few declarative lines, not a new file of duplicated flow. Only diverge from
the base when behavior is genuinely different, and say so explicitly in the PR.
## Crates (exactly three — see AGENTS.md)
## Crates (see AGENTS.md)
`litellm-core` **is** the LiteLLM SDK in Rust: it makes the LLM call.
`litellm-ai-gateway` is an HTTP/WebSocket server in front of it, and
`litellm-python-bridge` exposes it to the Python SDK. A crate is a **layer**, not
a route — add modules, not crates.
`litellm-python-bridge` exposes it to the Python SDK. `litellm-python-interop`
holds domain-neutral PyO3 primitives shared by Python-facing Rust code. A crate
is a layer or shared foundation, not a route; add modules, not crates.
## Core Boundary
@ -175,7 +176,7 @@ cd litellm-rust
cargo fmt --check
# the ai-gateway binary + server code is behind the `server` feature
cargo clippy -p litellm-ai-gateway --all-targets --features server -- -D warnings
cargo clippy -p litellm-core -p litellm-python-bridge --all-targets -- -D warnings
cargo clippy -p litellm-core -p litellm-python-interop -p litellm-python-bridge --all-targets -- -D warnings
cargo test --workspace
```

169
litellm-rust/Cargo.lock generated
View file

@ -919,6 +919,12 @@ version = "0.3.33"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "b231ed28831efb4a61a08580c4bc233ec56bc009f4cd8f52da2c3cb97df0c109"
[[package]]
name = "futures-timer"
version = "3.0.4"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "af43fadb8a98512d547e37b4e92e0ced13e205c061b87b4623eff01d918d6968"
[[package]]
name = "futures-util"
version = "0.3.33"
@ -972,6 +978,12 @@ dependencies = [
"wasm-bindgen",
]
[[package]]
name = "glob"
version = "0.3.4"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "e4eba85ea1d0a966a983acd07deee566e67395d2d96b6fb39e62b5a833f1eb0b"
[[package]]
name = "h2"
version = "0.3.27"
@ -1380,6 +1392,12 @@ dependencies = [
"wasm-bindgen",
]
[[package]]
name = "lazy_static"
version = "1.5.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "bbd2bcb4c963f2ddae06a2efc7e9f3591312473c50c6685e1f298068316e66fe"
[[package]]
name = "libc"
version = "0.2.186"
@ -1404,6 +1422,7 @@ dependencies = [
"tokio",
"tokio-tungstenite",
"tower",
"tracing",
]
[[package]]
@ -1423,6 +1442,7 @@ dependencies = [
"sha2 0.10.9",
"thiserror 2.0.19",
"tokio",
"tracing",
]
[[package]]
@ -1430,14 +1450,29 @@ name = "litellm-python-bridge"
version = "0.1.0"
dependencies = [
"criterion",
"futures-util",
"litellm-ai-gateway",
"litellm-core",
"litellm-python-interop",
"pyo3",
"pyo3-async-runtimes",
"pythonize",
"serde",
"serde_json",
"tokio",
"tokio-tungstenite",
"tracing",
"tracing-subscriber",
]
[[package]]
name = "litellm-python-interop"
version = "0.1.0"
dependencies = [
"pyo3",
"pythonize",
"rstest",
"serde",
"serde_json",
]
[[package]]
@ -1627,6 +1662,15 @@ dependencies = [
"zerocopy",
]
[[package]]
name = "proc-macro-crate"
version = "3.5.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "e67ba7e9b2b56446f1d419b1d807906278ffa1a658a8a5d8a39dcb1f5a78614f"
dependencies = [
"toml_edit",
]
[[package]]
name = "proc-macro2"
version = "1.0.107"
@ -1638,9 +1682,9 @@ dependencies = [
[[package]]
name = "pyo3"
version = "0.29.0"
version = "0.29.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "cd274650b21d4bfc26a0a47587962c1edb425f69287324355cd040c3ea66071c"
checksum = "4688ddedf473e32662b9b067670129a8afb8c18e351482c70d62ba4a88171e8b"
dependencies = [
"libc",
"once_cell",
@ -1666,18 +1710,18 @@ dependencies = [
[[package]]
name = "pyo3-build-config"
version = "0.29.0"
version = "0.29.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "c5e2a7d2f0d013342f295c048ad19237add5154a55b1c5a254c0ec93d4109078"
checksum = "f41027e41b4bd03f6e60f9f417fe24a6341a6bb744edd62b6f709f2a52ea30e9"
dependencies = [
"target-lexicon",
]
[[package]]
name = "pyo3-ffi"
version = "0.29.0"
version = "0.29.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "ca85c467da1bbc8d866eea5deff9cf29ea5f7785054a17da36e65bda9c05845b"
checksum = "e591a95526fead067432c3b3a33fc74770b87b1e04e73671090d9c2055a2b327"
dependencies = [
"libc",
"pyo3-build-config",
@ -1685,9 +1729,9 @@ dependencies = [
[[package]]
name = "pyo3-macros"
version = "0.29.0"
version = "0.29.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "9ac53762fd065daa3194dd09337a38bd793a188100fd1a9304c4ab312d901771"
checksum = "73225868fc1cd84eef2c3c230ddb91273bf1de46aeb8a4248da76d32a0924a1c"
dependencies = [
"proc-macro2",
"pyo3-macros-backend",
@ -1697,9 +1741,9 @@ dependencies = [
[[package]]
name = "pyo3-macros-backend"
version = "0.29.0"
version = "0.29.2"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "4ca3a1557399783172dc5bf39cfca835157732532cba56b71d2292161e53b362"
checksum = "571575aa3749fa6216757dd47d2a3e7ef360f329a40f0666a9fbd14889024952"
dependencies = [
"heck",
"proc-macro2",
@ -1899,6 +1943,12 @@ version = "0.8.11"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "d6f6ff9a378485b298a5286656da665ba74413d36db0979633275d2e708145d4"
[[package]]
name = "relative-path"
version = "1.9.3"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "ba39f3699c378cd8970968dcbff9c43159ea4cfbd88d43c00b22f2ef10a435d2"
[[package]]
name = "reqwest"
version = "0.12.28"
@ -1956,6 +2006,35 @@ dependencies = [
"windows-sys 0.52.0",
]
[[package]]
name = "rstest"
version = "0.26.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "f5a3193c063baaa2a95a33f03035c8a72b83d97a54916055ba22d35ed3839d49"
dependencies = [
"futures-timer",
"futures-util",
"rstest_macros",
]
[[package]]
name = "rstest_macros"
version = "0.26.1"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "9c845311f0ff7951c5506121a9ad75aec44d083c31583b2ea5a30bcb0b0abba0"
dependencies = [
"cfg-if",
"glob",
"proc-macro-crate",
"proc-macro2",
"quote",
"regex",
"relative-path",
"rustc_version",
"syn 2.0.119",
"unicode-ident",
]
[[package]]
name = "rustc-hash"
version = "2.1.3"
@ -2210,6 +2289,15 @@ dependencies = [
"digest 0.11.3",
]
[[package]]
name = "sharded-slab"
version = "0.1.7"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "f40ca3c46823713e0d4209592e8d6e826aa57e928f09752619fc696c499637f6"
dependencies = [
"lazy_static",
]
[[package]]
name = "shlex"
version = "2.0.1"
@ -2348,6 +2436,15 @@ dependencies = [
"syn 3.0.0",
]
[[package]]
name = "thread_local"
version = "1.1.10"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "1ad99c4c6d32803332c548b1af0540b357b3f5fc0be8f6c6bfe8b2e6ae784070"
dependencies = [
"cfg-if",
]
[[package]]
name = "time"
version = "0.3.53"
@ -2488,6 +2585,36 @@ dependencies = [
"tokio",
]
[[package]]
name = "toml_datetime"
version = "1.1.1+spec-1.1.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "3165f65f62e28e0115a00b2ebdd37eb6f3b641855f9d636d3cd4103767159ad7"
dependencies = [
"serde_core",
]
[[package]]
name = "toml_edit"
version = "0.25.13+spec-1.1.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "6975367e4d2ef766d86af01ffad14b622fecc8d4357a998fbc4deb6e9bacaf9b"
dependencies = [
"indexmap",
"toml_datetime",
"toml_parser",
"winnow",
]
[[package]]
name = "toml_parser"
version = "1.1.3+spec-1.1.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "1d38ac1cf9b95face32296c0a3ede1fdc270627c9d9c02a7274dd6d960dc4d56"
dependencies = [
"winnow",
]
[[package]]
name = "tower"
version = "0.5.3"
@ -2566,6 +2693,17 @@ dependencies = [
"once_cell",
]
[[package]]
name = "tracing-subscriber"
version = "0.3.23"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "cb7f578e5945fb242538965c2d0b04418d38ec25c79d160cd279bf0731c8d319"
dependencies = [
"sharded-slab",
"thread_local",
"tracing-core",
]
[[package]]
name = "try-lock"
version = "0.2.5"
@ -2903,6 +3041,15 @@ version = "0.52.6"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "589f6da84c646204747d1270a2a5661ea66ed1cced2631d546fdfb155959f9ec"
[[package]]
name = "winnow"
version = "1.0.4"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "23b97319f7b8343df12cc98938e5c3eb436064524c8d2b4e30a1d3a36eecdf81"
dependencies = [
"memchr",
]
[[package]]
name = "writeable"
version = "0.6.3"

16
litellm-rust/Cargo.toml
View file

@ -2,6 +2,7 @@
members = [
"crates/core",
"crates/ai-gateway",
"crates/python-interop",
"crates/python-bridge",
]
resolver = "2"
@ -13,14 +14,18 @@ license = "MIT"
repository = "https://github.com/BerriAI/litellm"
[workspace.dependencies]
tracing = "0.1"
tracing-subscriber = { version = "0.3", default-features = false, features = ["registry", "std"] }
litellm-core = { path = "crates/core" }
litellm-ai-gateway = { path = "crates/ai-gateway", default-features = false }
litellm-python-interop = { path = "crates/python-interop" }
axum = "0.7"
pyo3 = "0.29.0"
pyo3 = "0.29.2"
pyo3-async-runtimes = { version = "0.29.0", features = ["tokio-runtime"] }
pythonize = "0.29.0"
rand = "0.8"
reqwest = { version = "0.12", default-features = false, features = ["blocking", "json", "rustls-tls", "http2", "stream"] }
rstest = "0.26.1"
serde = { version = "1.0", features = ["derive"] }
serde_json = "1.0"
sha2 = "0.10"
@ -30,3 +35,12 @@ tokio = { version = "1", features = ["rt-multi-thread", "macros", "time", "net"]
tokio-tungstenite = { version = "0.24", default-features = false, features = ["connect", "rustls-tls-native-roots"] }
futures-util = { version = "0.3", default-features = false, features = ["sink", "std"] }
base64 = "0.22"
[profile.release]
opt-level = 3
lto = "thin"
codegen-units = 1
panic = "unwind"
debug = false
incremental = false
strip = "symbols"

8
litellm-rust/README.md
View file

@ -26,9 +26,10 @@ coverage and production evidence.
|-------|------|
| litellm-core | The SDK. Per-route entrypoints (`messages::messages()`), types, provider transforms (modules under `providers/`), provider resolution, auth, the provider HTTP call, and the router. |
| litellm-ai-gateway | The axum server (behind the `server` feature) and WebSocket hosts. Translates HTTP/WS to core entrypoints; no provider handlers. |
| litellm-python-bridge | PyO3 cdylib exposing Rust to the litellm Python SDK — marshals Python objects and calls core entrypoints. |
| litellm-python-interop | Domain-neutral PyO3 foundation for GIL handling and typed Python/Serde conversion. |
| litellm-python-bridge | PyO3 cdylib exposing LiteLLM Rust APIs to the Python SDK. Owns API registration, domain wiring, and Python exception mapping. |
Dependency direction (acyclic): litellm-core ← litellm-ai-gateway ← litellm-python-bridge.
Dependency direction is acyclic: `litellm-python-bridge` depends on the domain layers and `litellm-python-interop`; the interop foundation depends on no LiteLLM domain crate.
## Layout
@ -38,7 +39,8 @@ crates/
src/messages/ mod.rs (entrypoint), types, transformation, prepare, handler, client
src/providers/anthropic/messages/transformation.rs
ai-gateway/ Axum server + WebSocket hosts; calls core entrypoints.
python-bridge/ PyO3 bridge for Python LiteLLM.
python-interop/ Domain-neutral PyO3 conversion and GIL primitives.
python-bridge/ PyO3 API adapter for Python LiteLLM.
```
The folder shape follows the Python provider tree:

2
litellm-rust/crates/CODING_STANDARDS/PROVIDER_CODING_STANDARDS.md
View file

@ -54,6 +54,6 @@ Rules for adding or changing an LLM provider/route in `litellm-rust`. `messages`
cd litellm-rust
cargo fmt --check
cargo clippy -p litellm-ai-gateway --all-targets --features server -- -D warnings
cargo clippy -p litellm-core -p litellm-python-bridge --all-targets -- -D warnings
cargo clippy -p litellm-core -p litellm-python-interop -p litellm-python-bridge --all-targets -- -D warnings
cargo test --workspace
```

1
litellm-rust/crates/ai-gateway/Cargo.toml
View file

@ -14,6 +14,7 @@ path = "src/main.rs"
required-features = ["server"]
[dependencies]
tracing.workspace = true
litellm-core = { workspace = true, features = ["bedrock-auth"] }
# reqwest (rustls + json) is used by io/ocr and ships realtime logs to the
# Python proxy callbacks API.

7
litellm-rust/crates/ai-gateway/README.md
View file

@ -6,15 +6,16 @@ dials OpenAI upstream, and splices the two sockets frame-by-frame.
## Crates
`litellm-rust` is exactly three crates (a crate is a **layer**, not a route):
`litellm-rust` has four crates. A crate is a layer or shared foundation, not a route:
| Crate | Role |
|-------|------|
| litellm-core | The LiteLLM SDK in Rust — per-route entrypoints (`messages::messages()`) that resolve the provider, transform, and make the call; plus types, provider transforms, and the router. |
| litellm-ai-gateway | The Axum server (behind the `server` feature) and WebSocket hosts. Translates HTTP/WS to core entrypoints; no provider handlers. |
| litellm-python-bridge | PyO3 cdylib exposing Rust to the litellm Python SDK — marshals Python objects and calls core entrypoints. |
| litellm-python-interop | Domain-neutral PyO3 foundation for GIL handling and typed Python/Serde conversion. |
| litellm-python-bridge | PyO3 cdylib exposing LiteLLM Rust APIs to the Python SDK. |
Dependency direction (acyclic): litellm-core ← litellm-ai-gateway ← litellm-python-bridge.
Dependency direction is acyclic: `litellm-python-bridge` depends on the domain layers and `litellm-python-interop`; the interop foundation depends on no LiteLLM domain crate.
- **Client endpoint:** `wss://<host>/v1/realtime?model=<model>` (WebSocket)
- **Auth:** `Authorization: Bearer $LITELLM_MASTER_KEY` (fails closed if unset)

48
litellm-rust/crates/ai-gateway/src/audio_transcription/common_utils.rs
View file

@ -1,48 +0,0 @@
use std::collections::BTreeMap;
use litellm_core::CoreResult;
use litellm_core::audio_transcription::transformation::AudioTranscriptionProviderConfig;
use litellm_core::error::CoreError;
use litellm_core::providers::bedrock::audio_transcription::BEDROCK_AUDIO_TRANSCRIPTION_CONFIG;
use serde_json::{Map, Value};
pub(super) fn audio_transcription_provider_config(
provider: &str,
) -> Option<&'static dyn AudioTranscriptionProviderConfig> {
match provider {
"bedrock" => Some(&BEDROCK_AUDIO_TRANSCRIPTION_CONFIG),
_ => None,
}
}
pub(super) fn string_headers(
headers: Option<Map<String, Value>>,
) -> CoreResult<BTreeMap<String, String>> {
headers
.unwrap_or_default()
.into_iter()
.map(|(key, value)| {
value
.as_str()
.map(|value| (key.clone(), value.to_string()))
.ok_or_else(|| {
CoreError::InvalidRequest(format!(
"audio transcription extra_headers.{key} must be a string"
))
})
})
.collect()
}
pub(super) fn has_header(headers: &BTreeMap<String, String>, name: &str) -> bool {
headers.keys().any(|key| key.eq_ignore_ascii_case(name))
}
pub(super) fn truncate_error_body(body: &str) -> String {
let truncated: String = body.chars().take(256).collect();
if truncated.chars().count() == body.chars().count() {
truncated
} else {
format!("{truncated}... (truncated)")
}
}

89
litellm-rust/crates/ai-gateway/src/audio_transcription/handler.rs
View file

@ -1,89 +0,0 @@
use std::time::SystemTime;
use litellm_core::CoreResult;
use litellm_core::audio_transcription::transformation::AudioTranscriptionAuth;
use litellm_core::error::CoreError;
use litellm_core::providers::bedrock::audio_transcription::aws_auth_config;
use litellm_core::providers::bedrock::aws_base::{resolve_credentials, sign_bedrock_post};
use serde_json::Value;
use super::common_utils::truncate_error_body;
use super::types::ProviderAudioTranscriptionRequest;
use crate::client::http_client;
pub(crate) async fn execute_audio_transcription_provider_call(
request: ProviderAudioTranscriptionRequest,
) -> CoreResult<Value> {
let body = serde_json::to_vec(&request.body).map_err(|error| {
CoreError::InvalidRequest(format!("invalid audio request body: {error}"))
})?;
let mut request_builder = http_client().post(&request.url).body(body.clone());
for (key, value) in &request.upstream_headers {
request_builder = request_builder.header(key, value);
}
if let Some(duration) = request.timeout {
request_builder = request_builder.timeout(duration);
}
let response = request_builder
.send()
.await
.map_err(|error| CoreError::Network(error.to_string()))?;
let status = response.status();
let text = response
.text()
.await
.map_err(|error| CoreError::Network(error.to_string()))?;
if !status.is_success() {
return Err(CoreError::Http {
status: status.as_u16(),
body: truncate_error_body(&text),
});
}
let response_json: Value = serde_json::from_str(&text).map_err(|error| {
CoreError::InvalidResponse(format!("invalid audio response JSON: {error}"))
})?;
Ok(request
.config
.transform_transcription_response(&request.model, response_json)?
.into_json())
}
pub(crate) async fn sign_request(
request: &ProviderAudioTranscriptionRequest,
optional_params: &serde_json::Map<String, Value>,
) -> CoreResult<ProviderAudioTranscriptionRequest> {
let env_lookup = environment_lookup;
let auth = request
.config
.auth_strategy(&request.model, optional_params, &env_lookup)?;
let body = serde_json::to_vec(&request.body).map_err(|error| {
CoreError::InvalidRequest(format!("invalid audio request body: {error}"))
})?;
let mut headers = super::common_utils::string_headers(None)?;
headers.insert("Content-Type".to_string(), "application/json".to_string());
headers.extend(request.upstream_headers.iter().cloned());
match auth {
AudioTranscriptionAuth::Bearer => {}
AudioTranscriptionAuth::AwsSigV4 { region, .. } => {
let credentials =
resolve_credentials(aws_auth_config(optional_params, &env_lookup), &env_lookup)
.await?;
headers.extend(sign_bedrock_post(
&request.url,
&body,
&headers,
&region,
&credentials,
SystemTime::now(),
)?);
}
}
Ok(ProviderAudioTranscriptionRequest {
upstream_headers: headers.into_iter().collect(),
..request.clone()
})
}
pub(super) fn environment_lookup(key: &str) -> Option<String> {
std::env::var(key).ok()
}

138
litellm-rust/crates/ai-gateway/src/audio_transcription/hooks.rs
View file

@ -1,15 +1,14 @@
use litellm_core::audio_transcription::{
AudioTranscriptionRequest as CoreAudioTranscriptionRequest, ProviderAudioTranscriptionRequest,
prepare_audio_transcription_provider_call,
};
use litellm_core::call_lifecycle::{CallLifecycleContext, CallLifecycleHooks, CallLifecycleTiming};
use litellm_core::error::Error;
use serde_json::{Map, Value, json};
use std::future::Future;
use std::pin::Pin;
use litellm_core::CoreResult;
use litellm_core::audio_transcription::transformation::AudioTranscriptionAuth;
use litellm_core::call_lifecycle::{CallLifecycleContext, CallLifecycleHooks, CallLifecycleTiming};
use litellm_core::error::CoreError;
use serde_json::{Map, Value, json};
use super::common_utils::{audio_transcription_provider_config, has_header, string_headers};
use super::handler::sign_request;
use super::types::{PreparedAudioTranscriptionRequest, ProviderAudioTranscriptionRequest};
use super::types::PreparedAudioTranscriptionRequest;
use crate::integrations::custom_guardrail::{
CustomGuardrailRunner, GuardrailContext, GuardrailError, GuardrailRequest,
};
@ -26,7 +25,7 @@ pub(crate) struct AudioTranscriptionLifecycleHooks {
request_metadata: RequestMetadata,
}
type AudioFuture<'a, T> = Pin<Box<dyn Future<Output = CoreResult<T>> + Send + 'a>>;
type AudioFuture<'a, T> = Pin<Box<dyn Future<Output = Result<T, Error>> + Send + 'a>>;
type AudioLogFuture<'a> = Pin<Box<dyn Future<Output = ()> + Send + 'a>>;
impl AudioTranscriptionLifecycleHooks {
@ -45,7 +44,7 @@ impl AudioTranscriptionLifecycleHooks {
async fn run_pre_call_guardrails(
&self,
request: PreparedAudioTranscriptionRequest,
) -> CoreResult<PreparedAudioTranscriptionRequest> {
) -> Result<PreparedAudioTranscriptionRequest, Error> {
if self.guardrail_runner.is_empty() {
return Ok(request);
}
@ -63,17 +62,17 @@ impl AudioTranscriptionLifecycleHooks {
.await
.map_err(guardrail_error_to_core_error)?;
let Value::Object(mut data) = guardrail_request.data else {
return Err(CoreError::InvalidRequest(
return Err(Error::InvalidRequest(
"audio transcription pre_call guardrail must return an object".to_string(),
));
};
let audio = data.remove("audio").ok_or_else(|| {
CoreError::InvalidRequest("audio transcription guardrail removed audio".to_string())
Error::InvalidRequest("audio transcription guardrail removed audio".to_string())
})?;
let optional_params = match data.remove("optional_params") {
Some(Value::Object(value)) => value,
Some(_) => {
return Err(CoreError::InvalidRequest(
return Err(Error::InvalidRequest(
"audio transcription optional_params must be an object".to_string(),
));
}
@ -89,53 +88,36 @@ impl AudioTranscriptionLifecycleHooks {
async fn prepare_provider_request(
&self,
request: PreparedAudioTranscriptionRequest,
) -> CoreResult<ProviderAudioTranscriptionRequest> {
let config = audio_transcription_provider_config(&request.custom_llm_provider)
.ok_or_else(|| CoreError::InvalidProvider(request.custom_llm_provider.clone()))?;
let env_lookup = super::handler::environment_lookup;
let headers = string_headers(request.extra_headers)?;
let url = config.complete_url(
request.api_base.as_deref(),
&request.model,
&request.optional_params,
&env_lookup,
)?;
let filtered_params = config.map_transcription_params(&request.optional_params);
let body = config.transform_transcription_request(
&request.model,
request.audio,
filtered_params,
)?;
let auth = config.auth_strategy(&request.model, &request.optional_params, &env_lookup)?;
let mut upstream_headers = headers.into_iter().collect::<Vec<_>>();
if matches!(auth, AudioTranscriptionAuth::Bearer)
&& !has_header(
&upstream_headers
.iter()
.cloned()
.collect::<std::collections::BTreeMap<_, _>>(),
"authorization",
)
&& let Some(api_key) = request.api_key.as_deref()
{
upstream_headers.push(("Authorization".to_string(), format!("Bearer {api_key}")));
}
let provider_request = ProviderAudioTranscriptionRequest {
model: request.model,
config,
url,
body: body.body,
upstream_headers,
timeout: request.timeout,
};
let provider_request = self.run_during_call_guardrails(provider_request).await?;
sign_request(&provider_request, &request.optional_params).await
) -> Result<ProviderAudioTranscriptionRequest, Error> {
let PreparedAudioTranscriptionRequest {
model,
custom_llm_provider,
audio,
api_key,
api_base,
extra_headers,
optional_params,
timeout,
..
} = request;
let provider_request =
prepare_audio_transcription_provider_call(CoreAudioTranscriptionRequest {
model: &model,
audio,
api_key: api_key.as_deref(),
api_base: api_base.as_deref(),
custom_llm_provider: Some(&custom_llm_provider),
extra_headers,
optional_params,
timeout,
})?;
self.run_during_call_guardrails(provider_request).await
}
async fn run_during_call_guardrails(
&self,
request: ProviderAudioTranscriptionRequest,
) -> CoreResult<ProviderAudioTranscriptionRequest> {
) -> Result<ProviderAudioTranscriptionRequest, Error> {
if self.guardrail_runner.is_empty() {
return Ok(request);
}
@ -144,23 +126,23 @@ impl AudioTranscriptionLifecycleHooks {
.run_during_call(
&guardrail_context(&self.request_metadata),
GuardrailRequest::new(json!({
"model": request.model,
"custom_llm_provider": "bedrock",
"url": request.url,
"body": request.body,
"model": request.model(),
"custom_llm_provider": request.custom_llm_provider(),
"url": request.url(),
"body": request.body(),
})),
)
.await
.map_err(guardrail_error_to_core_error)?;
let Value::Object(mut data) = guardrail_request.data else {
return Err(CoreError::InvalidRequest(
return Err(Error::InvalidRequest(
"audio transcription during_call guardrail must return an object".to_string(),
));
};
let body = data.remove("body").ok_or_else(|| {
CoreError::InvalidRequest("audio transcription guardrail removed body".to_string())
Error::InvalidRequest("audio transcription guardrail removed body".to_string())
})?;
Ok(ProviderAudioTranscriptionRequest { body, ..request })
Ok(request.with_body(body))
}
fn logging_payload(
@ -241,7 +223,7 @@ impl CallLifecycleHooks<PreparedAudioTranscriptionRequest, ProviderAudioTranscri
fn async_log_failure_event<'a>(
&'a self,
context: &'a CallLifecycleContext,
error: &'a CoreError,
error: &'a Error,
timing: &'a CallLifecycleTiming,
) -> Self::FailureFuture<'a> {
Box::pin(async move {
@ -281,22 +263,22 @@ fn guardrail_context(metadata: &RequestMetadata) -> GuardrailContext {
}
}
fn guardrail_error_to_core_error(error: GuardrailError) -> CoreError {
CoreError::InvalidRequest(format!("{}: {}", error.kind, error.message))
fn guardrail_error_to_core_error(error: GuardrailError) -> Error {
Error::InvalidRequest(format!("{}: {}", error.kind, error.message))
}
fn core_error_kind(error: &CoreError) -> &'static str {
fn core_error_kind(error: &Error) -> &'static str {
match error {
CoreError::Auth(_) => "AuthError",
CoreError::InvalidProvider(_) => "InvalidProvider",
CoreError::InvalidRequest(_) => "InvalidRequest",
CoreError::InvalidType { .. } => "InvalidType",
CoreError::MissingField(_) => "MissingField",
CoreError::Http { .. } => "HttpError",
CoreError::InvalidResponse(_) => "InvalidResponse",
CoreError::Network(_) => "NetworkError",
CoreError::Connect(_) => "ConnectError",
CoreError::Routing(_) => "RoutingError",
CoreError::Unsupported(_) => "UnsupportedRequest",
Error::Auth(_) => "AuthError",
Error::InvalidProvider(_) => "InvalidProvider",
Error::InvalidRequest(_) => "InvalidRequest",
Error::InvalidType { .. } => "InvalidType",
Error::MissingField(_) => "MissingField",
Error::Http { .. } => "HttpError",
Error::InvalidResponse(_) => "InvalidResponse",
Error::Network(_) => "NetworkError",
Error::Connect(_) => "ConnectError",
Error::Routing(_) => "RoutingError",
Error::Unsupported(_) => "UnsupportedRequest",
}
}

8
litellm-rust/crates/ai-gateway/src/audio_transcription/mod.rs
View file

@ -1,19 +1,17 @@
use litellm_core::CoreResult;
use litellm_core::Error;
use litellm_core::audio_transcription::execute_audio_transcription_provider_call;
use litellm_core::call_lifecycle::CallLifecycle;
use serde_json::Value;
mod common_utils;
mod handler;
mod hooks;
mod prepare;
mod types;
pub use types::AudioTranscriptionRequest;
use handler::execute_audio_transcription_provider_call;
use prepare::{PreparedAudioTranscriptionCall, prepare_audio_transcription_call};
pub async fn audio_transcription(request: AudioTranscriptionRequest<'_>) -> CoreResult<Value> {
pub async fn audio_transcription(request: AudioTranscriptionRequest<'_>) -> Result<Value, Error> {
let PreparedAudioTranscriptionCall { request, hooks } =
prepare_audio_transcription_call(request);
CallLifecycle::default()

11
litellm-rust/crates/ai-gateway/src/audio_transcription/types.rs
View file

@ -1,7 +1,6 @@
use std::sync::Arc;
use std::time::Duration;
use litellm_core::audio_transcription::transformation::AudioTranscriptionProviderConfig;
use litellm_core::call_lifecycle::{CallLifecycleContext, CallLifecycleRequest};
use serde_json::{Map, Value};
@ -46,13 +45,3 @@ impl CallLifecycleRequest for PreparedAudioTranscriptionRequest {
)
}
}
#[derive(Clone)]
pub(crate) struct ProviderAudioTranscriptionRequest {
pub(crate) model: String,
pub(crate) config: &'static dyn AudioTranscriptionProviderConfig,
pub(crate) url: String,
pub(crate) body: Value,
pub(crate) upstream_headers: Vec<(String, String)>,
pub(crate) timeout: Option<Duration>,
}

43
litellm-rust/crates/ai-gateway/src/io/realtime.rs
View file

@ -15,8 +15,7 @@ use std::time::Duration;
use futures_util::stream::{SplitSink, SplitStream};
use futures_util::{Sink, SinkExt, Stream, StreamExt};
use litellm_core::CoreResult;
use litellm_core::error::CoreError;
use litellm_core::error::Error;
use litellm_core::realtime::transformation::RealtimeProviderConfig;
use litellm_core::realtime::types::RealtimeEvent;
use tokio::net::TcpStream;
@ -48,7 +47,7 @@ pub(crate) type UpstreamRx = SplitStream<UpstreamWs>;
/// Resolve the OpenAI API key from the explicit param or the environment.
///
/// Blank/whitespace values are treated as absent (guard at resolution time).
pub(crate) fn resolve_api_key(api_key: Option<&str>) -> CoreResult<String> {
pub(crate) fn resolve_api_key(api_key: Option<&str>) -> Result<String, Error> {
api_key
.map(str::trim)
.filter(|key| !key.is_empty())
@ -58,7 +57,7 @@ pub(crate) fn resolve_api_key(api_key: Option<&str>) -> CoreResult<String> {
.ok()
.filter(|key| !key.trim().is_empty())
})
.ok_or_else(|| CoreError::Auth(MISSING_KEY_MESSAGE.to_string()))
.ok_or_else(|| Error::Auth(MISSING_KEY_MESSAGE.to_string()))
}
/// Open the upstream WebSocket to OpenAI for `(model, api_key, api_base)`.
@ -70,24 +69,24 @@ pub(crate) async fn dial_upstream(
model: &str,
api_key: &str,
api_base: Option<&str>,
) -> CoreResult<UpstreamWs> {
) -> Result<UpstreamWs, Error> {
let url = OPENAI_REALTIME_CONFIG.complete_url(api_base, model);
let mut request = url
.as_str()
.into_client_request()
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
// GA realtime: only Authorization. The legacy OpenAI-Beta header triggers
// beta_api_shape_disabled, so we do not send it.
request.headers_mut().insert(
AUTHORIZATION,
HeaderValue::from_str(&format!("Bearer {api_key}"))
.map_err(|err| CoreError::Auth(err.to_string()))?,
.map_err(|err| Error::Auth(err.to_string()))?,
);
let (upstream, _response) = connect_async(request)
.await
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
Ok(upstream)
}
@ -96,22 +95,22 @@ pub(crate) async fn dial_upstream(
/// Used by the pool to pre-read OpenAI's unprompted `session.created`. Returns an
/// error on a non-text frame, a closed socket, or undecodable JSON so the pool can
/// discard a misbehaving socket rather than warm it.
pub(crate) async fn read_event(upstream_rx: &mut UpstreamRx) -> CoreResult<RealtimeEvent> {
pub(crate) async fn read_event(upstream_rx: &mut UpstreamRx) -> Result<RealtimeEvent, Error> {
loop {
let message = upstream_rx
.next()
.await
.ok_or_else(|| CoreError::Network("upstream closed before first event".to_string()))?
.map_err(|err| CoreError::Network(err.to_string()))?;
.ok_or_else(|| Error::Network("upstream closed before first event".to_string()))?
.map_err(|err| Error::Network(err.to_string()))?;
match message {
Message::Text(text) => {
return serde_json::from_str(&text)
.map_err(|err| CoreError::InvalidResponse(err.to_string()));
.map_err(|err| Error::InvalidResponse(err.to_string()));
}
// Ignore protocol frames (ping/pong) while waiting for the first event.
Message::Ping(_) | Message::Pong(_) => continue,
Message::Close(_) => {
return Err(CoreError::Network(
return Err(Error::Network(
"upstream closed before first event".to_string(),
));
}
@ -139,7 +138,7 @@ pub(crate) async fn splice<In, Out>(
mut observe: impl FnMut(&RealtimeEvent) + Send,
mut client_in: In,
mut client_out: Out,
) -> CoreResult<()>
) -> Result<(), Error>
where
In: Stream<Item = RealtimeEvent> + Unpin + Send,
Out: Sink<RealtimeEvent> + Unpin + Send,
@ -154,7 +153,7 @@ where
client_out
.send(outbound)
.await
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
}
}
@ -175,26 +174,26 @@ where
// inflate its own spend log. Logging observes upstream events only.
for outbound in config.transform_realtime_request(&event, model)?.events {
let payload = serde_json::to_string(&outbound)
.map_err(|err| CoreError::InvalidResponse(err.to_string()))?;
.map_err(|err| Error::InvalidResponse(err.to_string()))?;
upstream_tx
.send(Message::Text(payload))
.await
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
}
}
// upstream -> client
upstream_message = upstream_rx.next() => {
let Some(message) = upstream_message else { break }; // upstream closed
match message.map_err(|err| CoreError::Network(err.to_string()))? {
match message.map_err(|err| Error::Network(err.to_string()))? {
Message::Text(text) => {
let event: RealtimeEvent = serde_json::from_str(&text)
.map_err(|err| CoreError::InvalidResponse(err.to_string()))?;
.map_err(|err| Error::InvalidResponse(err.to_string()))?;
observe(&event);
for outbound in config.transform_realtime_response(&event, model)?.events {
client_out
.send(outbound)
.await
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
}
}
Message::Close(_) => break,
@ -225,7 +224,7 @@ pub async fn realtime<In, Out>(
observe: impl FnMut(&RealtimeEvent) + Send,
client_in: In,
client_out: Out,
) -> CoreResult<()>
) -> Result<(), Error>
where
In: Stream<Item = RealtimeEvent> + Unpin + Send,
Out: Sink<RealtimeEvent> + Unpin + Send,
@ -258,7 +257,7 @@ pub async fn realtime_warm<In, Out>(
observe: impl FnMut(&RealtimeEvent) + Send,
client_in: In,
client_out: Out,
) -> CoreResult<()>
) -> Result<(), Error>
where
In: Stream<Item = RealtimeEvent> + Unpin + Send,
Out: Sink<RealtimeEvent> + Unpin + Send,

4
litellm-rust/crates/ai-gateway/src/io/realtime_pool.rs
View file

@ -28,7 +28,7 @@ use std::sync::{Arc, Mutex};
use std::time::{Duration, Instant};
use futures_util::StreamExt;
use litellm_core::CoreResult;
use litellm_core::Error;
use litellm_core::realtime::types::RealtimeEvent;
use crate::io::realtime::{
@ -438,7 +438,7 @@ impl RealtimePool {
///
/// `key.api_key` is already resolved (non-blank). The first frame OpenAI sends
/// unprompted is `session.created`; we buffer exactly that and read nothing more.
async fn warm_one(key: &UpstreamKey) -> CoreResult<WarmConnection> {
async fn warm_one(key: &UpstreamKey) -> Result<WarmConnection, Error> {
let upstream: UpstreamWs =
dial_upstream(&key.model, &key.api_key, key.api_base.as_deref()).await?;
let (tx, mut rx) = upstream.split();

76
litellm-rust/crates/ai-gateway/src/io/responses_ws.rs
View file

@ -4,10 +4,10 @@ use std::time::Duration;
use futures_util::stream::{SplitSink, SplitStream};
use futures_util::{Sink, SinkExt, Stream, StreamExt};
use litellm_core::Error;
use litellm_core::providers::openai::responses::transformation::OPENAI_RESPONSES_WS_CONFIG;
use litellm_core::responses::types::ResponsesWsEvent;
use litellm_core::responses::websocket::ResponsesWebSocketProviderConfig;
use litellm_core::{CoreError, CoreResult};
use tokio::net::TcpStream;
use tokio::sync::Mutex;
use tokio_tungstenite::tungstenite::Message;
@ -37,51 +37,49 @@ impl ResponsesWebSocketConnection {
url: &str,
headers: &HashMap<String, String>,
timeout: Option<Duration>,
) -> CoreResult<Self> {
) -> Result<Self, Error> {
let mut request = url
.into_client_request()
.map_err(|error| CoreError::Network(error.to_string()))?;
.map_err(|error| Error::Network(error.to_string()))?;
for (name, value) in headers {
let header_name = name
.parse::<HeaderName>()
.map_err(|error| CoreError::InvalidRequest(error.to_string()))?;
.map_err(|error| Error::InvalidRequest(error.to_string()))?;
let header_value = HeaderValue::from_str(value)
.map_err(|error| CoreError::InvalidRequest(error.to_string()))?;
.map_err(|error| Error::InvalidRequest(error.to_string()))?;
request.headers_mut().insert(header_name, header_value);
}
let connect = connect_async(request);
let result = match timeout {
Some(timeout) => tokio::time::timeout(timeout, connect).await.map_err(|_| {
CoreError::Network("Responses WebSocket connection timed out".to_string())
Error::Network("Responses WebSocket connection timed out".to_string())
})?,
None => connect.await,
};
let (socket, _) = result.map_err(|error| match error {
tokio_tungstenite::tungstenite::Error::Http(response) => CoreError::Http {
tokio_tungstenite::tungstenite::Error::Http(response) => Error::Http {
status: response.status().as_u16(),
body: String::new(),
},
other => CoreError::Network(other.to_string()),
other => Error::Network(other.to_string()),
})?;
Ok(Self {
socket: Arc::new(Mutex::new(Some(socket))),
})
}
pub async fn send_text(&self, text: String) -> CoreResult<()> {
pub async fn send_text(&self, text: String) -> Result<(), Error> {
let mut socket = self.socket.lock().await;
let Some(socket) = socket.as_mut() else {
return Err(CoreError::Network(
"Responses WebSocket is closed".to_string(),
));
return Err(Error::Network("Responses WebSocket is closed".to_string()));
};
socket
.send(Message::Text(text))
.await
.map_err(|error| CoreError::Network(error.to_string()))
.map_err(|error| Error::Network(error.to_string()))
}
pub async fn recv_text(&self) -> CoreResult<Option<String>> {
pub async fn recv_text(&self) -> Result<Option<String>, Error> {
let mut socket_guard = self.socket.lock().await;
let Some(socket) = socket_guard.as_mut() else {
return Ok(None);
@ -90,27 +88,27 @@ impl ResponsesWebSocketConnection {
Some(Ok(Message::Text(text))) => Ok(Some(text)),
Some(Ok(Message::Binary(bytes))) => String::from_utf8(bytes.to_vec())
.map(Some)
.map_err(|error| CoreError::InvalidResponse(error.to_string())),
.map_err(|error| Error::InvalidResponse(error.to_string())),
Some(Ok(Message::Close(_))) | None => Ok(None),
Some(Ok(_)) => Ok(None),
Some(Err(error)) => Err(CoreError::Network(error.to_string())),
Some(Err(error)) => Err(Error::Network(error.to_string())),
}
}
pub async fn close(&self) -> CoreResult<()> {
pub async fn close(&self) -> Result<(), Error> {
let mut socket = self.socket.lock().await;
if let Some(socket) = socket.as_mut() {
socket
.close(None)
.await
.map_err(|error| CoreError::Network(error.to_string()))?;
.map_err(|error| Error::Network(error.to_string()))?;
}
*socket = None;
Ok(())
}
}
pub(crate) fn resolve_api_key(api_key: Option<&str>) -> CoreResult<String> {
pub(crate) fn resolve_api_key(api_key: Option<&str>) -> Result<String, Error> {
api_key
.map(str::trim)
.filter(|value| !value.is_empty())
@ -120,38 +118,38 @@ pub(crate) fn resolve_api_key(api_key: Option<&str>) -> CoreResult<String> {
.ok()
.filter(|value| !value.trim().is_empty())
})
.ok_or_else(|| CoreError::Auth(MISSING_KEY_MESSAGE.to_string()))
.ok_or_else(|| Error::Auth(MISSING_KEY_MESSAGE.to_string()))
}
async fn dial_upstream(
model: &str,
api_key: &str,
api_base: Option<&str>,
) -> CoreResult<ResponsesUpstreamWs> {
) -> Result<ResponsesUpstreamWs, Error> {
let url = OPENAI_RESPONSES_WS_CONFIG.complete_websocket_url(api_base, model);
let mut request = url
.as_str()
.into_client_request()
.map_err(|error| CoreError::Network(error.to_string()))?;
.map_err(|error| Error::Network(error.to_string()))?;
request.headers_mut().insert(
AUTHORIZATION,
HeaderValue::from_str(&format!("Bearer {api_key}"))
.map_err(|error| CoreError::Auth(error.to_string()))?,
.map_err(|error| Error::Auth(error.to_string()))?,
);
let result = tokio::time::timeout(
Duration::from_secs(DEFAULT_RESPONSES_WS_CONNECT_TIMEOUT_SECS),
connect_async(request),
)
.await
.map_err(|_| CoreError::Network("Responses WebSocket connection timed out".to_string()))?;
.map_err(|_| Error::Network("Responses WebSocket connection timed out".to_string()))?;
result
.map(|(socket, _)| socket)
.map_err(|error| match error {
tokio_tungstenite::tungstenite::Error::Http(response) => CoreError::Http {
tokio_tungstenite::tungstenite::Error::Http(response) => Error::Http {
status: response.status().as_u16(),
body: String::new(),
},
other => CoreError::Network(other.to_string()),
other => Error::Network(other.to_string()),
})
}
@ -166,7 +164,7 @@ impl ResponsesWebSocketStreaming {
observe: impl FnMut(&ResponsesWsEvent) + Send,
client_in: In,
client_out: Out,
) -> CoreResult<()>
) -> Result<(), Error>
where
In: Stream<Item = ResponsesWsEvent> + Unpin + Send,
Out: Sink<ResponsesWsEvent> + Unpin + Send,
@ -193,7 +191,7 @@ pub(crate) async fn splice<In, Out>(
mut observe: impl FnMut(&ResponsesWsEvent) + Send,
mut client_in: In,
mut client_out: Out,
) -> CoreResult<()>
) -> Result<(), Error>
where
In: Stream<Item = ResponsesWsEvent> + Unpin + Send,
Out: Sink<ResponsesWsEvent> + Unpin + Send,
@ -210,18 +208,18 @@ where
.events
{
let payload = serde_json::to_string(&outbound)
.map_err(|error| CoreError::InvalidResponse(error.to_string()))?;
.map_err(|error| Error::InvalidResponse(error.to_string()))?;
upstream_tx.send(Message::Text(payload))
.await
.map_err(|error| CoreError::Network(error.to_string()))?;
.map_err(|error| Error::Network(error.to_string()))?;
}
}
message = upstream_rx.next() => {
let Some(message) = message else { break };
match message.map_err(|error| CoreError::Network(error.to_string()))? {
match message.map_err(|error| Error::Network(error.to_string()))? {
Message::Text(text) => {
let event = serde_json::from_str::<ResponsesWsEvent>(&text)
.map_err(|error| CoreError::InvalidResponse(error.to_string()))?;
.map_err(|error| Error::InvalidResponse(error.to_string()))?;
observe(&event);
for outbound in OPENAI_RESPONSES_WS_CONFIG
.transform_ws_response(&event, model)?
@ -229,7 +227,7 @@ where
{
client_out.send(outbound)
.await
.map_err(|error| CoreError::Network(error.to_string()))?;
.map_err(|error| Error::Network(error.to_string()))?;
}
}
Message::Close(_) => break,
@ -252,7 +250,7 @@ pub async fn async_responses_websocket<In, Out>(
mut observe: impl FnMut(&ResponsesWsEvent) + Send,
client_in: In,
client_out: Out,
) -> CoreResult<()>
) -> Result<(), Error>
where
In: Stream<Item = ResponsesWsEvent> + Unpin + Send,
Out: Sink<ResponsesWsEvent> + Unpin + Send,
@ -267,11 +265,11 @@ where
.events
{
let payload = serde_json::to_string(&outbound)
.map_err(|error| CoreError::InvalidResponse(error.to_string()))?;
.map_err(|error| Error::InvalidResponse(error.to_string()))?;
upstream_tx
.send(Message::Text(payload))
.await
.map_err(|error| CoreError::Network(error.to_string()))?;
.map_err(|error| Error::Network(error.to_string()))?;
}
}
ResponsesWebSocketStreaming::bidirectional_forward(
@ -296,7 +294,7 @@ pub async fn responses_ws<In, Out>(
observe: impl FnMut(&ResponsesWsEvent) + Send,
client_in: In,
client_out: Out,
) -> CoreResult<()>
) -> Result<(), Error>
where
In: Stream<Item = ResponsesWsEvent> + Unpin + Send,
Out: Sink<ResponsesWsEvent> + Unpin + Send,
@ -514,7 +512,7 @@ mod tests {
)
.await
.expect_err("status error");
assert!(matches!(error, CoreError::Http { status: 401, .. }));
assert!(matches!(error, Error::Http { status: 401, .. }));
server.await.expect("server task");
}
@ -543,7 +541,7 @@ mod tests {
)
.await
.expect_err("status error");
assert!(matches!(error, CoreError::Http { status: 500, .. }));
assert!(matches!(error, Error::Http { status: 500, .. }));
server.await.expect("server task");
}
}

80
litellm-rust/crates/ai-gateway/src/ocr/common_utils.rs
View file

@ -3,8 +3,7 @@ use std::time::{Duration, Instant};
use base64::Engine;
use base64::engine::general_purpose::STANDARD as BASE64_STANDARD;
use litellm_core::CoreResult;
use litellm_core::error::CoreError;
use litellm_core::error::Error;
use litellm_core::ocr::transformation::OcrProviderConfig;
use reqwest::Url;
use serde_json::{Map, Value};
@ -33,6 +32,7 @@ pub(super) fn truncate_error_body(body: &str) -> String {
format!("{truncated}... (truncated)")
}
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
pub(super) fn ocr_provider_config(
provider: &str,
model: &str,
@ -56,7 +56,7 @@ fn is_azure_document_intelligence_model(model: &str) -> bool {
pub(super) fn string_headers(
extra_headers: Option<Map<String, Value>>,
) -> CoreResult<Vec<(String, String)>> {
) -> Result<Vec<(String, String)>, Error> {
extra_headers
.unwrap_or_default()
.into_iter()
@ -65,7 +65,7 @@ pub(super) fn string_headers(
.as_str()
.map(|value| (key.clone(), value.to_string()))
.ok_or_else(|| {
CoreError::InvalidRequest(format!(
Error::InvalidRequest(format!(
"OCR extra_headers.{key} must be a string, got {}",
litellm_core::error::json_type_name(&value)
))
@ -74,13 +74,7 @@ pub(super) fn string_headers(
.collect()
}
pub(super) fn has_header(headers: &[(String, String)], name: &str) -> bool {
headers
.iter()
.any(|(key, _)| key.eq_ignore_ascii_case(name))
}
fn document_url_field(document: &Value) -> CoreResult<Option<(&str, &str)>> {
fn document_url_field(document: &Value) -> Result<Option<(&str, &str)>, Error> {
let Some(object) = document.as_object() else {
return Ok(None);
};
@ -138,13 +132,13 @@ fn is_blocked_ip(ip: IpAddr) -> bool {
}
}
fn blocked_url_error(url: &Url) -> CoreError {
CoreError::InvalidRequest(format!(
fn blocked_url_error(url: &Url) -> Error {
Error::InvalidRequest(format!(
"OCR document URL rejected by SSRF protection: {url}"
))
}
async fn validate_safe_fetch_url(url: &Url) -> CoreResult<()> {
async fn validate_safe_fetch_url(url: &Url) -> Result<(), Error> {
if !matches!(url.scheme(), "http" | "https") {
return Err(blocked_url_error(url));
}
@ -162,7 +156,7 @@ async fn validate_safe_fetch_url(url: &Url) -> CoreResult<()> {
.ok_or_else(|| blocked_url_error(url))?;
let addresses = tokio::net::lookup_host((host, port))
.await
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
let mut saw_address = false;
for address in addresses {
saw_address = true;
@ -176,25 +170,25 @@ async fn validate_safe_fetch_url(url: &Url) -> CoreResult<()> {
Ok(())
}
fn redirect_location(response: &reqwest::Response, url: &Url) -> CoreResult<Url> {
fn redirect_location(response: &reqwest::Response, url: &Url) -> Result<Url, Error> {
let location = response
.headers()
.get(reqwest::header::LOCATION)
.and_then(|value| value.to_str().ok())
.ok_or_else(|| {
CoreError::InvalidResponse("OCR document redirect missing Location header".to_string())
Error::InvalidResponse("OCR document redirect missing Location header".to_string())
})?;
url.join(location)
.map_err(|err| CoreError::InvalidResponse(format!("invalid OCR document redirect: {err}")))
.map_err(|err| Error::InvalidResponse(format!("invalid OCR document redirect: {err}")))
}
async fn safe_get_document_url(url: &str) -> CoreResult<(Url, reqwest::Response)> {
async fn safe_get_document_url(url: &str) -> Result<(Url, reqwest::Response), Error> {
let client = reqwest::Client::builder()
.redirect(reqwest::redirect::Policy::none())
.build()
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
let mut current_url = Url::parse(url)
.map_err(|err| CoreError::InvalidRequest(format!("invalid OCR document URL: {err}")))?;
.map_err(|err| Error::InvalidRequest(format!("invalid OCR document URL: {err}")))?;
for _ in 0..MAX_SAFE_FETCH_REDIRECTS {
validate_safe_fetch_url(&current_url).await?;
@ -202,28 +196,28 @@ async fn safe_get_document_url(url: &str) -> CoreResult<(Url, reqwest::Response)
.get(current_url.clone())
.send()
.await
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
if !response.status().is_redirection() {
return Ok((current_url, response));
}
current_url = redirect_location(&response, &current_url)?;
}
Err(CoreError::InvalidRequest(
Err(Error::InvalidRequest(
"Too many redirects while fetching OCR document URL".to_string(),
))
}
fn enforce_download_size(content_length: u64, max_bytes: u64, url: &Url) -> CoreResult<()> {
fn enforce_download_size(content_length: u64, max_bytes: u64, url: &Url) -> Result<(), Error> {
if max_bytes == 0 {
return Err(CoreError::InvalidRequest(format!(
return Err(Error::InvalidRequest(format!(
"OCR document URL download is disabled (MAX_IMAGE_URL_DOWNLOAD_SIZE_MB=0). url={url}"
)));
}
if content_length > max_bytes {
let size_mb = content_length as f64 / (1024.0 * 1024.0);
let max_size_mb = max_bytes as f64 / (1024.0 * 1024.0);
return Err(CoreError::InvalidRequest(format!(
return Err(Error::InvalidRequest(format!(
"OCR document size ({size_mb:.2}MB) exceeds maximum allowed size ({max_size_mb:.2}MB). url={url}"
)));
}
@ -233,7 +227,7 @@ fn enforce_download_size(content_length: u64, max_bytes: u64, url: &Url) -> Core
async fn read_response_with_limit(
mut response: reqwest::Response,
url: &Url,
) -> CoreResult<Vec<u8>> {
) -> Result<Vec<u8>, Error> {
let max_bytes = max_document_download_bytes();
if let Some(content_length) = response.content_length() {
enforce_download_size(content_length, max_bytes, url)?;
@ -246,7 +240,7 @@ async fn read_response_with_limit(
while let Some(chunk) = response
.chunk()
.await
.map_err(|err| CoreError::Network(err.to_string()))?
.map_err(|err| Error::Network(err.to_string()))?
{
bytes_downloaded += chunk.len() as u64;
enforce_download_size(bytes_downloaded, max_bytes, url)?;
@ -255,7 +249,7 @@ async fn read_response_with_limit(
Ok(bytes)
}
pub(super) async fn convert_document_url_to_data_uri(document: Value) -> CoreResult<Value> {
pub(super) async fn convert_document_url_to_data_uri(document: Value) -> Result<Value, Error> {
let Some((field, url)) = document_url_field(&document)? else {
return Ok(document);
};
@ -267,7 +261,7 @@ pub(super) async fn convert_document_url_to_data_uri(document: Value) -> CoreRes
let status = response.status();
if !status.is_success() {
let body = response.text().await.unwrap_or_default();
return Err(CoreError::Http {
return Err(Error::Http {
status: status.as_u16(),
body: truncate_error_body(&body),
});
@ -290,7 +284,7 @@ pub(super) async fn convert_document_url_to_data_uri(document: Value) -> CoreRes
let mut transformed = document
.as_object()
.cloned()
.ok_or_else(|| CoreError::InvalidRequest("OCR document must be an object".to_string()))?;
.ok_or_else(|| Error::InvalidRequest("OCR document must be an object".to_string()))?;
transformed.insert(field.to_string(), Value::String(data_uri));
Ok(Value::Object(transformed))
}
@ -316,11 +310,11 @@ fn retry_after_secs(response: &reqwest::Response) -> u64 {
.unwrap_or(2)
}
fn operation_status(response_json: &Value) -> CoreResult<&str> {
fn operation_status(response_json: &Value) -> Result<&str, Error> {
let status = response_json
.get("status")
.and_then(Value::as_str)
.ok_or(CoreError::MissingField("status"))?;
.ok_or(Error::MissingField("status"))?;
match status {
"succeeded" => Ok("succeeded"),
"running" | "notStarted" => Ok("running"),
@ -330,11 +324,11 @@ fn operation_status(response_json: &Value) -> CoreResult<&str> {
.and_then(|error| error.get("message"))
.and_then(Value::as_str)
.unwrap_or("Unknown error");
Err(CoreError::InvalidResponse(format!(
Err(Error::InvalidResponse(format!(
"Azure Document Intelligence analysis failed: {message}"
)))
}
other => Err(CoreError::InvalidResponse(format!(
other => Err(Error::InvalidResponse(format!(
"Unknown operation status: {other}"
))),
}
@ -345,9 +339,9 @@ pub(super) async fn poll_document_intelligence(
original_url: &str,
headers: &[(String, String)],
timeout: Option<Duration>,
) -> CoreResult<Value> {
) -> Result<Value, Error> {
if !same_origin(operation_url, original_url) {
return Err(CoreError::InvalidResponse(
return Err(Error::InvalidResponse(
"Azure Document Intelligence: rejected cross-origin polling URL".to_string(),
));
}
@ -358,7 +352,7 @@ pub(super) async fn poll_document_intelligence(
));
loop {
if start.elapsed() > timeout {
return Err(CoreError::Network(format!(
return Err(Error::Network(format!(
"Azure Document Intelligence operation polling timed out after {} seconds",
timeout.as_secs()
)));
@ -373,21 +367,21 @@ pub(super) async fn poll_document_intelligence(
let response = request_builder
.send()
.await
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
let retry_after = retry_after_secs(&response);
let status = response.status();
let text = response
.text()
.await
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
if !status.is_success() {
return Err(CoreError::Http {
return Err(Error::Http {
status: status.as_u16(),
body: truncate_error_body(&text),
});
}
let response_json: Value = serde_json::from_str(&text).map_err(|err| {
CoreError::InvalidResponse(format!("invalid Azure DI poll response JSON: {err}"))
Error::InvalidResponse(format!("invalid Azure DI poll response JSON: {err}"))
})?;
if operation_status(&response_json)? == "succeeded" {
return Ok(response_json);
@ -426,7 +420,7 @@ mod tests {
assert!(matches!(
error,
CoreError::InvalidRequest(message)
Error::InvalidRequest(message)
if message.contains("SSRF protection")
));
}

27
litellm-rust/crates/ai-gateway/src/ocr/handler.rs
View file

@ -1,13 +1,19 @@
use litellm_core::CoreResult;
use litellm_core::error::CoreError;
use litellm_core::error::Error;
use litellm_core::http_utils::http_request;
use litellm_core::ocr::transformation::OcrResponseHandling;
use serde_json::Value;
use super::common_utils::{poll_document_intelligence, truncate_error_body};
use super::types::ProviderOcrRequest;
use super::hooks::OcrLifecycleHooks;
use super::types::PreparedOcrRequest;
use crate::client::http_client;
pub(crate) async fn execute_ocr_provider_call(request: ProviderOcrRequest) -> CoreResult<Value> {
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
pub(crate) async fn execute_ocr_provider_call(
request: PreparedOcrRequest,
hooks: &OcrLifecycleHooks,
) -> Result<Value, Error> {
let request = hooks.prepare_provider_request(request).await?;
let mut request_builder = http_client().post(&request.url).json(&request.body);
for (key, value) in &request.upstream_headers {
request_builder = request_builder.header(key, value);
@ -16,10 +22,9 @@ pub(crate) async fn execute_ocr_provider_call(request: ProviderOcrRequest) -> Co
request_builder = request_builder.timeout(duration);
}
let response = request_builder
.send()
let response = http_request(request_builder)
.await
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
let status = response.status();
if request.config.response_handling() == OcrResponseHandling::AzureDocumentIntelligencePoll
@ -31,7 +36,7 @@ pub(crate) async fn execute_ocr_provider_call(request: ProviderOcrRequest) -> Co
.and_then(|value| value.to_str().ok())
.map(str::to_string)
.ok_or_else(|| {
CoreError::InvalidResponse(
Error::InvalidResponse(
"Azure Document Intelligence returned 202 but no Operation-Location header found"
.to_string(),
)
@ -52,17 +57,17 @@ pub(crate) async fn execute_ocr_provider_call(request: ProviderOcrRequest) -> Co
let text = response
.text()
.await
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
if !status.is_success() {
return Err(CoreError::Http {
return Err(Error::Http {
status: status.as_u16(),
body: truncate_error_body(&text),
});
}
let response_json: Value = serde_json::from_str(&text)
.map_err(|err| CoreError::InvalidResponse(format!("invalid OCR response JSON: {err}")))?;
.map_err(|err| Error::InvalidResponse(format!("invalid OCR response JSON: {err}")))?;
Ok(request
.config

112
litellm-rust/crates/ai-gateway/src/ocr/hooks.rs
View file

@ -1,15 +1,10 @@
use litellm_core::call_lifecycle::{CallLifecycleContext, CallLifecycleHooks, CallLifecycleTiming};
use litellm_core::error::Error;
use serde_json::{Map, Value, json};
use std::future::Future;
use std::pin::Pin;
use litellm_core::CoreResult;
use litellm_core::call_lifecycle::{CallLifecycleContext, CallLifecycleHooks, CallLifecycleTiming};
use litellm_core::error::CoreError;
use litellm_core::ocr::transformation::OcrAuthStrategy;
use serde_json::{Map, Value, json};
use super::common_utils::{
convert_document_url_to_data_uri, has_header, ocr_provider_config, string_headers,
};
use super::common_utils::{convert_document_url_to_data_uri, string_headers};
use super::types::{PreparedOcrRequest, ProviderOcrRequest};
use crate::integrations::custom_guardrail::{
CustomGuardrailRunner, GuardrailContext, GuardrailError, GuardrailRequest,
@ -27,7 +22,7 @@ pub(crate) struct OcrLifecycleHooks {
request_metadata: RequestMetadata,
}
type OcrFuture<'a, T> = Pin<Box<dyn Future<Output = CoreResult<T>> + Send + 'a>>;
type OcrFuture<'a, T> = Pin<Box<dyn Future<Output = Result<T, Error>> + Send + 'a>>;
type OcrLogFuture<'a> = Pin<Box<dyn Future<Output = ()> + Send + 'a>>;
impl OcrLifecycleHooks {
@ -46,7 +41,7 @@ impl OcrLifecycleHooks {
async fn run_pre_call_guardrails(
&self,
request: PreparedOcrRequest,
) -> CoreResult<PreparedOcrRequest> {
) -> Result<PreparedOcrRequest, Error> {
if self.guardrail_runner.is_empty() {
return Ok(request);
}
@ -64,6 +59,10 @@ impl OcrLifecycleHooks {
.await
.map_err(guardrail_error_to_core_error)?;
let (document, optional_params) = parse_ocr_pre_call_guardrail_request(guardrail_request)?;
let optional_params = match &request.config {
Ok(config) => config.map_ocr_params(&optional_params),
Err(_) => optional_params,
};
Ok(PreparedOcrRequest {
document,
optional_params,
@ -71,25 +70,23 @@ impl OcrLifecycleHooks {
})
}
async fn prepare_provider_request(
pub(crate) async fn prepare_provider_request(
&self,
request: PreparedOcrRequest,
) -> CoreResult<ProviderOcrRequest> {
let config = ocr_provider_config(&request.custom_llm_provider, &request.model)
.ok_or_else(|| CoreError::InvalidProvider(request.custom_llm_provider.clone()))?;
) -> Result<ProviderOcrRequest, Error> {
let config = request.config?;
let env_lookup = |key: &str| std::env::var(key).ok();
let headers = string_headers(request.extra_headers)?;
let auth_strategy = config.auth_strategy();
let api_key = (!has_header(&headers, auth_strategy.header_name()))
.then(|| config.resolve_api_key(request.api_key.as_deref(), &env_lookup))
.transpose()?;
let upstream_headers = config.validate_environment(
string_headers(request.extra_headers)?,
request.api_key.as_deref(),
&env_lookup,
)?;
let url = config.complete_url(
request.api_base.as_deref(),
&request.model,
&request.optional_params,
&env_lookup,
)?;
let filtered_params = config.map_ocr_params(&request.optional_params);
let model = request.model.clone();
let custom_llm_provider = request.custom_llm_provider.clone();
let document = if config.requires_data_uri_document() {
@ -98,9 +95,8 @@ impl OcrLifecycleHooks {
request.document
};
let body = config
.transform_ocr_request(&request.model, document, filtered_params)?
.transform_ocr_request(&request.model, document, request.optional_params)?
.data;
let upstream_headers = upstream_headers(&headers, auth_strategy, api_key.as_deref());
let body = self
.run_during_call_guardrails(&model, &custom_llm_provider, &url, body)
.await?;
@ -120,7 +116,7 @@ impl OcrLifecycleHooks {
custom_llm_provider: &str,
url: &str,
body: Value,
) -> CoreResult<Value> {
) -> Result<Value, Error> {
if self.guardrail_runner.is_empty() {
return Ok(body);
}
@ -169,9 +165,9 @@ impl OcrLifecycleHooks {
}
}
impl CallLifecycleHooks<PreparedOcrRequest, ProviderOcrRequest, Value> for OcrLifecycleHooks {
impl CallLifecycleHooks<PreparedOcrRequest, PreparedOcrRequest, Value> for OcrLifecycleHooks {
type PreCallFuture<'a> = OcrFuture<'a, PreparedOcrRequest>;
type DuringCallFuture<'a> = OcrFuture<'a, ProviderOcrRequest>;
type DuringCallFuture<'a> = OcrFuture<'a, PreparedOcrRequest>;
type SuccessFuture<'a> = OcrLogFuture<'a>;
type FailureFuture<'a> = OcrLogFuture<'a>;
@ -188,7 +184,7 @@ impl CallLifecycleHooks<PreparedOcrRequest, ProviderOcrRequest, Value> for OcrLi
_context: &'a CallLifecycleContext,
request: PreparedOcrRequest,
) -> Self::DuringCallFuture<'a> {
Box::pin(async move { self.prepare_provider_request(request).await })
Box::pin(async move { Ok(request) })
}
fn async_log_success_event<'a>(
@ -217,7 +213,7 @@ impl CallLifecycleHooks<PreparedOcrRequest, ProviderOcrRequest, Value> for OcrLi
fn async_log_failure_event<'a>(
&'a self,
context: &'a CallLifecycleContext,
error: &'a CoreError,
error: &'a Error,
timing: &'a CallLifecycleTiming,
) -> Self::FailureFuture<'a> {
Box::pin(async move {
@ -249,21 +245,6 @@ impl CallLifecycleHooks<PreparedOcrRequest, ProviderOcrRequest, Value> for OcrLi
}
}
fn upstream_headers(
headers: &[(String, String)],
auth_strategy: OcrAuthStrategy,
api_key: Option<&str>,
) -> Vec<(String, String)> {
api_key
.map(|api_key| match auth_strategy {
OcrAuthStrategy::Bearer => ("Authorization".to_string(), format!("Bearer {api_key}")),
OcrAuthStrategy::Header(header_name) => (header_name.to_string(), api_key.to_string()),
})
.into_iter()
.chain(headers.iter().cloned())
.collect()
}
fn guardrail_context(metadata: &RequestMetadata) -> GuardrailContext {
GuardrailContext {
call_type: CallType::Ocr,
@ -278,19 +259,19 @@ fn guardrail_context(metadata: &RequestMetadata) -> GuardrailContext {
fn parse_ocr_pre_call_guardrail_request(
request: GuardrailRequest,
) -> CoreResult<(Value, Map<String, Value>)> {
) -> Result<(Value, Map<String, Value>), Error> {
let Value::Object(mut data) = request.data else {
return Err(CoreError::InvalidRequest(
return Err(Error::InvalidRequest(
"OCR pre_call guardrail must return an object".to_string(),
));
};
let document = data.remove("document").ok_or_else(|| {
CoreError::InvalidRequest("OCR pre_call guardrail removed document".to_string())
Error::InvalidRequest("OCR pre_call guardrail removed document".to_string())
})?;
let optional_params = match data.remove("optional_params") {
Some(Value::Object(params)) => params,
Some(_) => {
return Err(CoreError::InvalidRequest(
return Err(Error::InvalidRequest(
"OCR pre_call guardrail optional_params must be an object".to_string(),
));
}
@ -299,33 +280,32 @@ fn parse_ocr_pre_call_guardrail_request(
Ok((document, optional_params))
}
fn parse_ocr_during_call_guardrail_request(request: GuardrailRequest) -> CoreResult<Value> {
fn parse_ocr_during_call_guardrail_request(request: GuardrailRequest) -> Result<Value, Error> {
let Value::Object(mut data) = request.data else {
return Err(CoreError::InvalidRequest(
return Err(Error::InvalidRequest(
"OCR during_call guardrail must return an object".to_string(),
));
};
data.remove("body").ok_or_else(|| {
CoreError::InvalidRequest("OCR during_call guardrail removed body".to_string())
})
data.remove("body")
.ok_or_else(|| Error::InvalidRequest("OCR during_call guardrail removed body".to_string()))
}
fn guardrail_error_to_core_error(error: GuardrailError) -> CoreError {
CoreError::InvalidRequest(format!("{}: {}", error.kind, error.message))
fn guardrail_error_to_core_error(error: GuardrailError) -> Error {
Error::InvalidRequest(format!("{}: {}", error.kind, error.message))
}
fn core_error_kind(error: &CoreError) -> &'static str {
fn core_error_kind(error: &Error) -> &'static str {
match error {
CoreError::Auth(_) => "AuthError",
CoreError::InvalidProvider(_) => "InvalidProvider",
CoreError::InvalidRequest(_) => "InvalidRequest",
CoreError::InvalidType { .. } => "InvalidType",
CoreError::MissingField(_) => "MissingField",
CoreError::Http { .. } => "HttpError",
CoreError::InvalidResponse(_) => "InvalidResponse",
CoreError::Network(_) => "NetworkError",
CoreError::Connect(_) => "ConnectError",
CoreError::Routing(_) => "RoutingError",
CoreError::Unsupported(_) => "UnsupportedRequest",
Error::Auth(_) => "AuthError",
Error::InvalidProvider(_) => "InvalidProvider",
Error::InvalidRequest(_) => "InvalidRequest",
Error::InvalidType { .. } => "InvalidType",
Error::MissingField(_) => "MissingField",
Error::Http { .. } => "HttpError",
Error::InvalidResponse(_) => "InvalidResponse",
Error::Network(_) => "NetworkError",
Error::Connect(_) => "ConnectError",
Error::Routing(_) => "RoutingError",
Error::Unsupported(_) => "UnsupportedRequest",
}
}

9
litellm-rust/crates/ai-gateway/src/ocr/mod.rs
View file

@ -1,4 +1,4 @@
use litellm_core::CoreResult;
use litellm_core::Error;
use litellm_core::call_lifecycle::CallLifecycle;
use serde_json::Value;
@ -13,10 +13,13 @@ pub use types::OcrRequest;
use handler::execute_ocr_provider_call;
use prepare::{PreparedOcrCall, prepare_ocr_call};
pub async fn ocr(request: OcrRequest<'_>) -> CoreResult<Value> {
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
pub async fn ocr(request: OcrRequest<'_>) -> Result<Value, Error> {
let PreparedOcrCall { request, hooks } = prepare_ocr_call(request);
CallLifecycle::default()
.run_request(request, &hooks, execute_ocr_provider_call)
.run_request(request, &hooks, |request| {
execute_ocr_provider_call(request, &hooks)
})
.await
}

20
litellm-rust/crates/ai-gateway/src/ocr/prepare.rs
View file

@ -3,6 +3,7 @@ use std::time::{SystemTime, UNIX_EPOCH};
use litellm_core::routing_utils::provider::{CustomLlmProvider, get_custom_llm_provider};
use super::common_utils::ocr_provider_config;
use super::hooks::OcrLifecycleHooks;
use super::types::{OcrRequest, PreparedOcrRequest};
use crate::integrations::custom_guardrail::CustomGuardrailRunner;
@ -13,6 +14,7 @@ pub(crate) struct PreparedOcrCall {
pub(crate) hooks: OcrLifecycleHooks,
}
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
pub(crate) fn prepare_ocr_call(request: OcrRequest<'_>) -> PreparedOcrCall {
let call_id = request
.litellm_call_id
@ -25,9 +27,25 @@ pub(crate) fn prepare_ocr_call(request: OcrRequest<'_>) -> PreparedOcrCall {
});
let model = provider_info.model.to_string();
let custom_llm_provider = provider_info.custom_llm_provider.to_string();
let config = ocr_provider_config(&custom_llm_provider, &model)
.ok_or_else(|| litellm_core::Error::InvalidProvider(custom_llm_provider.clone()));
let optional_params = match &config {
Ok(config) => {
let supported = config.supported_ocr_params();
config.map_ocr_params(
&request
.optional_params
.into_iter()
.filter(|(name, _)| supported.contains(&name.as_str()))
.collect(),
)
}
Err(_) => request.optional_params,
};
PreparedOcrCall {
request: PreparedOcrRequest {
config,
model,
custom_llm_provider,
litellm_call_id: call_id,
@ -35,7 +53,7 @@ pub(crate) fn prepare_ocr_call(request: OcrRequest<'_>) -> PreparedOcrCall {
api_key: request.api_key.map(str::to_string),
api_base: request.api_base.map(str::to_string),
extra_headers: request.extra_headers,
optional_params: request.optional_params,
optional_params,
timeout: request.timeout,
},
hooks: OcrLifecycleHooks::new(

11
litellm-rust/crates/ai-gateway/src/ocr/tests.rs
View file

@ -1,13 +1,14 @@
use std::sync::{Arc, Mutex};
use std::time::Duration;
use litellm_core::error::CoreError;
use litellm_core::error::Error;
use litellm_core::http_utils::has_header;
use litellm_core::ocr::transformation::OcrResponseHandling;
use serde_json::{Map, Value, json};
use tokio::io::{AsyncReadExt, AsyncWriteExt};
use tokio::net::{TcpListener, TcpStream};
use super::common_utils::{has_header, ocr_provider_config, string_headers, truncate_error_body};
use super::common_utils::{ocr_provider_config, string_headers, truncate_error_body};
use super::{OcrRequest, ocr};
use crate::integrations::custom_guardrail::{
CustomGuardrail, GuardrailContext, GuardrailDecision, GuardrailError, GuardrailEventHook,
@ -395,7 +396,7 @@ async fn ocr_lifecycle_runs_failure_hook_on_provider_error() {
.await
.expect_err("provider error propagates");
assert!(matches!(err, CoreError::Http { status: 500, .. }));
assert!(matches!(err, Error::Http { status: 500, .. }));
server.await.expect("server task completes");
assert_eq!(
logger.events(),
@ -439,7 +440,7 @@ async fn ocr_lifecycle_pre_call_block_skips_provider_socket() {
.await
.expect_err("guardrail blocks request");
assert!(matches!(err, CoreError::InvalidRequest(_)));
assert!(matches!(err, Error::InvalidRequest(_)));
assert_eq!(guardrail.events(), vec!["async_pre_call_hook"]);
assert_eq!(
logger.events(),
@ -607,7 +608,7 @@ fn string_headers_rejects_non_string_values() {
let err = string_headers(Some(headers)).expect_err("non-string header rejected");
assert_eq!(
err,
CoreError::InvalidRequest(
Error::InvalidRequest(
"OCR extra_headers.x-retry-count must be a string, got number".to_string()
)
);

1
litellm-rust/crates/ai-gateway/src/ocr/types.rs
View file

@ -25,6 +25,7 @@ pub struct OcrRequest<'a> {
}
pub(crate) struct PreparedOcrRequest {
pub(crate) config: Result<&'static dyn OcrProviderConfig, litellm_core::Error>,
pub(crate) model: String,
pub(crate) custom_llm_provider: String,
pub(crate) litellm_call_id: String,

12
litellm-rust/crates/ai-gateway/src/python/config.rs
View file

@ -6,33 +6,31 @@
//! (and recorded in [`crate::gil`]); the realtime hot path never touches Python.
//!
//! Compiled only under the `python-config` feature.
use litellm_core::CoreResult;
use litellm_core::error::CoreError;
use litellm_core::error::Error;
use litellm_core::router::{Deployment, Router};
use pyo3::prelude::*;
use crate::gil;
/// Load the router's `model_list` from `config_path` via the Python reader.
pub fn load_router_from_config(config_path: &str) -> CoreResult<Router> {
pub fn load_router_from_config(config_path: &str) -> Result<Router, Error> {
gil::record_acquisition();
Python::attach(|py| {
let model_list = py
.import("litellm.proxy.read_model_list")
.and_then(|module| module.getattr("read_model_list"))
.and_then(|reader| reader.call1((config_path,)))
.map_err(|err| CoreError::Routing(format!("read_model_list failed: {err}")))?;
.map_err(|err| Error::Routing(format!("read_model_list failed: {err}")))?;
let model_list_json: String = py
.import("json")
.and_then(|json| json.getattr("dumps"))
.and_then(|dumps| dumps.call1((model_list,)))
.and_then(|encoded| encoded.extract())
.map_err(|err| CoreError::Routing(format!("serializing model_list failed: {err}")))?;
.map_err(|err| Error::Routing(format!("serializing model_list failed: {err}")))?;
let deployments: Vec<Deployment> = serde_json::from_str(&model_list_json)
.map_err(|err| CoreError::Routing(format!("parsing model_list failed: {err}")))?;
.map_err(|err| Error::Routing(format!("parsing model_list failed: {err}")))?;
Ok(Router::new(deployments))
})

38
litellm-rust/crates/ai-gateway/src/routes/messages/mod.rs
View file

@ -9,7 +9,7 @@ use axum::http::StatusCode;
use axum::http::header::{CACHE_CONTROL, CONTENT_TYPE, HeaderMap, HeaderValue};
use axum::response::{IntoResponse, Response};
use axum::routing::post;
use litellm_core::CoreError;
use litellm_core::Error;
use serde_json::{Map, Value};
use crate::auth::RequireMasterKey;
@ -46,7 +46,7 @@ fn stream_response(upstream: reqwest::Response) -> Result<Response, MessagesRout
let mut response = Response::builder()
.status(
StatusCode::from_u16(upstream.status().as_u16()).map_err(|error| {
MessagesRouteError(CoreError::InvalidResponse(format!(
MessagesRouteError(Error::InvalidResponse(format!(
"invalid upstream response status: {error}"
)))
})?,
@ -58,13 +58,13 @@ fn stream_response(upstream: reqwest::Response) -> Result<Response, MessagesRout
response
.body(Body::from_stream(upstream.bytes_stream()))
.map_err(|error| {
MessagesRouteError(CoreError::InvalidResponse(format!(
MessagesRouteError(Error::InvalidResponse(format!(
"failed to build streaming response: {error}"
)))
})
}
fn forwarded_headers(headers: &HeaderMap) -> Result<Option<Map<String, Value>>, CoreError> {
fn forwarded_headers(headers: &HeaderMap) -> Result<Option<Map<String, Value>>, Error> {
let forwarded = headers
.iter()
.filter(|(name, _)| {
@ -74,19 +74,19 @@ fn forwarded_headers(headers: &HeaderMap) -> Result<Option<Map<String, Value>>,
})
.map(|(name, value)| {
let value = value.to_str().map_err(|_| {
CoreError::InvalidRequest(format!("invalid value for header {}", name.as_str()))
Error::InvalidRequest(format!("invalid value for header {}", name.as_str()))
})?;
Ok((name.to_string(), Value::String(value.to_string())))
})
.collect::<Result<Map<_, _>, CoreError>>()?;
.collect::<Result<Map<_, _>, Error>>()?;
Ok((!forwarded.is_empty()).then_some(forwarded))
}
#[derive(Debug)]
struct MessagesRouteError(CoreError);
struct MessagesRouteError(Error);
impl From<CoreError> for MessagesRouteError {
fn from(error: CoreError) -> Self {
impl From<Error> for MessagesRouteError {
fn from(error: Error) -> Self {
Self(error)
}
}
@ -94,28 +94,28 @@ impl From<CoreError> for MessagesRouteError {
impl IntoResponse for MessagesRouteError {
fn into_response(self) -> Response {
let (status, message) = match self.0 {
CoreError::InvalidRequest(message) => (StatusCode::BAD_REQUEST, message),
CoreError::InvalidProvider(_) | CoreError::Routing(_) => (
Error::InvalidRequest(message) => (StatusCode::BAD_REQUEST, message),
Error::InvalidProvider(_) | Error::Routing(_) => (
StatusCode::NOT_FOUND,
"no messages deployment is configured for this model".to_string(),
),
CoreError::Auth(_) => (
Error::Auth(_) => (
StatusCode::BAD_GATEWAY,
"messages provider authentication failed".to_string(),
),
CoreError::Http { .. }
| CoreError::Network(_)
| CoreError::Connect(_)
| CoreError::InvalidResponse(_)
| CoreError::InvalidType { .. }
| CoreError::MissingField(_) => (
Error::Http { .. }
| Error::Network(_)
| Error::Connect(_)
| Error::InvalidResponse(_)
| Error::InvalidType { .. }
| Error::MissingField(_) => (
StatusCode::BAD_GATEWAY,
"messages provider request failed".to_string(),
),
// The gateway has no Python implementation to decline to, so a
// request the core cannot serve is reported to the caller. The
// reason is a fixed internal string, never provider content.
CoreError::Unsupported(reason) => (
Error::Unsupported(reason) => (
StatusCode::BAD_REQUEST,
format!("messages request is not supported: {reason}"),
),

16
litellm-rust/crates/ai-gateway/src/routes/messages/service.rs
View file

@ -1,10 +1,10 @@
use std::sync::Arc;
use litellm_core::Error;
use litellm_core::constants::ANTHROPIC_MESSAGES_PROVIDER;
use litellm_core::messages::types::MessagesRequest;
use litellm_core::messages::{messages, messages_stream};
use litellm_core::router::Router;
use litellm_core::{CoreError, CoreResult};
use serde_json::{Map, Value};
pub(crate) enum MessagesResponse {
@ -16,16 +16,16 @@ pub async fn run(
router: &Arc<Router>,
body: Value,
extra_headers: Option<Map<String, Value>>,
) -> CoreResult<MessagesResponse> {
) -> Result<MessagesResponse, Error> {
let model = body
.get("model")
.and_then(Value::as_str)
.map(str::trim)
.filter(|model| !model.is_empty())
.ok_or_else(|| CoreError::InvalidRequest("messages body requires a model".to_string()))?;
let deployment = router.get_available_deployment(model).ok_or_else(|| {
CoreError::Routing(format!("no deployment available for model '{model}'"))
})?;
.ok_or_else(|| Error::InvalidRequest("messages body requires a model".to_string()))?;
let deployment = router
.get_available_deployment(model)
.ok_or_else(|| Error::Routing(format!("no deployment available for model '{model}'")))?;
let provider_model = deployment.litellm_params.model.as_str();
let upstream_model = provider_model
.split_once('/')
@ -37,7 +37,7 @@ pub async fn run(
};
let mut body = body;
body.as_object_mut()
.ok_or_else(|| CoreError::InvalidRequest("messages body must be an object".to_string()))?
.ok_or_else(|| Error::InvalidRequest("messages body must be an object".to_string()))?
.insert(
"model".to_string(),
Value::String(upstream_model.to_string()),
@ -60,6 +60,6 @@ pub async fn run(
serde_json::to_value(response)
.map(MessagesResponse::Json)
.map_err(|err| {
CoreError::InvalidResponse(format!("failed to serialize messages response: {err}"))
Error::InvalidResponse(format!("failed to serialize messages response: {err}"))
})
}

11
litellm-rust/crates/ai-gateway/src/routes/realtime/service.rs
View file

@ -11,8 +11,7 @@ use std::time::Duration;
use crate::io::realtime_pool::{RealtimePool, upstream_key};
use futures_util::{Sink, Stream};
use litellm_core::CoreResult;
use litellm_core::error::CoreError;
use litellm_core::error::Error;
use litellm_core::realtime::types::RealtimeEvent;
use litellm_core::router::Router;
@ -29,15 +28,15 @@ pub async fn run<In, Out>(
observe: impl FnMut(&RealtimeEvent) + Send,
client_in: In,
client_out: Out,
) -> CoreResult<()>
) -> Result<(), Error>
where
In: Stream<Item = RealtimeEvent> + Unpin + Send,
Out: Sink<RealtimeEvent> + Unpin + Send,
<Out as Sink<RealtimeEvent>>::Error: std::fmt::Display,
{
let deployment = router.get_available_deployment(model).ok_or_else(|| {
CoreError::Routing(format!("no deployment available for model '{model}'"))
})?;
let deployment = router
.get_available_deployment(model)
.ok_or_else(|| Error::Routing(format!("no deployment available for model '{model}'")))?;
let params = &deployment.litellm_params;
// Strip a leading `openai/` so the OpenAI-only realtime fn gets the bare model.
let provider_model = params

12
litellm-rust/crates/ai-gateway/src/routes/responses/service.rs
View file

@ -2,13 +2,13 @@ use std::sync::Arc;
use std::time::Duration;
use futures_util::{Sink, Stream};
use litellm_core::Error;
use litellm_core::call_lifecycle::{CallLifecycle, CallLifecycleContext};
use litellm_core::responses::instrumentation::{
ResponsesWsCallbackPayload, ResponsesWsInstrumentation, ResponsesWsLogOutcome,
ResponsesWsMetadata,
};
use litellm_core::responses::types::ResponsesWsEvent;
use litellm_core::{CoreError, CoreResult};
use crate::integrations::custom_logger::{
CallbackTiming, CallbackValue, CustomLogger, CustomLoggerRunner, LoggingError, ModelCallDetails,
@ -26,22 +26,22 @@ pub async fn run<In, Out>(
metadata: RequestMetadata,
client_in: In,
client_out: Out,
) -> CoreResult<()>
) -> Result<(), Error>
where
In: Stream<Item = ResponsesWsEvent> + Unpin + Send,
Out: Sink<ResponsesWsEvent> + Unpin + Send,
Out::Error: std::fmt::Display,
{
let deployment = router.get_available_deployment(model).ok_or_else(|| {
CoreError::Routing(format!("no deployment available for model '{model}'"))
})?;
let deployment = router
.get_available_deployment(model)
.ok_or_else(|| Error::Routing(format!("no deployment available for model '{model}'")))?;
let params = &deployment.litellm_params;
let provider_model = params
.model
.strip_prefix("openai/")
.unwrap_or(&params.model);
if params.model.contains('/') && !params.model.starts_with("openai/") {
return Err(CoreError::InvalidProvider(
return Err(Error::InvalidProvider(
"Responses WebSocket route supports OpenAI deployments only".to_string(),
));
}

1
litellm-rust/crates/core/Cargo.toml
View file

@ -11,6 +11,7 @@ reqwest.workspace = true
serde.workspace = true
serde_json.workspace = true
thiserror.workspace = true
tracing.workspace = true
sha2.workspace = true
aws-config = { version = "1.9.0", default-features = false, features = ["rustls", "rt-tokio"], optional = true }
aws-credential-types = { version = "1.3.0", features = ["hardcoded-credentials"], optional = true }

14
litellm-rust/crates/core/src/audio_transcription/client.rs Normal file
View file

@ -0,0 +1,14 @@
use std::sync::OnceLock;
use std::time::Duration;
use crate::constants::AUDIO_TRANSCRIPTION_TIMEOUT_SECS;
pub(super) fn http_client() -> &'static reqwest::Client {
static CLIENT: OnceLock<reqwest::Client> = OnceLock::new();
CLIENT.get_or_init(|| {
reqwest::Client::builder()
.timeout(Duration::from_secs(AUDIO_TRANSCRIPTION_TIMEOUT_SECS))
.build()
.unwrap_or_else(|_| reqwest::Client::new())
})
}

91
litellm-rust/crates/core/src/audio_transcription/handler.rs Normal file
View file

@ -0,0 +1,91 @@
use serde_json::Value;
use crate::error::Error;
use crate::http_utils::{http_request, truncate_error_body};
use super::client::http_client;
use super::types::ProviderAudioTranscriptionRequest;
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
pub async fn execute_audio_transcription_provider_call(
request: ProviderAudioTranscriptionRequest,
) -> Result<Value, Error> {
let body = serde_json::to_vec(&request.body)
.map_err(|error| Error::InvalidRequest(format!("invalid audio request body: {error}")))?;
let headers = signed_headers(&request, &body).await?;
let mut request_builder = http_client().post(&request.url).body(body);
for (key, value) in headers {
request_builder = request_builder.header(key, value);
}
if let Some(duration) = request.timeout {
request_builder = request_builder.timeout(duration);
}
let response = http_request(request_builder)
.await
.map_err(|error| Error::Network(error.to_string()))?;
let status = response.status();
let text = response
.text()
.await
.map_err(|error| Error::Network(error.to_string()))?;
if !status.is_success() {
return Err(Error::Http {
status: status.as_u16(),
body: truncate_error_body(&text),
});
}
let response_json = serde_json::from_str(&text)
.map_err(|error| Error::InvalidResponse(format!("invalid audio response JSON: {error}")))?;
Ok(request
.config
.transform_transcription_response(&request.model, response_json)?
.into_json())
}
#[cfg(feature = "bedrock-auth")]
async fn signed_headers(
request: &ProviderAudioTranscriptionRequest,
body: &[u8],
) -> Result<Vec<(String, String)>, Error> {
use std::collections::BTreeMap;
use std::time::SystemTime;
use crate::audio_transcription::transformation::AudioTranscriptionAuth;
use crate::providers::bedrock::audio_transcription::aws_auth_config;
use crate::providers::bedrock::aws_base::{resolve_credentials, sign_bedrock_post};
let AudioTranscriptionAuth::AwsSigV4 { region, .. } = &request.auth else {
return Ok(request.upstream_headers.clone());
};
let env_lookup = |key: &str| std::env::var(key).ok();
let credentials = resolve_credentials(
aws_auth_config(&request.optional_params, &env_lookup),
&env_lookup,
)
.await?;
let unsigned: BTreeMap<String, String> = request.upstream_headers.iter().cloned().collect();
let signature = sign_bedrock_post(
&request.url,
body,
&unsigned,
region,
&credentials,
SystemTime::now(),
)?;
Ok(unsigned.into_iter().chain(signature).collect())
}
#[cfg(not(feature = "bedrock-auth"))]
async fn signed_headers(
request: &ProviderAudioTranscriptionRequest,
_body: &[u8],
) -> Result<Vec<(String, String)>, Error> {
use crate::audio_transcription::transformation::AudioTranscriptionAuth;
match request.auth {
AudioTranscriptionAuth::AwsSigV4 { .. } => Err(Error::Unsupported(
"AWS SigV4 requires the bedrock-auth feature",
)),
AudioTranscriptionAuth::Bearer => Ok(request.upstream_headers.clone()),
}
}

19
litellm-rust/crates/core/src/audio_transcription/mod.rs
View file

@ -1,2 +1,21 @@
use crate::Error;
mod client;
mod handler;
mod prepare;
pub mod transformation;
pub mod types;
use serde_json::Value;
pub use handler::execute_audio_transcription_provider_call;
pub use prepare::prepare_audio_transcription_provider_call;
pub use types::{AudioTranscriptionRequest, ProviderAudioTranscriptionRequest};
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
pub async fn audio_transcription(request: AudioTranscriptionRequest<'_>) -> Result<Value, Error> {
execute_audio_transcription_provider_call(prepare_audio_transcription_provider_call(request)?)
.await
}
#[cfg(test)]
mod tests;

74
litellm-rust/crates/core/src/audio_transcription/prepare.rs Normal file
View file

@ -0,0 +1,74 @@
use crate::error::Error;
use crate::http_utils::{has_header, string_headers};
#[cfg(feature = "bedrock-auth")]
use crate::providers::bedrock::audio_transcription::BEDROCK_AUDIO_TRANSCRIPTION_CONFIG;
use crate::routing_utils::provider::{CustomLlmProvider, get_custom_llm_provider};
use super::transformation::{AudioTranscriptionAuth, AudioTranscriptionProviderConfig};
use super::types::{AudioTranscriptionRequest, ProviderAudioTranscriptionRequest};
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
fn provider_config(provider: &str) -> Option<&'static dyn AudioTranscriptionProviderConfig> {
#[cfg(feature = "bedrock-auth")]
if provider == "bedrock" {
return Some(&BEDROCK_AUDIO_TRANSCRIPTION_CONFIG);
}
let _ = provider;
None
}
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
pub fn prepare_audio_transcription_provider_call(
request: AudioTranscriptionRequest<'_>,
) -> Result<ProviderAudioTranscriptionRequest, Error> {
let provider_info = get_custom_llm_provider(request.model, request.custom_llm_provider)
.or_else(|| {
request
.custom_llm_provider
.map(|provider| CustomLlmProvider {
model: request.model,
custom_llm_provider: provider,
})
})
.ok_or_else(|| {
Error::InvalidProvider(
"unable to resolve custom_llm_provider for audio transcription request".to_string(),
)
})?;
let model = provider_info.model.to_string();
let config = provider_config(provider_info.custom_llm_provider)
.ok_or_else(|| Error::InvalidProvider(provider_info.custom_llm_provider.to_string()))?;
let env_lookup = |key: &str| std::env::var(key).ok();
let mut headers = string_headers("audio transcription", request.extra_headers)?;
let auth = config.auth_strategy(&model, &request.optional_params, &env_lookup)?;
if matches!(auth, AudioTranscriptionAuth::Bearer)
&& !has_header(&headers, "authorization")
&& let Some(api_key) = request.api_key
{
headers.push(("Authorization".to_string(), format!("Bearer {api_key}")));
}
if !has_header(&headers, "content-type") {
headers.push(("Content-Type".to_string(), "application/json".to_string()));
}
let url = config.complete_url(
request.api_base,
&model,
&request.optional_params,
&env_lookup,
)?;
let filtered_params = config.map_transcription_params(&request.optional_params);
let transformed =
config.transform_transcription_request(&model, request.audio, filtered_params)?;
Ok(ProviderAudioTranscriptionRequest {
model,
custom_llm_provider: provider_info.custom_llm_provider.to_string(),
config,
url,
body: transformed.body,
upstream_headers: headers,
auth,
#[cfg(feature = "bedrock-auth")]
optional_params: request.optional_params,
timeout: request.timeout,
})
}

50
litellm-rust/crates/core/src/audio_transcription/tests.rs Normal file
View file

@ -0,0 +1,50 @@
use std::io::{Read, Write};
use std::net::TcpListener;
use std::thread;
use serde_json::{Map, json};
use super::audio_transcription;
use super::types::AudioTranscriptionRequest;
#[tokio::test]
async fn bedrock_request_is_signed_and_contains_audio() {
let listener = TcpListener::bind("127.0.0.1:0").expect("listener");
let address = listener.local_addr().expect("address");
let server = thread::spawn(move || {
let (mut stream, _) = listener.accept().expect("connection");
let mut request = Vec::new();
let mut buffer = [0_u8; 16_384];
let count = stream.read(&mut buffer).expect("request");
request.extend_from_slice(&buffer[..count]);
let request = String::from_utf8_lossy(&request);
assert!(request.contains("POST /model/mistral.voxtral-mini-3b-2507/converse"));
assert!(request.contains("authorization: AWS4-HMAC-SHA256"));
assert!(request.contains("x-amz-date:"));
assert!(request.contains("\"bytes\":\"AQI=\""));
assert!(request.contains("Transcribe the audio. Respond with only the transcript."));
let response = b"HTTP/1.1 200 OK\r\nContent-Type: application/json\r\nContent-Length: 53\r\nConnection: close\r\n\r\n{\"output\":{\"message\":{\"content\":[{\"text\":\"hello\"}]}}}";
stream.write_all(response).expect("response");
});
let optional_params = Map::from_iter([
("aws_access_key_id".to_string(), json!("access-key")),
("aws_secret_access_key".to_string(), json!("secret-key")),
("aws_region_name".to_string(), json!("us-east-1")),
]);
let api_base = format!("http://{address}");
let response = audio_transcription(AudioTranscriptionRequest {
model: "mistral.voxtral-mini-3b-2507",
audio: json!({"data": "AQI=", "format": "wav", "filename": "audio.wav"}),
api_key: None,
api_base: Some(&api_base),
custom_llm_provider: Some("bedrock"),
extra_headers: None,
optional_params,
timeout: None,
})
.await
.expect("transcription");
assert_eq!(response, json!({"text": "hello"}));
server.join().expect("server");
}

12
litellm-rust/crates/core/src/audio_transcription/transformation.rs
View file

@ -1,7 +1,6 @@
use crate::Error;
use serde_json::{Map, Value};
use crate::CoreResult;
use super::types::{AudioTranscriptionRequestData, AudioTranscriptionResponseData};
#[derive(Clone, Debug, PartialEq, Eq)]
@ -16,6 +15,7 @@ pub enum AudioTranscriptionAuth {
pub trait AudioTranscriptionProviderConfig: Sync {
fn supported_transcription_params(&self) -> &'static [&'static str];
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
fn map_transcription_params(&self, params: &Map<String, Value>) -> Map<String, Value> {
params
.iter()
@ -32,13 +32,13 @@ pub trait AudioTranscriptionProviderConfig: Sync {
model: &str,
audio: Value,
optional_params: Map<String, Value>,
) -> CoreResult<AudioTranscriptionRequestData>;
) -> Result<AudioTranscriptionRequestData, Error>;
fn transform_transcription_response(
&self,
model: &str,
response_json: Value,
) -> CoreResult<AudioTranscriptionResponseData>;
) -> Result<AudioTranscriptionResponseData, Error>;
fn complete_url(
&self,
@ -46,12 +46,12 @@ pub trait AudioTranscriptionProviderConfig: Sync {
model: &str,
optional_params: &Map<String, Value>,
env_lookup: &dyn Fn(&str) -> Option<String>,
) -> CoreResult<String>;
) -> Result<String, Error>;
fn auth_strategy(
&self,
model: &str,
optional_params: &Map<String, Value>,
env_lookup: &dyn Fn(&str) -> Option<String>,
) -> CoreResult<AudioTranscriptionAuth>;
) -> Result<AudioTranscriptionAuth, Error>;
}

53
litellm-rust/crates/core/src/audio_transcription/types.rs
View file

@ -1,5 +1,56 @@
use std::time::Duration;
use serde::{Deserialize, Serialize};
use serde_json::Value;
use serde_json::{Map, Value};
use super::transformation::{AudioTranscriptionAuth, AudioTranscriptionProviderConfig};
pub struct AudioTranscriptionRequest<'a> {
pub model: &'a str,
pub audio: Value,
pub api_key: Option<&'a str>,
pub api_base: Option<&'a str>,
pub custom_llm_provider: Option<&'a str>,
pub extra_headers: Option<Map<String, Value>>,
pub optional_params: Map<String, Value>,
pub timeout: Option<Duration>,
}
#[derive(Clone)]
pub struct ProviderAudioTranscriptionRequest {
pub(super) model: String,
pub(super) custom_llm_provider: String,
pub(super) config: &'static dyn AudioTranscriptionProviderConfig,
pub(super) url: String,
pub(super) body: Value,
pub(super) upstream_headers: Vec<(String, String)>,
pub(super) auth: AudioTranscriptionAuth,
#[cfg(feature = "bedrock-auth")]
pub(super) optional_params: Map<String, Value>,
pub(super) timeout: Option<Duration>,
}
impl ProviderAudioTranscriptionRequest {
pub fn model(&self) -> &str {
&self.model
}
pub fn custom_llm_provider(&self) -> &str {
&self.custom_llm_provider
}
pub fn url(&self) -> &str {
&self.url
}
pub fn body(&self) -> &Value {
&self.body
}
pub fn with_body(self, body: Value) -> Self {
Self { body, ..self }
}
}
#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
pub struct AudioTranscriptionRequestData {

34
litellm-rust/crates/core/src/call_lifecycle/mod.rs
View file

@ -1,7 +1,7 @@
use std::future::Future;
use std::time::{Instant, SystemTime, UNIX_EPOCH};
use crate::{CoreError, CoreResult};
use crate::Error;
pub mod types;
@ -11,14 +11,14 @@ pub use types::{
};
pub trait CallLifecycleHooks<InitialReq, ProviderReq, Resp>: Send + Sync {
type PreCallFuture<'a>: Future<Output = CoreResult<InitialReq>> + Send + 'a
type PreCallFuture<'a>: Future<Output = Result<InitialReq, Error>> + Send + 'a
where
Self: 'a,
InitialReq: 'a,
ProviderReq: 'a,
Resp: 'a;
type DuringCallFuture<'a>: Future<Output = CoreResult<ProviderReq>> + Send + 'a
type DuringCallFuture<'a>: Future<Output = Result<ProviderReq, Error>> + Send + 'a
where
Self: 'a,
InitialReq: 'a,
@ -56,7 +56,7 @@ pub trait CallLifecycleHooks<InitialReq, ProviderReq, Resp>: Send + Sync {
fn async_log_failure_event<'a>(
&'a self,
context: &'a CallLifecycleContext,
error: &'a CoreError,
error: &'a Error,
timing: &'a CallLifecycleTiming,
) -> Self::FailureFuture<'a>;
}
@ -86,12 +86,12 @@ impl<'a> CallLifecycle<'a> {
request: InitialReq,
hooks: &Hooks,
provider_call: ProviderCall,
) -> CoreResult<Resp>
) -> Result<Resp, Error>
where
InitialReq: CallLifecycleRequest,
Hooks: CallLifecycleHooks<InitialReq, ProviderReq, Resp>,
ProviderCall: FnOnce(ProviderReq) -> ProviderFuture,
ProviderFuture: Future<Output = CoreResult<Resp>>,
ProviderFuture: Future<Output = Result<Resp, Error>>,
{
let context = request.lifecycle_context();
self.run(context, request, hooks, provider_call).await
@ -103,11 +103,11 @@ impl<'a> CallLifecycle<'a> {
request: InitialReq,
hooks: &Hooks,
provider_call: ProviderCall,
) -> CoreResult<Resp>
) -> Result<Resp, Error>
where
Hooks: CallLifecycleHooks<InitialReq, ProviderReq, Resp>,
ProviderCall: FnOnce(ProviderReq) -> ProviderFuture,
ProviderFuture: Future<Output = CoreResult<Resp>>,
ProviderFuture: Future<Output = Result<Resp, Error>>,
{
let call_start = epoch_seconds();
let mut phases = Vec::new();
@ -166,7 +166,7 @@ impl<'a> CallLifecycle<'a> {
&self,
context: &CallLifecycleContext,
hooks: &Hooks,
error: &CoreError,
error: &Error,
call_start: f64,
phases: &mut Vec<CallLifecyclePhaseTiming>,
) where
@ -251,8 +251,8 @@ mod tests {
}
impl CallLifecycleHooks<String, String, String> for RecordingHooks {
type PreCallFuture<'a> = BoxFuture<'a, CoreResult<String>>;
type DuringCallFuture<'a> = BoxFuture<'a, CoreResult<String>>;
type PreCallFuture<'a> = BoxFuture<'a, Result<String, Error>>;
type DuringCallFuture<'a> = BoxFuture<'a, Result<String, Error>>;
type SuccessFuture<'a> = BoxFuture<'a, ()>;
type FailureFuture<'a> = BoxFuture<'a, ()>;
@ -294,7 +294,7 @@ mod tests {
fn async_log_failure_event<'a>(
&'a self,
_context: &'a CallLifecycleContext,
_error: &'a CoreError,
_error: &'a Error,
_timing: &'a CallLifecycleTiming,
) -> Self::FailureFuture<'a> {
Box::pin(async move {
@ -304,8 +304,8 @@ mod tests {
}
impl CallLifecycleHooks<RecordingRequest, String, String> for RecordingHooks {
type PreCallFuture<'a> = BoxFuture<'a, CoreResult<RecordingRequest>>;
type DuringCallFuture<'a> = BoxFuture<'a, CoreResult<String>>;
type PreCallFuture<'a> = BoxFuture<'a, Result<RecordingRequest, Error>>;
type DuringCallFuture<'a> = BoxFuture<'a, Result<String, Error>>;
type SuccessFuture<'a> = BoxFuture<'a, ()>;
type FailureFuture<'a> = BoxFuture<'a, ()>;
@ -345,7 +345,7 @@ mod tests {
fn async_log_failure_event<'a>(
&'a self,
_context: &'a CallLifecycleContext,
_error: &'a CoreError,
_error: &'a Error,
_timing: &'a CallLifecycleTiming,
) -> Self::FailureFuture<'a> {
Box::pin(async move {
@ -383,13 +383,13 @@ mod tests {
"request".to_string(),
&hooks,
|_request| async move {
Err::<String, CoreError>(CoreError::Network("provider down".to_string()))
Err::<String, Error>(Error::Network("provider down".to_string()))
},
)
.await
.expect_err("call fails");
assert_eq!(error, CoreError::Network("provider down".to_string()));
assert_eq!(error, Error::Network("provider down".to_string()));
assert_eq!(hooks.events(), vec!["pre_call", "during_call", "failure"]);
}

8
litellm-rust/crates/core/src/chat_completions/common_utils.rs
View file

@ -1,13 +1,13 @@
use serde_json::{Map, Value};
use crate::error::CoreResult;
use crate::Error;
use crate::http_utils::string_headers as shared_string_headers;
use crate::providers::anthropic::chat_completions::transformation::ANTHROPIC_CHAT_COMPLETIONS_CONFIG;
use serde_json::{Map, Value};
use super::transformation::ChatCompletionsProviderConfig;
const HEADER_CONTEXT: &str = "chat completions";
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
pub(super) fn chat_completions_provider_config(
provider: &str,
) -> Option<&'static dyn ChatCompletionsProviderConfig> {
@ -23,6 +23,6 @@ pub(super) fn chat_completions_provider_config(
pub(super) fn string_headers(
extra_headers: Option<Map<String, Value>>,
) -> CoreResult<Vec<(String, String)>> {
) -> Result<Vec<(String, String)>, Error> {
shared_string_headers(HEADER_CONTEXT, extra_headers)
}

40
litellm-rust/crates/core/src/chat_completions/handler.rs
View file

@ -1,19 +1,23 @@
use serde_json::Value;
use crate::error::{CoreError, CoreResult};
use crate::http_utils::truncate_error_body;
use crate::error::Error;
use crate::http_utils::{http_request, truncate_error_body};
use super::client::http_client;
use super::prepare::prepare_provider_request;
use super::transformation::ChatCompletionsAuth;
use super::types::{
ChatCompletionsResponse, ProviderChatCompletionsRequest, ProviderChatResponseData,
ResolvedChatCompletionsRequest,
};
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
pub(super) async fn execute_chat_completions_provider_call(
request: ProviderChatCompletionsRequest,
) -> CoreResult<ChatCompletionsResponse> {
request: ResolvedChatCompletionsRequest<'_>,
) -> Result<ChatCompletionsResponse, Error> {
let request = prepare_provider_request(request)?;
let body = serde_json::to_vec(&request.body).map_err(|err| {
CoreError::InvalidRequest(format!(
Error::InvalidRequest(format!(
"failed to serialize chat completions request: {err}"
))
})?;
@ -27,14 +31,14 @@ pub(super) async fn execute_chat_completions_provider_call(
request_builder = request_builder.timeout(duration);
}
let response = request_builder.send().await.map_err(|err| {
let response = http_request(request_builder).await.map_err(|err| {
// Failing to establish the connection means the request never went out,
// so the host can still serve it. Everything else here, a timeout
// above all, may have reached the provider and been answered.
if err.is_connect() || err.is_builder() {
CoreError::Connect(err.to_string())
Error::Connect(err.to_string())
} else {
CoreError::Network(err.to_string())
Error::Network(err.to_string())
}
})?;
@ -42,17 +46,17 @@ pub(super) async fn execute_chat_completions_provider_call(
let text = response
.text()
.await
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
if !status.is_success() {
return Err(CoreError::Http {
return Err(Error::Http {
status: status.as_u16(),
body: truncate_error_body(&text),
});
}
let body: Value = serde_json::from_str(&text).map_err(|err| {
CoreError::InvalidResponse(format!("invalid chat completions response JSON: {err}"))
Error::InvalidResponse(format!("invalid chat completions response JSON: {err}"))
})?;
request
.config
@ -69,10 +73,10 @@ pub(super) async fn execute_chat_completions_provider_call(
/// second kind has already been billed, and a host that keeps a reference
/// implementation must not retry those, so collapse them to one variant that
/// can only mean the provider was already called.
pub(super) fn as_response_error(err: CoreError) -> CoreError {
pub(super) fn as_response_error(err: Error) -> Error {
match err {
already @ (CoreError::InvalidResponse(_) | CoreError::Http { .. }) => already,
other => CoreError::InvalidResponse(other.to_string()),
already @ (Error::InvalidResponse(_) | Error::Http { .. }) => already,
other => Error::InvalidResponse(other.to_string()),
}
}
@ -80,7 +84,7 @@ pub(super) fn as_response_error(err: CoreError) -> CoreError {
pub(super) async fn signed_headers(
request: &ProviderChatCompletionsRequest,
body: &[u8],
) -> CoreResult<Vec<(String, String)>> {
) -> Result<Vec<(String, String)>, Error> {
use std::collections::BTreeMap;
use std::time::SystemTime;
@ -101,7 +105,7 @@ pub(super) async fn signed_headers(
.iter()
.any(|(name, _)| is_sigv4_computed_header(name))
{
return Err(CoreError::Unsupported(
return Err(Error::Unsupported(
"request forwards a header AWS SigV4 computes",
));
}
@ -137,9 +141,9 @@ pub(super) async fn signed_headers(
pub(super) async fn signed_headers(
request: &ProviderChatCompletionsRequest,
_body: &[u8],
) -> CoreResult<Vec<(String, String)>> {
) -> Result<Vec<(String, String)>, Error> {
match &request.auth {
ChatCompletionsAuth::AwsSigV4 { .. } => Err(CoreError::Unsupported(
ChatCompletionsAuth::AwsSigV4 { .. } => Err(Error::Unsupported(
"AWS SigV4 requires the bedrock-auth feature",
)),
_ => Ok(request.upstream_headers.clone()),

10
litellm-rust/crates/core/src/chat_completions/mod.rs
View file

@ -6,6 +6,7 @@
//! credentials, and it resolves the provider, translates the conversation,
//! calls the provider, and returns a typed OpenAI-shaped response.
use crate::Error;
mod client;
mod common_utils;
pub mod conversation;
@ -17,16 +18,15 @@ pub mod types;
use serde_json::{Map, Value};
use crate::error::CoreResult;
use handler::execute_chat_completions_provider_call;
use prepare::{parse_messages, prepare_chat_completions_call, resolve_provider_config};
use prepare::{parse_messages, resolve_provider_config, resolve_request};
use types::{ChatCompletionsRequest, ChatCompletionsResponse};
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
pub async fn chat_completions(
request: ChatCompletionsRequest<'_>,
) -> CoreResult<ChatCompletionsResponse> {
execute_chat_completions_provider_call(prepare_chat_completions_call(request)?).await
) -> Result<ChatCompletionsResponse, Error> {
execute_chat_completions_provider_call(resolve_request(request)?).await
}
/// Whether the core would accept this request, without resolving credentials or

63
litellm-rust/crates/core/src/chat_completions/prepare.rs
View file

@ -1,17 +1,20 @@
use serde_json::Value;
use crate::error::{CoreError, CoreResult};
use crate::error::Error;
use crate::http_utils::has_header;
use crate::routing_utils::provider::{CustomLlmProvider, get_custom_llm_provider};
use super::common_utils::{chat_completions_provider_config, string_headers};
use super::transformation::{ChatCompletionsAuth, ChatCompletionsProviderConfig};
use super::types::{ChatCompletionsRequest, ChatMessage, ProviderChatCompletionsRequest};
use super::types::{
ChatCompletionsRequest, ChatMessage, ProviderChatCompletionsRequest,
ResolvedChatCompletionsRequest,
};
pub(super) fn resolve_provider_config<'a>(
model: &'a str,
custom_llm_provider: Option<&'a str>,
) -> CoreResult<(String, &'static dyn ChatCompletionsProviderConfig)> {
) -> Result<(String, &'static dyn ChatCompletionsProviderConfig), Error> {
let provider_info = get_custom_llm_provider(model, custom_llm_provider)
.or_else(|| {
custom_llm_provider.map(|provider| CustomLlmProvider {
@ -20,41 +23,56 @@ pub(super) fn resolve_provider_config<'a>(
})
})
.ok_or_else(|| {
CoreError::InvalidProvider(
Error::InvalidProvider(
"unable to resolve custom_llm_provider for chat completions request".to_string(),
)
})?;
let config = chat_completions_provider_config(provider_info.custom_llm_provider)
.ok_or_else(|| CoreError::InvalidProvider(provider_info.custom_llm_provider.to_string()))?;
.ok_or_else(|| Error::InvalidProvider(provider_info.custom_llm_provider.to_string()))?;
Ok((provider_info.model.to_string(), config))
}
pub(super) fn parse_messages(messages: Value) -> CoreResult<Vec<ChatMessage>> {
serde_json::from_value(messages).map_err(|err| {
CoreError::InvalidRequest(format!("invalid chat completions messages: {err}"))
})
pub(super) fn parse_messages(messages: Value) -> Result<Vec<ChatMessage>, Error> {
serde_json::from_value(messages)
.map_err(|err| Error::InvalidRequest(format!("invalid chat completions messages: {err}")))
}
pub(super) fn prepare_chat_completions_call(
pub(super) fn resolve_request(
request: ChatCompletionsRequest<'_>,
) -> CoreResult<ProviderChatCompletionsRequest> {
) -> Result<ResolvedChatCompletionsRequest<'_>, Error> {
let (model, config) = resolve_provider_config(request.model, request.custom_llm_provider)?;
let env_lookup = |key: &str| std::env::var(key).ok();
let messages = parse_messages(request.messages)?;
if messages.is_empty() {
return Err(CoreError::InvalidRequest(
return Err(Error::InvalidRequest(
"chat completions requires at least one message".to_string(),
));
}
if let Some(reason) = config.unsupported_reason(&messages, &request.optional_params) {
return Err(CoreError::Unsupported(reason.0));
return Err(Error::Unsupported(reason.0));
}
Ok(ResolvedChatCompletionsRequest {
model,
config,
messages,
optional_params: request.optional_params,
api_key: request.api_key,
api_base: request.api_base,
extra_headers: request.extra_headers,
timeout: request.timeout,
})
}
let mut headers = string_headers(request.extra_headers)?;
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
fn validate_environment(
request: &ResolvedChatCompletionsRequest<'_>,
model: &str,
config: &dyn ChatCompletionsProviderConfig,
) -> Result<(Vec<(String, String)>, ChatCompletionsAuth), Error> {
let env_lookup = |key: &str| std::env::var(key).ok();
let mut headers = string_headers(request.extra_headers.clone())?;
let auth = config.auth(
request.api_key,
&model,
model,
&request.optional_params,
&env_lookup,
)?;
@ -95,7 +113,16 @@ pub(super) fn prepare_chat_completions_call(
headers.push(((*name).to_string(), (*value).to_string()));
}
}
Ok((headers, auth))
}
pub(super) fn prepare_provider_request(
request: ResolvedChatCompletionsRequest<'_>,
) -> Result<ProviderChatCompletionsRequest, Error> {
let (headers, auth) = validate_environment(&request, &request.model, request.config)?;
let model = request.model;
let config = request.config;
let env_lookup = |key: &str| std::env::var(key).ok();
let url = config.complete_url(
request.api_base,
&model,
@ -103,7 +130,7 @@ pub(super) fn prepare_chat_completions_call(
&env_lookup,
)?;
let transformed =
config.transform_request(&model, messages, request.optional_params.clone())?;
config.transform_request(&model, request.messages, request.optional_params.clone())?;
Ok(ProviderChatCompletionsRequest {
model,

50
litellm-rust/crates/core/src/chat_completions/tests.rs
View file

@ -1,10 +1,16 @@
use serde_json::{Map, Value, json};
use crate::error::CoreError;
use crate::error::Error;
use super::prepare::prepare_chat_completions_call;
use super::prepare::{prepare_provider_request, resolve_request};
use super::transformation::ChatCompletionsAuth;
use super::types::ChatCompletionsRequest;
use super::types::{ChatCompletionsRequest, ProviderChatCompletionsRequest};
fn prepare_chat_completions_call(
request: ChatCompletionsRequest<'_>,
) -> Result<ProviderChatCompletionsRequest, Error> {
prepare_provider_request(resolve_request(request)?)
}
fn request<'a>(
model: &'a str,
@ -29,7 +35,7 @@ fn request<'a>(
/// `ProviderChatCompletionsRequest` deliberately has no `Debug` (its headers
/// carry resolved credentials), so unwrap the failure case by hand.
fn decline(request: ChatCompletionsRequest<'_>) -> CoreError {
fn decline(request: ChatCompletionsRequest<'_>) -> Error {
match prepare_chat_completions_call(request) {
Err(error) => error,
Ok(prepared) => panic!("expected a decline, prepared a call to {}", prepared.url),
@ -196,7 +202,7 @@ fn declines_an_unsupported_request_before_resolving_credentials() {
call.api_key = None;
// No api_key is set and no env is consulted: the gate must run first, so the
// error is the decline rather than a missing-credential error.
assert_eq!(decline(call), CoreError::Unsupported("streaming"));
assert_eq!(decline(call), Error::Unsupported("streaming"));
}
#[test]
@ -208,7 +214,7 @@ fn rejects_an_unknown_provider() {
json!([{"role": "user", "content": "hi"}]),
json!({}),
)),
CoreError::InvalidProvider("openai".to_string())
Error::InvalidProvider("openai".to_string())
);
}
@ -221,7 +227,7 @@ fn rejects_a_model_with_no_resolvable_provider() {
json!([{"role": "user", "content": "hi"}]),
json!({}),
)),
CoreError::InvalidProvider(_)
Error::InvalidProvider(_)
));
}
@ -234,7 +240,7 @@ fn rejects_an_empty_or_malformed_message_list() {
json!([]),
json!({}),
)),
CoreError::InvalidRequest("chat completions requires at least one message".to_string())
Error::InvalidRequest("chat completions requires at least one message".to_string())
);
assert!(matches!(
decline(request(
@ -243,7 +249,7 @@ fn rejects_an_empty_or_malformed_message_list() {
json!("not a list"),
json!({}),
)),
CoreError::InvalidRequest(_)
Error::InvalidRequest(_)
));
}
@ -258,7 +264,7 @@ fn rejects_non_string_extra_headers() {
call.extra_headers = Some(Map::from_iter([("x-trace".to_string(), json!(7))]));
assert_eq!(
decline(call),
CoreError::InvalidRequest(
Error::InvalidRequest(
"chat completions extra_headers.x-trace must be a string, got number".to_string()
)
);
@ -374,7 +380,7 @@ async fn a_forwarded_header_the_signer_computes_declines_to_python() {
.await
.expect_err("{forwarded} should decline instead of being signed");
assert!(
matches!(error, CoreError::Unsupported(_)),
matches!(error, Error::Unsupported(_)),
"{forwarded} declined as {error:?}, which the host would not fall back on"
);
}
@ -727,7 +733,7 @@ mod round_trip {
.expect_err("response cannot be normalized");
handle.await.expect("server task");
assert!(
matches!(err, CoreError::InvalidResponse(_)),
matches!(err, Error::InvalidResponse(_)),
"expected a post-send error, got {err:?}"
);
}
@ -745,7 +751,7 @@ mod round_trip {
.expect_err("response cannot be normalized");
handle.await.expect("server task");
assert!(
matches!(err, CoreError::InvalidResponse(_)),
matches!(err, Error::InvalidResponse(_)),
"expected a post-send error, got {err:?}"
);
}
@ -763,7 +769,7 @@ mod round_trip {
.expect_err("upstream rejects");
handle.await.expect("server task");
assert!(
matches!(err, CoreError::Http { status: 429, .. }),
matches!(err, Error::Http { status: 429, .. }),
"expected a 429, got {err:?}"
);
}
@ -787,7 +793,7 @@ mod round_trip {
.await
.expect_err("nothing is listening");
assert!(
matches!(err, CoreError::Connect(_)),
matches!(err, Error::Connect(_)),
"expected a pre-send connect failure, got {err:?}"
);
}
@ -797,24 +803,24 @@ mod round_trip {
use crate::chat_completions::handler::as_response_error;
for original in [
CoreError::MissingField("usage"),
CoreError::Unsupported("non-text response content block"),
CoreError::InvalidRequest("whatever".to_string()),
CoreError::Auth("whatever".to_string()),
Error::MissingField("usage"),
Error::Unsupported("non-text response content block"),
Error::InvalidRequest("whatever".to_string()),
Error::Auth("whatever".to_string()),
] {
let label = format!("{original:?}");
assert!(
matches!(as_response_error(original), CoreError::InvalidResponse(_)),
matches!(as_response_error(original), Error::InvalidResponse(_)),
"{label} must not stay retryable once the provider has answered"
);
}
// An upstream status is already unambiguous, so it survives intact.
assert!(matches!(
as_response_error(CoreError::Http {
as_response_error(Error::Http {
status: 500,
body: "boom".to_string()
}),
CoreError::Http { status: 500, .. }
Error::Http { status: 500, .. }
));
}
}

24
litellm-rust/crates/core/src/chat_completions/transformation.rs
View file

@ -1,7 +1,6 @@
use crate::Error;
use serde_json::{Map, Value};
use crate::error::CoreResult;
use super::types::{
ChatCompletionsResponse, ChatMessage, ChatMessageContent, ProviderChatRequestData,
ProviderChatResponseData,
@ -39,7 +38,7 @@ pub trait ChatCompletionsProviderConfig: Sync {
model: &str,
optional_params: &Map<String, Value>,
env_lookup: &dyn Fn(&str) -> Option<String>,
) -> CoreResult<String>;
) -> Result<String, Error>;
fn auth(
&self,
@ -47,7 +46,7 @@ pub trait ChatCompletionsProviderConfig: Sync {
model: &str,
optional_params: &Map<String, Value>,
env_lookup: &dyn Fn(&str) -> Option<String>,
) -> CoreResult<ChatCompletionsAuth>;
) -> Result<ChatCompletionsAuth, Error>;
fn default_headers(&self) -> &'static [(&'static str, &'static str)] {
&[("content-type", "application/json")]
@ -63,9 +62,8 @@ pub trait ChatCompletionsProviderConfig: Sync {
false
}
/// Provider parameter names (post-mapping) the Rust path knows how to place
/// in the upstream body. Anything outside this set declines the request.
fn supported_params(&self) -> &'static [&'static str];
/// Supported OpenAI parameter names paired with their provider names.
fn supported_openai_params(&self) -> &'static [(&'static str, &'static str)];
/// Parameters consumed as call configuration (credentials, endpoints)
/// rather than placed in the body. Accepted, never serialized.
@ -79,7 +77,7 @@ pub trait ChatCompletionsProviderConfig: Sync {
optional_params: &Map<String, Value>,
) -> Option<Unsupported> {
unsupported_param(
self.supported_params(),
self.supported_openai_params(),
self.config_params(),
optional_params,
)
@ -91,17 +89,17 @@ pub trait ChatCompletionsProviderConfig: Sync {
model: &str,
messages: Vec<ChatMessage>,
optional_params: Map<String, Value>,
) -> CoreResult<ProviderChatRequestData>;
) -> Result<ProviderChatRequestData, Error>;
fn transform_response(
&self,
model: &str,
response: ProviderChatResponseData,
) -> CoreResult<ChatCompletionsResponse>;
) -> Result<ChatCompletionsResponse, Error>;
}
pub fn unsupported_param(
supported: &'static [&'static str],
supported: &'static [(&'static str, &'static str)],
config: &'static [&'static str],
optional_params: &Map<String, Value>,
) -> Option<Unsupported> {
@ -116,7 +114,9 @@ pub fn unsupported_param(
.keys()
.any(|key| {
key != STREAM_PARAM
&& !supported.contains(&key.as_str())
&& !supported
.iter()
.any(|(_, provider_name)| *provider_name == key)
&& !config.contains(&key.as_str())
})
.then_some(Unsupported("unrecognized request parameter"))

11
litellm-rust/crates/core/src/chat_completions/types.rs
View file

@ -22,6 +22,17 @@ pub struct ChatCompletionsRequest<'a> {
pub timeout: Option<Duration>,
}
pub(super) struct ResolvedChatCompletionsRequest<'a> {
pub(super) model: String,
pub(super) config: &'static dyn ChatCompletionsProviderConfig,
pub(super) messages: Vec<ChatMessage>,
pub(super) optional_params: Map<String, Value>,
pub(super) api_key: Option<&'a str>,
pub(super) api_base: Option<&'a str>,
pub(super) extra_headers: Option<Map<String, Value>>,
pub(super) timeout: Option<Duration>,
}
pub(super) struct ProviderChatCompletionsRequest {
pub(super) model: String,
pub(super) config: &'static dyn ChatCompletionsProviderConfig,

2
litellm-rust/crates/core/src/constants.rs
View file

@ -30,6 +30,8 @@ pub(crate) const CHAT_COMPLETIONS_TIMEOUT_SECS: u64 = 600;
/// Connect timeout for chat completions provider calls, in seconds.
pub(crate) const CHAT_COMPLETIONS_CONNECT_TIMEOUT_SECS: u64 = 10;
pub(crate) const AUDIO_TRANSCRIPTION_TIMEOUT_SECS: u64 = 600;
/// `object` field every non-streaming chat completion response carries.
pub const CHAT_COMPLETION_OBJECT: &str = "chat.completion";

8
litellm-rust/crates/core/src/error.rs
View file

@ -1,9 +1,7 @@
use thiserror::Error;
use thiserror::Error as ThisError;
pub type CoreResult<T> = Result<T, CoreError>;
#[derive(Debug, Error, PartialEq, Eq)]
pub enum CoreError {
#[derive(Debug, ThisError, PartialEq, Eq)]
pub enum Error {
#[error("expected {expected}, got {actual}")]
InvalidType {
expected: &'static str,

15
litellm-rust/crates/core/src/http_utils.rs
View file

@ -3,7 +3,14 @@
use serde_json::{Map, Value};
use crate::constants::UPSTREAM_ERROR_BODY_MAX_CHARS;
use crate::error::{CoreError, CoreResult, json_type_name};
use crate::error::{Error, json_type_name};
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
pub async fn http_request(
request: reqwest::RequestBuilder,
) -> Result<reqwest::Response, reqwest::Error> {
request.send().await
}
/// Bound an upstream error body before it crosses a host boundary, so provider
/// bodies stay data-minimized.
@ -18,7 +25,7 @@ pub fn truncate_error_body(body: &str) -> String {
pub fn string_headers(
context: &'static str,
extra_headers: Option<Map<String, Value>>,
) -> CoreResult<Vec<(String, String)>> {
) -> Result<Vec<(String, String)>, Error> {
extra_headers
.unwrap_or_default()
.into_iter()
@ -27,7 +34,7 @@ pub fn string_headers(
.as_str()
.map(|value| (key.clone(), value.to_string()))
.ok_or_else(|| {
CoreError::InvalidRequest(format!(
Error::InvalidRequest(format!(
"{context} extra_headers.{key} must be a string, got {}",
json_type_name(&value)
))
@ -81,7 +88,7 @@ mod tests {
let err = string_headers("chat completions", Some(headers)).expect_err("non-string value");
assert_eq!(
err,
CoreError::InvalidRequest(
Error::InvalidRequest(
"chat completions extra_headers.x-trace must be a string, got number".to_string()
)
);

2
litellm-rust/crates/core/src/lib.rs
View file

@ -13,4 +13,4 @@ pub mod responses;
pub mod router;
pub mod routing_utils;
pub use error::{CoreError, CoreResult};
pub use error::Error;

8
litellm-rust/crates/core/src/messages/common_utils.rs
View file

@ -1,9 +1,8 @@
use serde_json::{Map, Value};
use crate::error::CoreResult;
use crate::Error;
use crate::http_utils::string_headers as shared_string_headers;
use crate::providers::anthropic::messages::transformation::ANTHROPIC_MESSAGES_CONFIG;
use crate::providers::azure_ai::messages::transformation::AZURE_ANTHROPIC_MESSAGES_CONFIG;
use serde_json::{Map, Value};
use super::transformation::AnthropicMessagesProviderConfig;
@ -11,6 +10,7 @@ pub(super) use crate::http_utils::{has_bearer_auth, has_header, truncate_error_b
const HEADER_CONTEXT: &str = "messages";
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
pub(super) fn messages_provider_config(
provider: &str,
) -> Option<&'static dyn AnthropicMessagesProviderConfig> {
@ -23,6 +23,6 @@ pub(super) fn messages_provider_config(
pub(super) fn string_headers(
extra_headers: Option<Map<String, Value>>,
) -> CoreResult<Vec<(String, String)>> {
) -> Result<Vec<(String, String)>, Error> {
shared_string_headers(HEADER_CONTEXT, extra_headers)
}

42
litellm-rust/crates/core/src/messages/handler.rs
View file

@ -1,13 +1,17 @@
use crate::constants::ANTHROPIC_MESSAGES_PROVIDER;
use crate::error::{CoreError, CoreResult};
use crate::error::Error;
use crate::http_utils::http_request;
use super::client::http_client;
use super::common_utils::truncate_error_body;
use super::types::{AnthropicMessagesResponse, ProviderMessagesRequest};
use super::prepare::prepare_provider_request;
use super::types::{AnthropicMessagesResponse, MessagesRequest};
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
pub(super) async fn execute_messages_provider_call(
request: ProviderMessagesRequest,
) -> CoreResult<AnthropicMessagesResponse> {
request: MessagesRequest<'_>,
) -> Result<AnthropicMessagesResponse, Error> {
let request = prepare_provider_request(request)?;
let mut request_builder = http_client().post(&request.url).json(&request.body);
for (key, value) in &request.upstream_headers {
request_builder = request_builder.header(key, value);
@ -16,35 +20,34 @@ pub(super) async fn execute_messages_provider_call(
request_builder = request_builder.timeout(duration);
}
let response = request_builder
.send()
let response = http_request(request_builder)
.await
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
let status = response.status();
let text = response
.text()
.await
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
if !status.is_success() {
return Err(CoreError::Http {
return Err(Error::Http {
status: status.as_u16(),
body: truncate_error_body(&text),
});
}
let response = serde_json::from_str(&text).map_err(|err| {
CoreError::InvalidResponse(format!("invalid messages response JSON: {err}"))
})?;
let response = serde_json::from_str(&text)
.map_err(|err| Error::InvalidResponse(format!("invalid messages response JSON: {err}")))?;
request.config.transform_response(&request.model, response)
}
pub(super) async fn execute_messages_provider_stream(
request: ProviderMessagesRequest,
) -> CoreResult<reqwest::Response> {
request: MessagesRequest<'_>,
) -> Result<reqwest::Response, Error> {
let request = prepare_provider_request(request)?;
if request.provider != ANTHROPIC_MESSAGES_PROVIDER {
return Err(CoreError::InvalidRequest(
return Err(Error::InvalidRequest(
"streaming messages is not supported for this provider".to_string(),
));
}
@ -57,17 +60,16 @@ pub(super) async fn execute_messages_provider_stream(
request_builder = request_builder.timeout(duration);
}
let response = request_builder
.send()
let response = http_request(request_builder)
.await
.map_err(|err| CoreError::Network(err.to_string()))?;
.map_err(|err| Error::Network(err.to_string()))?;
let status = response.status();
if !status.is_success() {
let text = response
.text()
.await
.map_err(|err| CoreError::Network(err.to_string()))?;
return Err(CoreError::Http {
.map_err(|err| Error::Network(err.to_string()))?;
return Err(Error::Http {
status: status.as_u16(),
body: truncate_error_body(&text),
});

13
litellm-rust/crates/core/src/messages/mod.rs
View file

@ -7,6 +7,7 @@
//! is the streaming variant; it hands the raw upstream response back so a host
//! can splice the event stream to its own caller.
use crate::Error;
mod client;
mod common_utils;
mod handler;
@ -14,18 +15,16 @@ mod prepare;
pub mod transformation;
pub mod types;
use crate::error::CoreResult;
use handler::{execute_messages_provider_call, execute_messages_provider_stream};
use prepare::prepare_messages_call;
use types::{AnthropicMessagesResponse, MessagesRequest};
pub async fn messages(request: MessagesRequest<'_>) -> CoreResult<AnthropicMessagesResponse> {
execute_messages_provider_call(prepare_messages_call(request)?).await
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
pub async fn messages(request: MessagesRequest<'_>) -> Result<AnthropicMessagesResponse, Error> {
execute_messages_provider_call(request).await
}
pub async fn messages_stream(request: MessagesRequest<'_>) -> CoreResult<reqwest::Response> {
execute_messages_provider_stream(prepare_messages_call(request)?).await
pub async fn messages_stream(request: MessagesRequest<'_>) -> Result<reqwest::Response, Error> {
execute_messages_provider_stream(request).await
}
#[cfg(test)]

71
litellm-rust/crates/core/src/messages/prepare.rs
View file

@ -1,13 +1,14 @@
use crate::error::{CoreError, CoreResult};
use crate::error::Error;
use crate::routing_utils::provider::{CustomLlmProvider, get_custom_llm_provider};
use super::common_utils::{has_bearer_auth, has_header, messages_provider_config, string_headers};
use super::transformation::MessagesAuthStrategy;
use super::transformation::{AnthropicMessagesProviderConfig, MessagesAuthStrategy};
use super::types::{MessagesRequest, ProviderMessagesRequest};
use serde_json::{Map, Value};
pub(super) fn prepare_messages_call(
pub(super) fn prepare_provider_request(
request: MessagesRequest<'_>,
) -> CoreResult<ProviderMessagesRequest> {
) -> Result<ProviderMessagesRequest, Error> {
let provider_info = get_custom_llm_provider(request.model, request.custom_llm_provider)
.or_else(|| {
request
@ -18,7 +19,7 @@ pub(super) fn prepare_messages_call(
})
})
.ok_or_else(|| {
CoreError::InvalidProvider(
Error::InvalidProvider(
"unable to resolve custom_llm_provider for messages request".to_string(),
)
})?;
@ -26,16 +27,49 @@ pub(super) fn prepare_messages_call(
let provider = provider_info.custom_llm_provider;
let config = messages_provider_config(provider)
.ok_or_else(|| CoreError::InvalidProvider(provider.to_string()))?;
.ok_or_else(|| Error::InvalidProvider(provider.to_string()))?;
let env_lookup = |key: &str| std::env::var(key).ok();
let mut headers = string_headers(request.extra_headers)?;
let headers =
validate_environment(config, request.extra_headers, request.api_key, &env_lookup)?;
let typed_request = serde_json::from_value(request.body).map_err(|err| {
Error::InvalidRequest(format!("invalid Anthropic messages request: {err}"))
})?;
let transformed = config.transform_request(typed_request)?;
let body = serde_json::to_value(transformed).map_err(|err| {
Error::InvalidRequest(format!(
"failed to serialize Anthropic messages request: {err}"
))
})?;
let url = config.complete_url(request.api_base, &model, &env_lookup)?;
Ok(ProviderMessagesRequest {
provider: provider.to_string(),
model,
config,
url,
body,
upstream_headers: headers,
timeout: request.timeout,
})
}
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
fn validate_environment(
config: &dyn AnthropicMessagesProviderConfig,
extra_headers: Option<Map<String, Value>>,
api_key: Option<&str>,
env_lookup: &dyn Fn(&str) -> Option<String>,
) -> Result<Vec<(String, String)>, Error> {
let mut headers = string_headers(extra_headers)?;
let auth_strategy = config.auth_strategy();
let already_authorized = has_header(&headers, auth_strategy.header_name())
|| (config.accepts_bearer_auth() && has_bearer_auth(&headers));
if !already_authorized {
let api_key = config.resolve_api_key(request.api_key, &env_lookup)?;
let api_key = config.resolve_api_key(api_key, env_lookup)?;
let auth_header = match auth_strategy {
MessagesAuthStrategy::Bearer => {
("authorization".to_string(), format!("Bearer {api_key}"))
@ -51,24 +85,5 @@ pub(super) fn prepare_messages_call(
}
}
let url = config.complete_url(request.api_base, &model, &env_lookup)?;
let typed_request = serde_json::from_value(request.body).map_err(|err| {
CoreError::InvalidRequest(format!("invalid Anthropic messages request: {err}"))
})?;
let transformed = config.transform_request(typed_request)?;
let body = serde_json::to_value(transformed).map_err(|err| {
CoreError::InvalidRequest(format!(
"failed to serialize Anthropic messages request: {err}"
))
})?;
Ok(ProviderMessagesRequest {
provider: provider.to_string(),
model,
config,
url,
body,
upstream_headers: headers,
timeout: request.timeout,
})
Ok(headers)
}

10
litellm-rust/crates/core/src/messages/tests.rs
View file

@ -4,7 +4,7 @@ use serde_json::{Map, Value, json};
use tokio::io::{AsyncReadExt, AsyncWriteExt};
use tokio::net::{TcpListener, TcpStream};
use crate::error::CoreError;
use crate::error::Error;
use super::common_utils::{
has_bearer_auth, has_header, messages_provider_config, string_headers, truncate_error_body,
@ -77,7 +77,7 @@ fn truncate_error_body_caps_long_payloads() {
fn string_headers_rejects_non_string_values() {
let headers = json!({"x-count": 3}).as_object().unwrap().clone();
let err = string_headers(Some(headers)).expect_err("non-string header rejected");
assert!(matches!(err, CoreError::InvalidRequest(_)));
assert!(matches!(err, Error::InvalidRequest(_)));
}
#[test]
@ -341,7 +341,7 @@ async fn messages_requires_auth_when_no_key_and_no_header() {
.await
.expect_err("missing auth errors");
assert!(matches!(err, CoreError::Auth(_)));
assert!(matches!(err, Error::Auth(_)));
}
#[tokio::test]
@ -420,7 +420,7 @@ async fn messages_maps_provider_error_status_to_http_error() {
.await
.expect_err("provider error propagates");
assert!(matches!(err, CoreError::Http { status: 401, .. }));
assert!(matches!(err, Error::Http { status: 401, .. }));
}
#[tokio::test]
@ -437,5 +437,5 @@ async fn messages_rejects_unsupported_provider() {
.await
.expect_err("unsupported provider errors");
assert!(matches!(err, CoreError::InvalidProvider(provider) if provider == "openai"));
assert!(matches!(err, Error::InvalidProvider(provider) if provider == "openai"));
}

13
litellm-rust/crates/core/src/messages/transformation.rs
View file

@ -1,6 +1,5 @@
use crate::error::CoreResult;
use super::types::{AnthropicMessagesRequest, AnthropicMessagesResponse};
use crate::Error;
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub enum MessagesAuthStrategy {
@ -23,13 +22,13 @@ pub trait AnthropicMessagesProviderConfig: Sync {
api_base: Option<&str>,
model: &str,
env_lookup: &dyn Fn(&str) -> Option<String>,
) -> CoreResult<String>;
) -> Result<String, Error>;
fn resolve_api_key(
&self,
api_key: Option<&str>,
env_lookup: &dyn Fn(&str) -> Option<String>,
) -> CoreResult<String>;
) -> Result<String, Error>;
fn auth_strategy(&self) -> MessagesAuthStrategy {
MessagesAuthStrategy::Header("x-api-key")
@ -46,18 +45,20 @@ pub trait AnthropicMessagesProviderConfig: Sync {
]
}
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
fn transform_request(
&self,
request: AnthropicMessagesRequest,
) -> CoreResult<AnthropicMessagesRequest> {
) -> Result<AnthropicMessagesRequest, Error> {
Ok(request)
}
#[tracing::instrument(target = "litellm::function_trace", level = "trace", skip_all)]
fn transform_response(
&self,
_model: &str,
response: AnthropicMessagesResponse,
) -> CoreResult<AnthropicMessagesResponse> {
) -> Result<AnthropicMessagesResponse, Error> {
Ok(response)
}
}

Some files were not shown because too many files have changed in this diff Show more