mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-18 09:45:23 +00:00
refactor(proxy): meet the AGENTS.md rules Greptile flagged on the startup-message change
The policy listing builds its rows as tuples in two helpers instead of appending to a list; plain output is unchanged (same bytes as before across 3000 random policy and attachment configs). The explanatory comments added with this change are gone, and the docstring of print_startup_message now says the record is whole and unredacted. Tests: the migration-diff JSON test no longer runs the fake prisma CLI; it checks the skipped-diff notice in-process, and the fixture extension it needed is reverted. The json_logs export and deprecation-notice CLI tests run the real proxy module and a real config file instead of a mocked ProxyConfig and proxy module, and the --health/--test notice tests no longer patch our helpers. Each still fails when the export, the hold or call_on_close is removed. New test parameters are typed.
This commit is contained in:
parent
ea2ac4a136
commit
d796d095bf
11 changed files with 184 additions and 206 deletions
|
|
@ -908,7 +908,7 @@ def print_startup_message(message: str) -> None:
|
|||
stdout is read by a log pipeline, so the message goes out as one INFO record of the proxy logger
|
||||
instead, on stdout like the print whatever handlers the logging config gave that logger, with the
|
||||
ANSI color codes and surrounding blank lines removed, located at the caller and emitted at any log
|
||||
level; a blank spacer produces no record.
|
||||
level, whole and unredacted as the print showed it; a blank spacer produces no record.
|
||||
"""
|
||||
if not litellm.json_logs:
|
||||
print(message) # noqa: T201 # plain mode: this is the operator-terminal print it replaces
|
||||
|
|
@ -928,20 +928,8 @@ def print_startup_message(message: str) -> None:
|
|||
None,
|
||||
func=caller.f_code.co_name if caller is not None else None,
|
||||
)
|
||||
# The print this replaces shows at any log level, so the record skips the level checks of the
|
||||
# proxy logger (at the root's WARNING in the CLI process until initialize() runs) and of its JSON
|
||||
# handler (LITELLM_LOG). It also skips the logger's stdout cap: that cap bounds echoed request
|
||||
# payloads, while a startup message is operator text the print showed whole (the required migration
|
||||
# SQL, the policy list). It skips secret redaction too, which the print never had and whose sk-
|
||||
# pattern turns a model named risk-assessment-gpt4o into riREDACTED. The processed record counts as
|
||||
# scrubbed, so the handler's secret filter and the formatter leave it alone too.
|
||||
if not _process_record(record, base64_limit=_get_max_base64_length_stdout_log(), text_limit=0, redact=False):
|
||||
return
|
||||
# The print wrote to stdout whatever the logging config said, so the record goes through the proxy's
|
||||
# own JSON handler on stdout rather than the proxy logger's handlers: uvicorn applies an operator
|
||||
# --log_config after proxy_cli's turn_on_json() and before initialize() re-installs the JSON handler
|
||||
# through load_config(), and in that window, where the banner and the feedback box are emitted, a
|
||||
# config that names "LiteLLM Proxy" has left the logger with no handler or with the operator's own.
|
||||
_startup_json_handler().handle(record)
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -8,6 +8,7 @@ Configuration structure:
|
|||
|
||||
import sys
|
||||
from collections.abc import Mapping, Sequence
|
||||
from itertools import chain
|
||||
from typing import TYPE_CHECKING, Any, Final, Optional
|
||||
|
||||
from litellm._logging import print_startup_message, verbose_proxy_logger
|
||||
|
|
@ -26,6 +27,42 @@ _yellow_color_code: Final = "\033[93m"
|
|||
_reset_color_code: Final = "\033[0m"
|
||||
|
||||
|
||||
def _policy_lines(policy_name: str, policy_data: Mapping[str, object]) -> tuple[str, ...]:
|
||||
guardrails: Final = policy_data.get("guardrails", {})
|
||||
inherit: Final = policy_data.get("inherit")
|
||||
condition: Final = policy_data.get("condition")
|
||||
description: Final = policy_data.get("description")
|
||||
|
||||
guardrails_add: Final = guardrails.get("add", []) if isinstance(guardrails, dict) else []
|
||||
guardrails_remove: Final = guardrails.get("remove", []) if isinstance(guardrails, dict) else []
|
||||
model_condition: Final = condition.get("model") if isinstance(condition, dict) else None
|
||||
inherit_str: Final = f" (inherits: {inherit})" if inherit else ""
|
||||
|
||||
return (
|
||||
f"{_blue_color_code} - {policy_name}{inherit_str}{_reset_color_code}",
|
||||
*((f" description: {description}",) if description else ()),
|
||||
*((f" guardrails.add: {guardrails_add}",) if guardrails_add else ()),
|
||||
*((f" guardrails.remove: {guardrails_remove}",) if guardrails_remove else ()),
|
||||
*((f" condition.model: {model_condition}",) if model_condition else ()),
|
||||
)
|
||||
|
||||
|
||||
def _attachment_line(attachment: Mapping[str, object]) -> str:
|
||||
policy: Final = attachment.get("policy", "unknown")
|
||||
scope: Final = attachment.get("scope")
|
||||
teams: Final = attachment.get("teams")
|
||||
keys: Final = attachment.get("keys")
|
||||
models: Final = attachment.get("models")
|
||||
|
||||
scope_parts: Final = (
|
||||
*(("scope=* (global)",) if scope == "*" else ()),
|
||||
*((f"teams={teams}",) if teams else ()),
|
||||
*((f"keys={keys}",) if keys else ()),
|
||||
*((f"models={models}",) if models else ()),
|
||||
)
|
||||
return f" - {policy} -> {', '.join(scope_parts) if scope_parts else 'all'}"
|
||||
|
||||
|
||||
def _print_policies_on_startup(
|
||||
policies_config: Mapping[str, Mapping[str, object]],
|
||||
policy_attachments_config: Sequence[Mapping[str, object]] | None = None,
|
||||
|
|
@ -33,60 +70,29 @@ def _print_policies_on_startup(
|
|||
"""
|
||||
Print loaded policies to console on startup (similar to model list).
|
||||
"""
|
||||
lines: Final = [
|
||||
header: Final = (
|
||||
f"{_green_color_code}\nLiteLLM Policy Engine: Loaded {len(policies_config)} policies{_reset_color_code}\n"
|
||||
]
|
||||
|
||||
for policy_name, policy_data in policies_config.items():
|
||||
guardrails = policy_data.get("guardrails", {})
|
||||
inherit = policy_data.get("inherit")
|
||||
condition = policy_data.get("condition")
|
||||
description = policy_data.get("description")
|
||||
|
||||
guardrails_add = guardrails.get("add", []) if isinstance(guardrails, dict) else []
|
||||
guardrails_remove = guardrails.get("remove", []) if isinstance(guardrails, dict) else []
|
||||
inherit_str = f" (inherits: {inherit})" if inherit else ""
|
||||
|
||||
lines.append(f"{_blue_color_code} - {policy_name}{inherit_str}{_reset_color_code}")
|
||||
if description:
|
||||
lines.append(f" description: {description}")
|
||||
if guardrails_add:
|
||||
lines.append(f" guardrails.add: {guardrails_add}")
|
||||
if guardrails_remove:
|
||||
lines.append(f" guardrails.remove: {guardrails_remove}")
|
||||
if condition:
|
||||
model_condition = condition.get("model") if isinstance(condition, dict) else None
|
||||
if model_condition:
|
||||
lines.append(f" condition.model: {model_condition}")
|
||||
|
||||
# Print attachments
|
||||
if policy_attachments_config:
|
||||
lines.append(
|
||||
f"\n{_yellow_color_code}Policy Attachments: {len(policy_attachments_config)} attachment(s){_reset_color_code}"
|
||||
)
|
||||
policy_lines: Final = tuple(
|
||||
chain.from_iterable(
|
||||
_policy_lines(policy_name, policy_data) for policy_name, policy_data in policies_config.items()
|
||||
)
|
||||
for attachment in policy_attachments_config:
|
||||
policy = attachment.get("policy", "unknown")
|
||||
scope = attachment.get("scope")
|
||||
teams = attachment.get("teams")
|
||||
keys = attachment.get("keys")
|
||||
models = attachment.get("models")
|
||||
)
|
||||
|
||||
scope_parts = []
|
||||
if scope == "*":
|
||||
scope_parts.append("scope=* (global)")
|
||||
if teams:
|
||||
scope_parts.append(f"teams={teams}")
|
||||
if keys:
|
||||
scope_parts.append(f"keys={keys}")
|
||||
if models:
|
||||
scope_parts.append(f"models={models}")
|
||||
scope_str = ", ".join(scope_parts) if scope_parts else "all"
|
||||
|
||||
lines.append(f" - {policy} -> {scope_str}")
|
||||
lines.append("")
|
||||
print_startup_message("\n".join(lines))
|
||||
if policy_attachments_config:
|
||||
print_startup_message(
|
||||
"\n".join(
|
||||
(
|
||||
header,
|
||||
*policy_lines,
|
||||
f"\n{_yellow_color_code}Policy Attachments: {len(policy_attachments_config)} attachment(s){_reset_color_code}",
|
||||
*(_attachment_line(attachment) for attachment in policy_attachments_config),
|
||||
"",
|
||||
)
|
||||
)
|
||||
)
|
||||
else:
|
||||
print_startup_message("\n".join(lines))
|
||||
print_startup_message("\n".join((header, *policy_lines)))
|
||||
print_startup_message(
|
||||
f"\n{_yellow_color_code}Warning: No policy_attachments configured. Policies will not be applied to any requests.{_reset_color_code}\n",
|
||||
)
|
||||
|
|
|
|||
|
|
@ -1323,9 +1323,6 @@ def run_server(
|
|||
import litellm
|
||||
|
||||
litellm.json_logs = True
|
||||
# Workers that uvicorn or granian spawn import litellm afresh and read this at import,
|
||||
# so they log JSON from their first record instead of plain text until load_config
|
||||
# reaches the json_logs key.
|
||||
os.environ["JSON_LOGS"] = "true"
|
||||
|
||||
litellm.turn_on_json()
|
||||
|
|
|
|||
|
|
@ -9627,8 +9627,6 @@ async def initialize(
|
|||
prisma_client
|
||||
from litellm.proxy.common_utils.banner import LITELLM_BANNER, show_banner
|
||||
|
||||
# show_banner() also serves the `lite` shell, which runs against an uninitialized SDK (#46803), so
|
||||
# the JSON-mode choice stays out of it.
|
||||
if litellm.json_logs:
|
||||
print_startup_message(LITELLM_BANNER)
|
||||
else:
|
||||
|
|
|
|||
|
|
@ -31,15 +31,10 @@ from tests.integration._support.process import owned_proxy_process, refused_boot
|
|||
|
||||
JSON_RECORD: Final = TypeAdapter(dict[str, JsonValue])
|
||||
PLAIN_LOG_LINE: Final = re.compile(r"^\d\d:\d\d:\d\d - LiteLLM")
|
||||
# prisma-client-py starts the Prisma query engine on the proxy's own stdout, and the engine announces
|
||||
# its build hash there in two writes (the text, then the newline), so with two workers the two
|
||||
# announcements land glued together or glued to a neighbor's line. The text is not litellm's and is
|
||||
# removed before the lines are read.
|
||||
PRISMA_ENGINE_TEXT: Final = re.compile(r"query-engine [0-9a-f]{40}")
|
||||
STARTED_WORKER: Final = "Started server process"
|
||||
MODEL_LIST_HEADER: Final = "Proxy initialized with Config, Set models:"
|
||||
CACHE_LINE: Final = "Setting Cache on Proxy"
|
||||
# printed by the CLI process before any worker exists, so it proves the launcher itself logs JSON
|
||||
CONNECTION_BUDGET_LINE: Final = "Postgres connection budget"
|
||||
REDIS_AUTH_CACHE_LINE: Final = "enable_redis_auth_cache"
|
||||
BANNER_FRAGMENT: Final = "██╗"
|
||||
|
|
@ -48,8 +43,6 @@ WORKER_COUNTS: Final = (1, 2)
|
|||
UNKNOWN_MODEL_STATUS: Final = 400
|
||||
BURST_SIZE: Final = 16
|
||||
CACHE_SETTINGS: Final[Mapping[str, JsonValue]] = {"cache": True, "cache_params": {"type": "redis"}}
|
||||
# the policy list and the required migration SQL are the startup listings that outgrow the stdout cap
|
||||
# (MAX_STRING_LENGTH_STDOUT_LOG, 4096 by default) that bounds echoed request payloads
|
||||
STDOUT_CAP: Final = 4096
|
||||
POLICY_COUNT: Final = 60
|
||||
POLICY_LIST_HEADER: Final = "LiteLLM Policy Engine: Loaded"
|
||||
|
|
@ -57,8 +50,6 @@ TRUNCATION_MARKER: Final = "litellm_truncated"
|
|||
DEPRECATED_FLAG: Final = "--enforce_prisma_migration_check"
|
||||
DEPRECATED_FLAG_NOTICE: Final = "--enforce_prisma_migration_check is deprecated and has no effect"
|
||||
UNPARSEABLE_CONFIG: Final = "model_list: [\n - bad: yaml: here\n"
|
||||
# an operator logging config of the usual shape: it sets the proxy logger's level and gives it no handler
|
||||
# of its own, and formats everything else, uvicorn included, as JSON through the root handler
|
||||
OPERATOR_LOG_CONFIG: Final[Mapping[str, JsonValue]] = {
|
||||
"version": 1,
|
||||
"disable_existing_loggers": False,
|
||||
|
|
@ -67,7 +58,6 @@ OPERATOR_LOG_CONFIG: Final[Mapping[str, JsonValue]] = {
|
|||
"loggers": {"LiteLLM Proxy": {"level": "INFO"}},
|
||||
"root": {"level": "INFO", "handlers": ["stdout"]},
|
||||
}
|
||||
# an operator logging config that silences the proxy logger with a handler of its own and keeps the rest JSON
|
||||
OPERATOR_SILENCING_LOG_CONFIG: Final[Mapping[str, JsonValue]] = {
|
||||
"version": 1,
|
||||
"disable_existing_loggers": False,
|
||||
|
|
@ -223,8 +213,6 @@ def test_json_logs_env_keeps_every_line_json(gateway: Gateway, tmp_path: Path, w
|
|||
@pytest.mark.parametrize("workers", WORKER_COUNTS)
|
||||
def test_json_logs_config_keeps_every_line_json(gateway: Gateway, tmp_path: Path, workers: int) -> None:
|
||||
models: Final = _models()
|
||||
# cache is listed before json_logs on purpose: the cache branch logs the Redis auth-cache line
|
||||
# while the config is still being read, so JSON mode must not depend on the key order.
|
||||
config: Final = _write_config(tmp_path, gateway.upstream_url, models, {**CACHE_SETTINGS, "json_logs": True})
|
||||
log: Final = _boot_serve_and_stop(gateway, tmp_path, config, {"LITELLM_LOG": "INFO"}, workers, models)
|
||||
_assert_every_line_is_json(log, workers, models)
|
||||
|
|
|
|||
|
|
@ -100,8 +100,6 @@ if not earlier_calls and os.environ.get("FAKE_PRISMA_HANG_FIRST"):
|
|||
grandchild = subprocess.Popen([sys.executable, "-c", "import time; time.sleep(600)"])
|
||||
pathlib.Path(os.environ["FAKE_PRISMA_GRANDCHILD_PIDFILE"]).write_text(str(grandchild.pid))
|
||||
time.sleep(600)
|
||||
if os.environ.get("FAKE_PRISMA_STDOUT"):
|
||||
sys.stdout.write(os.environ["FAKE_PRISMA_STDOUT"])
|
||||
sys.exit(0)
|
||||
"""
|
||||
|
||||
|
|
@ -113,8 +111,6 @@ class FakePrismaCli:
|
|||
With FAKE_PRISMA_HANG_FIRST set it hangs on its first call from a process tree
|
||||
of its own, the way the real CLI wraps Node around a Rust schema engine, so a
|
||||
timeout that kills only the direct child leaves the rest of that tree running.
|
||||
With FAKE_PRISMA_STDOUT set it writes that text to stdout, the way `migrate diff
|
||||
--script` prints the SQL that would close a schema gap.
|
||||
"""
|
||||
|
||||
calls_file: Path
|
||||
|
|
@ -146,7 +142,6 @@ def fake_prisma_cli(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> Generato
|
|||
monkeypatch.setenv("FAKE_PRISMA_GRANDCHILD_PIDFILE", str(cli.grandchild_pidfile))
|
||||
monkeypatch.setenv("LITELLM_PRISMA_COMMAND_TIMEOUT", "1")
|
||||
monkeypatch.delenv("FAKE_PRISMA_HANG_FIRST", raising=False)
|
||||
monkeypatch.delenv("FAKE_PRISMA_STDOUT", raising=False)
|
||||
yield cli
|
||||
if cli.grandchild_pidfile.exists():
|
||||
try:
|
||||
|
|
|
|||
|
|
@ -77,35 +77,31 @@ def test_migrate_diff_without_the_prisma_runner_skips_instead_of_crashing_boot(m
|
|||
assert check_prisma_schema_diff_helper("postgresql://u:p@localhost:9/x") == (False, [])
|
||||
|
||||
|
||||
def test_migrate_diff_announces_the_schema_changes_as_one_json_record(fake_prisma_cli, monkeypatch, capfd):
|
||||
def test_a_skipped_migration_diff_is_one_json_record_in_json_mode(
|
||||
monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""
|
||||
The "Changes to DB Schema detected" notice and its SQL went to stdout as plaintext;
|
||||
with json_logs on they are one INFO record, so a log pipeline keeps parsing. The
|
||||
loggers sit at WARNING, as they do at boot without LITELLM_LOG, and the record
|
||||
still goes out: the print it replaces had no level.
|
||||
With json_logs on, the notice that the migration diff was skipped is one INFO record on stdout,
|
||||
also while the loggers sit at WARNING, as they do at boot without LITELLM_LOG.
|
||||
"""
|
||||
from litellm.proxy.db.check_migration import check_prisma_schema_diff_helper
|
||||
|
||||
monkeypatch.setenv(
|
||||
"FAKE_PRISMA_STDOUT", '-- AlterTable\nALTER TABLE "LiteLLM_UserTable" ADD COLUMN "nickname" TEXT;\n'
|
||||
)
|
||||
monkeypatch.setitem(sys.modules, "litellm_proxy_extras.prisma_toolchain", None)
|
||||
monkeypatch.setattr(litellm, "json_logs", True)
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook) # turn_on_json swaps it; put it back at teardown
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook)
|
||||
turn_on_json()
|
||||
verbose_logger.setLevel(logging.WARNING)
|
||||
verbose_proxy_logger.setLevel(logging.WARNING)
|
||||
|
||||
assert check_prisma_schema_diff_helper("postgresql://u:p@localhost:9/x") == (
|
||||
True,
|
||||
['ALTER TABLE "LiteLLM_UserTable" ADD COLUMN "nickname" TEXT;'],
|
||||
)
|
||||
assert check_prisma_schema_diff_helper("postgresql://u:p@localhost:9/x") == (False, [])
|
||||
|
||||
out, err = capfd.readouterr()
|
||||
assert err == ""
|
||||
lines: Final = [raw for raw in out.splitlines() if raw.strip()]
|
||||
assert len(lines) == 1, lines
|
||||
record: Final = json.loads(lines[0])
|
||||
assert record["level"] == "INFO"
|
||||
assert record["message"] == (
|
||||
'Changes to DB Schema detected\nRequired SQL commands:\nALTER TABLE "LiteLLM_UserTable" ADD COLUMN "nickname" TEXT;'
|
||||
assert (record["level"], record["message"]) == (
|
||||
"INFO",
|
||||
"Skipping the migration diff: litellm-proxy-extras has no Prisma runner. "
|
||||
"Error: import of litellm_proxy_extras.prisma_toolchain halted; None in sys.modules",
|
||||
)
|
||||
|
|
|
|||
|
|
@ -53,7 +53,12 @@ NO_ATTACHMENTS_WARNING: Final = (
|
|||
],
|
||||
ids=["with-attachments", "without-attachments"],
|
||||
)
|
||||
def test_policy_table_prints_unchanged_in_plain_mode(monkeypatch, capfd, attachments, expected):
|
||||
def test_policy_table_prints_unchanged_in_plain_mode(
|
||||
monkeypatch: pytest.MonkeyPatch,
|
||||
capfd: pytest.CaptureFixture[str],
|
||||
attachments: Sequence[Mapping[str, object]] | None,
|
||||
expected: str,
|
||||
) -> None:
|
||||
"""Operators without JSON logs keep the colored policy table byte for byte."""
|
||||
monkeypatch.setattr(litellm, "json_logs", False)
|
||||
|
||||
|
|
@ -62,10 +67,12 @@ def test_policy_table_prints_unchanged_in_plain_mode(monkeypatch, capfd, attachm
|
|||
assert capfd.readouterr() == (expected, "")
|
||||
|
||||
|
||||
def test_policy_table_is_one_json_record_in_json_mode(monkeypatch, capfd):
|
||||
def test_policy_table_is_one_json_record_in_json_mode(
|
||||
monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""With JSON logs on the whole table, attachments included, is one INFO record without color codes."""
|
||||
monkeypatch.setattr(litellm, "json_logs", True)
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook) # turn_on_json swaps it; put it back at teardown
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook)
|
||||
turn_on_json()
|
||||
|
||||
_print_policies_on_startup(POLICIES, ATTACHMENTS)
|
||||
|
|
@ -88,7 +95,9 @@ def test_policy_table_is_one_json_record_in_json_mode(monkeypatch, capfd):
|
|||
)
|
||||
|
||||
|
||||
def test_missing_attachments_warning_is_a_second_json_record(monkeypatch, capfd):
|
||||
def test_missing_attachments_warning_is_a_second_json_record(
|
||||
monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""Without attachments the table and the yellow warning are two INFO records on stdout, where the two
|
||||
prints put them."""
|
||||
monkeypatch.setattr(litellm, "json_logs", True)
|
||||
|
|
|
|||
|
|
@ -10,6 +10,7 @@ from unittest.mock import AsyncMock, MagicMock, patch
|
|||
|
||||
import click
|
||||
import fastapi
|
||||
import httpx
|
||||
import pytest
|
||||
|
||||
|
||||
|
|
@ -2878,20 +2879,24 @@ class TestRunServerDbSetup:
|
|||
mock_setup_database.assert_called_once_with(use_migrate=True, use_v2_resolver=True)
|
||||
assert ("--enforce_prisma_migration_check is deprecated and has no effect" in capsys.readouterr().out) is warned
|
||||
|
||||
@pytest.mark.parametrize("argument", ("--health", "--test"), ids=("health", "test"))
|
||||
def test_the_retired_enforce_prisma_migration_check_opt_in_warns_on_health_and_test(self, argument, capsys):
|
||||
"""`--health` and `--test` return before the config is read, where a server start emits the
|
||||
deprecation line; they still print it, as they did when the notice came before those returns."""
|
||||
from litellm.proxy.proxy_cli import ProxyInitializationHelpers, run_server
|
||||
def test_the_retired_enforce_prisma_migration_check_opt_in_warns_on_health(
|
||||
self, capsys: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""`--health` returns before the config is read, where a server start emits the deprecation line;
|
||||
it still prints it, as it did when the notice came before that return."""
|
||||
with patch("httpx.get", return_value=httpx.Response(200, json={})) as mock_get:
|
||||
run_server.main(["--health", "--enforce_prisma_migration_check"], standalone_mode=False)
|
||||
|
||||
with (
|
||||
patch.object(ProxyInitializationHelpers, "run_health_check") as mock_health_check,
|
||||
patch.object(ProxyInitializationHelpers, "run_test_chat_completion") as mock_test_chat,
|
||||
):
|
||||
run_server.main([argument, "--enforce_prisma_migration_check"], standalone_mode=False)
|
||||
mock_get.assert_called_once_with(url="http://0.0.0.0:4000/health")
|
||||
assert "--enforce_prisma_migration_check is deprecated and has no effect" in capsys.readouterr().out
|
||||
|
||||
def test_the_retired_enforce_prisma_migration_check_opt_in_warns_on_test(
|
||||
self, capsys: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""`--test` returns before the config is read too, and prints the deprecation line before its request."""
|
||||
with pytest.raises(ValueError, match="Invalid test value"):
|
||||
run_server.main(["--test", "--enforce_prisma_migration_check"], standalone_mode=False)
|
||||
|
||||
expected_calls = (1, 0) if argument == "--health" else (0, 1)
|
||||
assert (mock_health_check.call_count, mock_test_chat.call_count) == expected_calls
|
||||
assert "--enforce_prisma_migration_check is deprecated and has no effect" in capsys.readouterr().out
|
||||
|
||||
@patch("subprocess.run")
|
||||
|
|
@ -3956,43 +3961,35 @@ class TestValidateConfigFlag:
|
|||
mock_popen.assert_not_called()
|
||||
|
||||
|
||||
def _boot_from_config_file(config_path: Path, *flags: str) -> str | None:
|
||||
env: Final = {k: v for k, v in os.environ.items() if k not in ("DATABASE_URL", "DIRECT_URL", "JSON_LOGS")}
|
||||
with patch.dict(os.environ, env, clear=True):
|
||||
run_server.main(["--config", str(config_path), "--skip_server_startup", *flags], standalone_mode=False)
|
||||
return os.environ.get("JSON_LOGS")
|
||||
|
||||
|
||||
class TestJsonLogsConfigReachesSpawnedWorkers:
|
||||
"""`litellm_settings.json_logs: true` turns JSON on inside the CLI process, but the workers that
|
||||
uvicorn or granian spawn are fresh interpreters that pick plain or JSON from the JSON_LOGS
|
||||
environment variable at import. The CLI has to export it, or every worker logs plain text until
|
||||
its own load_config reaches the key."""
|
||||
|
||||
@staticmethod
|
||||
def _json_logs_env_after_boot(config_path: Path, litellm_settings: dict[str, object]) -> str | None:
|
||||
config_path.write_text(yaml.dump({"model_list": [], "litellm_settings": litellm_settings}))
|
||||
mock_proxy_config: Final = MagicMock()
|
||||
mock_proxy_config.return_value.get_config = AsyncMock(return_value=yaml.safe_load(config_path.read_text()))
|
||||
mock_proxy_module: Final = MagicMock(
|
||||
app=MagicMock(),
|
||||
ProxyConfig=mock_proxy_config,
|
||||
KeyManagementSettings=MagicMock(),
|
||||
save_worker_config=MagicMock(),
|
||||
)
|
||||
clean_env: Final = {k: v for k, v in os.environ.items() if k not in ("DATABASE_URL", "DIRECT_URL", "JSON_LOGS")}
|
||||
with (
|
||||
patch.dict(os.environ, clean_env, clear=True),
|
||||
patch.dict(
|
||||
"sys.modules",
|
||||
{"proxy_server": mock_proxy_module, "litellm.proxy.proxy_server": mock_proxy_module},
|
||||
),
|
||||
patch("atexit.register"),
|
||||
patch("litellm.turn_on_json"), # test-quality-ok: it swaps every logger handler in this pytest worker
|
||||
):
|
||||
run_server.main(["--config", str(config_path), "--local", "--skip_server_startup"], standalone_mode=False)
|
||||
return os.environ.get("JSON_LOGS")
|
||||
|
||||
def test_json_logs_true_exports_json_logs_to_spawned_workers(self, tmp_path: Path, monkeypatch) -> None:
|
||||
monkeypatch.setattr(litellm, "json_logs", False) # run_server flips it on; undone at teardown
|
||||
assert self._json_logs_env_after_boot(tmp_path / "config.yaml", {"json_logs": True}) == "true"
|
||||
|
||||
def test_plain_logs_config_leaves_json_logs_unset(self, tmp_path: Path, monkeypatch) -> None:
|
||||
def test_json_logs_true_exports_json_logs_to_spawned_workers(
|
||||
self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch
|
||||
) -> None:
|
||||
monkeypatch.setattr(litellm, "json_logs", False)
|
||||
assert self._json_logs_env_after_boot(tmp_path / "config.yaml", {}) is None
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook)
|
||||
config_path: Final = tmp_path / "config.yaml"
|
||||
config_path.write_text(yaml.dump({"model_list": [], "litellm_settings": {"json_logs": True}}))
|
||||
|
||||
assert _boot_from_config_file(config_path) == "true"
|
||||
|
||||
def test_plain_logs_config_leaves_json_logs_unset(self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
|
||||
monkeypatch.setattr(litellm, "json_logs", False)
|
||||
config_path: Final = tmp_path / "config.yaml"
|
||||
config_path.write_text(yaml.dump({"model_list": [], "litellm_settings": {}}))
|
||||
|
||||
assert _boot_from_config_file(config_path) is None
|
||||
|
||||
|
||||
class TestEnforcePrismaMigrationCheckNoticeAroundTheConfigRead:
|
||||
|
|
@ -4001,68 +3998,53 @@ class TestEnforcePrismaMigrationCheckNoticeAroundTheConfigRead:
|
|||
and a boot whose config read fails still prints it before the error, as it did before the wait."""
|
||||
|
||||
NOTICE: Final = "--enforce_prisma_migration_check is deprecated and has no effect"
|
||||
FLAG: Final = "--enforce_prisma_migration_check"
|
||||
|
||||
@staticmethod
|
||||
def _boot(config_path: Path, get_config: AsyncMock) -> None:
|
||||
mock_proxy_config: Final = MagicMock()
|
||||
mock_proxy_config.return_value.get_config = get_config
|
||||
mock_proxy_module: Final = MagicMock(
|
||||
app=MagicMock(),
|
||||
ProxyConfig=mock_proxy_config,
|
||||
KeyManagementSettings=MagicMock(),
|
||||
save_worker_config=MagicMock(),
|
||||
)
|
||||
clean_env: Final = {k: v for k, v in os.environ.items() if k not in ("DATABASE_URL", "DIRECT_URL", "JSON_LOGS")}
|
||||
with (
|
||||
patch.dict(os.environ, clean_env, clear=True),
|
||||
patch.dict(
|
||||
"sys.modules",
|
||||
{"proxy_server": mock_proxy_module, "litellm.proxy.proxy_server": mock_proxy_module},
|
||||
),
|
||||
patch("atexit.register"),
|
||||
patch("litellm.turn_on_json"), # test-quality-ok: it swaps every logger handler in this pytest worker
|
||||
):
|
||||
run_server.main(
|
||||
["--config", str(config_path), "--local", "--skip_server_startup", "--enforce_prisma_migration_check"],
|
||||
standalone_mode=False,
|
||||
)
|
||||
|
||||
def test_a_config_read_that_fails_still_prints_the_notice(self, tmp_path: Path, monkeypatch, capsys) -> None:
|
||||
def test_a_config_read_that_fails_still_prints_the_notice(
|
||||
self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch, capsys: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
monkeypatch.setattr(litellm, "json_logs", False)
|
||||
missing: Final = tmp_path / "missing.yaml"
|
||||
|
||||
with pytest.raises(FileNotFoundError):
|
||||
self._boot(missing, AsyncMock(side_effect=FileNotFoundError(f"Config file not found: {missing}")))
|
||||
with pytest.raises(Exception, match="Config file not found"):
|
||||
_boot_from_config_file(tmp_path / "missing.yaml", self.FLAG)
|
||||
|
||||
assert capsys.readouterr().out.count(self.NOTICE) == 1
|
||||
|
||||
def test_a_plain_config_prints_the_notice_once(self, tmp_path: Path, monkeypatch, capsys) -> None:
|
||||
def test_a_plain_config_prints_the_notice_once(
|
||||
self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch, capsys: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
monkeypatch.setattr(litellm, "json_logs", False)
|
||||
config_path: Final = tmp_path / "config.yaml"
|
||||
config_path.write_text(yaml.dump({"model_list": [], "litellm_settings": {}}))
|
||||
|
||||
self._boot(tmp_path / "config.yaml", AsyncMock(return_value={"model_list": [], "litellm_settings": {}}))
|
||||
_boot_from_config_file(config_path, self.FLAG)
|
||||
|
||||
assert capsys.readouterr().out.count(self.NOTICE) == 1
|
||||
|
||||
def test_a_config_only_json_logs_makes_the_notice_one_json_record(
|
||||
self, tmp_path: Path, monkeypatch, capsys
|
||||
self, tmp_path: Path, monkeypatch: pytest.MonkeyPatch, capsys: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
monkeypatch.setattr(litellm, "json_logs", False) # run_server flips it on; undone at teardown
|
||||
config: Final = {"model_list": [], "litellm_settings": {"json_logs": True}}
|
||||
monkeypatch.setattr(litellm, "json_logs", False)
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook)
|
||||
config_path: Final = tmp_path / "config.yaml"
|
||||
config_path.write_text(yaml.dump({"model_list": [], "litellm_settings": {"json_logs": True}}))
|
||||
|
||||
self._boot(tmp_path / "config.yaml", AsyncMock(return_value=config))
|
||||
_boot_from_config_file(config_path, self.FLAG)
|
||||
|
||||
notice_lines: Final = [line for line in capsys.readouterr().out.splitlines() if self.NOTICE in line]
|
||||
assert len(notice_lines) == 1, notice_lines
|
||||
assert self.NOTICE in json.loads(notice_lines[0])["message"]
|
||||
|
||||
|
||||
def test_granian_flag_notices_are_json_records_in_json_mode(monkeypatch, capfd):
|
||||
def test_granian_flag_notices_are_json_records_in_json_mode(
|
||||
monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""The yellow notices for the flags Granian ignores (`--max_requests_before_restart`, `--ciphers`) are
|
||||
printed while the launcher starts; with json_logs on they are INFO records on stdout like every other
|
||||
line, not plaintext between the records."""
|
||||
pytest.importorskip("granian")
|
||||
monkeypatch.setattr(litellm, "json_logs", True)
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook) # turn_on_json swaps it; put it back at teardown
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook)
|
||||
turn_on_json()
|
||||
|
||||
with (
|
||||
|
|
|
|||
|
|
@ -15588,11 +15588,13 @@ async def test_spend_capture_rate_check_job_clears_the_gauge_once_the_setting_is
|
|||
]
|
||||
|
||||
|
||||
def test_feedback_box_is_one_json_record_in_json_mode(monkeypatch, capfd):
|
||||
def test_feedback_box_is_one_json_record_in_json_mode(
|
||||
monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""The feedback box is a dozen terminal lines; with JSON logs on it has to land as one parseable
|
||||
record, not a plaintext line per row."""
|
||||
monkeypatch.setattr(litellm, "json_logs", True)
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook) # turn_on_json swaps it; put it back at teardown
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook)
|
||||
turn_on_json()
|
||||
|
||||
proxy_server_module.generate_feedback_box()
|
||||
|
|
@ -15608,7 +15610,9 @@ def test_feedback_box_is_one_json_record_in_json_mode(monkeypatch, capfd):
|
|||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_initialize_prints_the_banner_between_blank_lines_in_plain_mode(monkeypatch, capfd):
|
||||
async def test_initialize_prints_the_banner_between_blank_lines_in_plain_mode(
|
||||
monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""Operators without JSON logs see the banner exactly as before, through show_banner()'s flushing
|
||||
click.echo."""
|
||||
monkeypatch.setattr(litellm, "json_logs", False)
|
||||
|
|
@ -15622,11 +15626,13 @@ async def test_initialize_prints_the_banner_between_blank_lines_in_plain_mode(mo
|
|||
|
||||
|
||||
@pytest.mark.asyncio
|
||||
async def test_initialize_prints_the_banner_as_one_json_record_in_json_mode(monkeypatch, capfd):
|
||||
async def test_initialize_prints_the_banner_as_one_json_record_in_json_mode(
|
||||
monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""With JSON logs on, the six banner rows are one INFO record with their indentation kept, so the
|
||||
first line a worker writes parses like every other."""
|
||||
monkeypatch.setattr(litellm, "json_logs", True)
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook) # turn_on_json swaps it; put it back at teardown
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook)
|
||||
monkeypatch.setenv("LITELLM_DONT_SHOW_FEEDBACK_BOX", "true")
|
||||
turn_on_json()
|
||||
|
||||
|
|
|
|||
|
|
@ -1827,7 +1827,9 @@ def test_print_verbose_formats_args_only_when_set_verbose_is_on(monkeypatch, cap
|
|||
assert capsys.readouterr().out == "key k1 value cached-vector\n"
|
||||
|
||||
|
||||
def test_print_startup_message_prints_plain_mode_text_unchanged(monkeypatch, capfd):
|
||||
def test_print_startup_message_prints_plain_mode_text_unchanged(
|
||||
monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""Operators without JSON logs keep the colored, spaced terminal output exactly as before."""
|
||||
monkeypatch.setattr(litellm, "json_logs", False)
|
||||
|
||||
|
|
@ -1838,12 +1840,14 @@ def test_print_startup_message_prints_plain_mode_text_unchanged(monkeypatch, cap
|
|||
assert err == ""
|
||||
|
||||
|
||||
def test_print_startup_message_is_one_json_record_in_json_mode(monkeypatch, capfd):
|
||||
def test_print_startup_message_is_one_json_record_in_json_mode(
|
||||
monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""With JSON logs on, a multi-line colored startup message is one INFO record with the colors and the
|
||||
surrounding blank lines removed, so a log pipeline parses it like every other line. The logger sits
|
||||
at WARNING, as it does in the CLI process, and the record still goes out: a print has no level."""
|
||||
monkeypatch.setattr(litellm, "json_logs", True)
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook) # turn_on_json swaps it; put it back at teardown
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook)
|
||||
turn_on_json()
|
||||
verbose_proxy_logger.setLevel(logging.WARNING)
|
||||
|
||||
|
|
@ -1859,11 +1863,12 @@ def test_print_startup_message_is_one_json_record_in_json_mode(monkeypatch, capf
|
|||
assert record["message"] == "LiteLLM: Proxy initialized with Config, Set models:\n gpt-4o"
|
||||
assert record["level"] == "INFO"
|
||||
assert record["component"] == "LiteLLM Proxy"
|
||||
# the record names the call site, not the helper, so operators can find the message
|
||||
assert str(record["logger"]).startswith("test_logging.py:"), record["logger"]
|
||||
|
||||
|
||||
def test_print_startup_message_yellow_notice_is_an_info_record_on_stdout(monkeypatch, capfd):
|
||||
def test_print_startup_message_yellow_notice_is_an_info_record_on_stdout(
|
||||
monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""An operator notice printed in yellow (an ignored flag, a deprecation) stays on stdout in JSON mode,
|
||||
where the print put it, as one INFO record with the color codes removed."""
|
||||
monkeypatch.setattr(litellm, "json_logs", True)
|
||||
|
|
@ -1882,7 +1887,9 @@ def test_print_startup_message_yellow_notice_is_an_info_record_on_stdout(monkeyp
|
|||
assert record["level"] == "INFO"
|
||||
|
||||
|
||||
def test_print_startup_message_shows_above_the_json_handler_level(monkeypatch, capfd):
|
||||
def test_print_startup_message_shows_above_the_json_handler_level(
|
||||
monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""`LITELLM_LOG=WARNING` sets the JSON handler to WARNING, which drops the INFO logger lines; the
|
||||
startup notices still go out, as the prints they replace did at every level."""
|
||||
monkeypatch.setattr(litellm, "json_logs", True)
|
||||
|
|
@ -1901,8 +1908,8 @@ def test_print_startup_message_shows_above_the_json_handler_level(monkeypatch, c
|
|||
|
||||
|
||||
def test_print_startup_message_emits_json_after_an_operator_log_config_removed_the_handlers(
|
||||
request, monkeypatch, capfd
|
||||
):
|
||||
request: pytest.FixtureRequest, monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""uvicorn applies an operator `--log_config` after proxy_cli's turn_on_json() and before initialize()
|
||||
re-installs the JSON handler through load_config(); a config that names "LiteLLM Proxy" leaves the
|
||||
logger without handlers in that window, where the banner and the feedback box are emitted, and they
|
||||
|
|
@ -1912,7 +1919,7 @@ def test_print_startup_message_emits_json_after_an_operator_log_config_removed_t
|
|||
monkeypatch.setattr(litellm, "json_logs", True)
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook)
|
||||
turn_on_json()
|
||||
request.addfinalizer(turn_on_json) # put the JSON handler back for the tests that follow
|
||||
request.addfinalizer(turn_on_json)
|
||||
logging.config.dictConfig(
|
||||
{"version": 1, "disable_existing_loggers": False, "loggers": {"LiteLLM Proxy": {"level": "INFO"}}}
|
||||
)
|
||||
|
|
@ -1943,8 +1950,8 @@ OPERATOR_HANDLERS_ON_THE_PROXY_LOGGER: Final = {
|
|||
|
||||
@pytest.mark.parametrize("handler", sorted(OPERATOR_HANDLERS_ON_THE_PROXY_LOGGER))
|
||||
def test_print_startup_message_stays_on_stdout_under_an_operator_handler_on_the_proxy_logger(
|
||||
request, monkeypatch, capfd, handler: str
|
||||
):
|
||||
request: pytest.FixtureRequest, monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str], handler: str
|
||||
) -> None:
|
||||
"""An operator `--log_config` that gives "LiteLLM Proxy" a handler of its own, a NullHandler that silences
|
||||
it or a plain WARNING-level handler on stderr, is in force when the banner and the feedback box are
|
||||
emitted; the print they replace wrote to stdout whatever the logging config said, so the record still
|
||||
|
|
@ -1954,7 +1961,7 @@ def test_print_startup_message_stays_on_stdout_under_an_operator_handler_on_the_
|
|||
monkeypatch.setattr(litellm, "json_logs", True)
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook)
|
||||
turn_on_json()
|
||||
request.addfinalizer(turn_on_json) # put the JSON handler back for the tests that follow
|
||||
request.addfinalizer(turn_on_json)
|
||||
logging.config.dictConfig(
|
||||
{
|
||||
"version": 1,
|
||||
|
|
@ -1980,13 +1987,15 @@ def test_print_startup_message_stays_on_stdout_under_an_operator_handler_on_the_
|
|||
assert record["component"] == "LiteLLM Proxy"
|
||||
|
||||
|
||||
def test_print_startup_message_keeps_a_listing_longer_than_the_stdout_cap_whole(monkeypatch, capfd):
|
||||
def test_print_startup_message_keeps_a_listing_longer_than_the_stdout_cap_whole(
|
||||
monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""The required migration SQL and the policy list run past MAX_STRING_LENGTH_STDOUT_LOG on a drifted
|
||||
deployment. The prints they replace showed them whole, and that cap exists for echoed request payloads,
|
||||
so the JSON record carries the whole listing too."""
|
||||
monkeypatch.setenv("MAX_STRING_LENGTH_STDOUT_LOG", "500")
|
||||
monkeypatch.setattr(litellm, "json_logs", True)
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook) # turn_on_json swaps it; put it back at teardown
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook)
|
||||
turn_on_json()
|
||||
commands: Final = tuple(f'ALTER TABLE "LiteLLM_Table{n}" ADD COLUMN "column_{n}" TEXT;' for n in range(40))
|
||||
listing: Final = "\n".join(("Changes to DB Schema detected", "Required SQL commands:", *commands))
|
||||
|
|
@ -2003,11 +2012,13 @@ def test_print_startup_message_keeps_a_listing_longer_than_the_stdout_cap_whole(
|
|||
assert record["message"] == listing
|
||||
|
||||
|
||||
def test_print_startup_message_keeps_a_model_name_the_secret_patterns_would_match(monkeypatch, capfd):
|
||||
def test_print_startup_message_keeps_a_model_name_the_secret_patterns_would_match(
|
||||
monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""The model list printed `risk-assessment-gpt4o` as configured; the sk- key pattern matches inside
|
||||
that name, so the JSON record has to carry the name as the print did, not `riREDACTED`."""
|
||||
monkeypatch.setattr(litellm, "json_logs", True)
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook) # turn_on_json swaps it; put it back at teardown
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook)
|
||||
turn_on_json()
|
||||
assert secret_redaction.redact_string("risk-assessment-gpt4o") != "risk-assessment-gpt4o"
|
||||
|
||||
|
|
@ -2020,7 +2031,9 @@ def test_print_startup_message_keeps_a_model_name_the_secret_patterns_would_matc
|
|||
assert json.loads(lines[0])["message"] == " risk-assessment-gpt4o"
|
||||
|
||||
|
||||
def test_print_startup_message_blank_spacer_emits_no_json_record(monkeypatch, capfd):
|
||||
def test_print_startup_message_blank_spacer_emits_no_json_record(
|
||||
monkeypatch: pytest.MonkeyPatch, capfd: pytest.CaptureFixture[str]
|
||||
) -> None:
|
||||
"""A spacer that only pads the terminal has nothing to say to a log pipeline."""
|
||||
monkeypatch.setattr(litellm, "json_logs", True)
|
||||
monkeypatch.setattr(sys, "excepthook", sys.excepthook)
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue