Action item from the v1.100.0 OOM RCA. LIT-6780 recorded "running without
--detailed_debug was not tried" as unverified; the fix PR closing it stated
"Only happens with --detailed_debug on" as fact without running that test, and
the untested half is where the customer-facing OOM lived. Nothing in the
template asked for the hedge, so it disappeared between the ticket and review.
The section asks for each untested claim plus what breaks if it is wrong, and
for any hedge on the linked ticket to be carried forward or explicitly closed
out, so reviewers and coding agents have something concrete to attack.
Co-Authored-By: Claude Code <noreply@anthropic.com>
create-release.yml is back to calling the chaos test through no mechanism at
all; it never called it. Also drops the E2E_REDIS_CHAOS opt-in gate itself:
the redis_chaos marker still exists for -m selection and is still excluded
from the per-PR selector by path (tests/e2e/(ui|claude_code|load)/), but the
test no longer needs an env var to run once its file is targeted.
Co-Authored-By: Claude Code <noreply@anthropic.com>
Now that create-release.yml gates stable and RC releases on this test directly,
the weekly schedule is redundant: every release gets a run against its own
commit instead of whatever happened to be on the default branch that Saturday.
workflow_dispatch stays for manual runs.
Co-Authored-By: Claude Code <noreply@anthropic.com>
The chaos test only ran on a weekly cron, so a release could be cut from a
commit it had never covered. Making it callable lets create-release.yml run it
against the exact commit being tagged and refuse to tag if it fails.
Dev, nightly, alpha and beta tags skip the gate: they are cut far more often
than stable and RC tags, and the weekly schedule already covers the default
branch. Input validation moves into the gate job so a malformed tag or SHA
fails before spending a multi-minute chaos run.
Co-Authored-By: Claude Code <noreply@anthropic.com>
Pad the locust payload to tens of KB so per-request bookkeeping cost scales with
body size instead of hiding behind a 40-byte prompt. Turn on
use_redis_transaction_buffer in the chaos config and JSON_LOGS in the workflow so
the spend buffer, pod lock, and JSON-encoded breaker tracebacks are all part of
the measured chaos cost. Add a log-bytes-per-request budget alongside latency,
RSS, and CPU, reading the proxy's log file size at each phase split; its ceiling
is uncalibrated since no chaos run has measured it yet.
Co-Authored-By: Claude Code <noreply@anthropic.com>
The sequential version sent one request at a time, so a Redis outage never
reached the concurrency where the failed-tracking alert body actually grows.
This drives the proxy with locust against one model group of three mock
deployments, two failing at order 1 and one serving at order 2, so every
request spends its retries on the failing pair and lands on the serving
deployment through the order-based fallback. Two phases, a healthy baseline
and a CLIENT PAUSE WRITE window, and every request must succeed in both.
Latency, RSS and CPU are reported as p50/p90/p99 per phase rather than
asserted on: RSS and CPU come from psutil on the proxy's process tree, since
a multi-worker proxy serves /metrics from the prometheus multiprocess
collector and that drops the process collector's series. Thresholds stay open
until weekly runs give real baselines.
Co-Authored-By: Claude Code <noreply@anthropic.com>
* feat(proxy): offload spend tracking to a pod-local spend worker sidecar
py-spy on the gateway showed the post-response _PROXY_track_cost_callback,
spend-log and DBSpendUpdateWriter work running on the inference workers'
event loop, so a DB or Redis stall backed up the request path.
When LITELLM_SPEND_WORKER_ENABLED=true, _ProxyDBLogger serializes one compact
typed SpendEvent per success and hands it to a SpendEventProducer that ships
it over a unix socket (default) or loopback-only TCP to a sidecar started as
`python -m gateway.spend_worker`. The sidecar runs the unchanged
_ProxyDBLogger pipeline against the pod's PgBouncer (pooled_database_url).
When the sidecar is unreachable, the buffer is full, or the gateway shuts
down with events still queued or in flight, the producer applies
LITELLM_SPEND_WORKER_ON_UNAVAILABLE (fallback in-process, or drop). The
sidecar half-closes producers on SIGTERM and drains, the producer treats
EOF as unavailable, and the gateway flushes buffered spend counters on
shutdown. The sidecar honors LITELLM_LOG so its writes are visible in its
own process log.
Helm: both charts gain an opt-in spend-worker sidecar container sharing an
emptyDir socket dir, and the componentized chart's HPA uses a
ContainerResource CPU metric scoped to the gateway container so sidecar
CPU does not drive inference scaling.
* feat(terraform): opt-in spend-worker sidecar for the AWS and GCP gateway stacks
Adds spend_worker_* inputs to both modules. On ECS Fargate the sidecar is a second, non-essential container in the gateway task; on Cloud Run it is a second container in the gateway service. Both listen on loopback TCP, share the gateway's DB/Redis/secret env, and set LITELLM_JOB_ROLE=spend_worker. Disabled by default. Plan-only tests cover both, and the terraform CI workflow now runs the gcp module too
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* test(proxy): retrieve a completed batch in the in-process spend path test
The base now defers cost tracking for batches that are still in flight, so an in_progress batch never reaches update_database
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* refactor(proxy): rename the spend worker sidecar to collector
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(proxy): run the collector from the installed litellm package and finish in-flight fallbacks on shutdown
The sidecar command becomes python -m litellm.proxy.collector so the classic image, whose runtime
stage copies only the installed package, can run it. The module now assembles DATABASE_URL and the
pod-local pgbouncer URL itself, replacing gateway/collector.py
The componentized collector sidecar inherits gateway.volumeMounts so custom CA mounts reach it.
SpendEventProducer shields an in-progress fallback from the writer task cancellation so close()
no longer loses an event already handed to the in-process pipeline
Helpers used across modules (address_argument, should_store_prompts_and_responses_in_spend_logs,
flush_spend_counters_on_shutdown) become public so the change adds no reportPrivateUsage errors
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* ci(terraform): drop the gcp job duplicated by the aws/gcp matrix
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(collector): keep metrics env off the classic sidecar and reject shared loopback ports
The classic chart no longer hands PROMETHEUS_METRICS_PORT and the billing metrics env to the collector container, and gives it the same /.npm scratch mount as the proxy on a read-only root. AWS and GCP now refuse a plan where the spend collector and the metrics sidecar bind the same loopback port. A regression test drives a sidecar crash mid-stream on asyncio and uvloop and checks no event is billed by both the sidecar and the in-process fallback; the producer docstring spells out why a failed drain() cannot double count
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* style(proxy): format pooled_database_url after the pgbouncer rebase
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(proxy): keep the cache-hit preset key and survive dead producers on collector drain
Cache hits updated the logging object after the early return, so the offloaded spend event carried
preset_cache_key=None and the collector re-hashed reconstructed kwargs. Also guard write_eof() against
producer transports uvloop already closed so one dead connection cannot abort the drain
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(terraform): keep the gcp collector port off the metrics sidecar health port
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(proxy): collector connects to Postgres directly under IAM or Entra token auth
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(proxy): mark the collector's DATABASE_URL as pooled when it uses the pod's pgbouncer
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---------
Co-authored-by: yassin <yassin@berri.ai>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* feat(proxy): share database connections across workers with an in-container pgbouncer
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(proxy): parse pgbouncer options iteratively to satisfy the recursion gate
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(proxy): refuse pgbouncer with token db auth and retry failed pooler restarts
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(proxy): build pgbouncer 1.25.2 from a pinned source archive and verify pooler replacements
The public Wolfi repository only carries pgbouncer 1.24.1-r3, which the image
scan rejects (CVE-2026-6664, CVE-2026-6665, CVE-2026-6666, CVE-2025-12819).
All three images now compile the checksummed 1.25.2 release in a builder stage.
The supervisor now waits for a replacement pooler to listen before treating it
as recovered, ends and retries one that never does, and takes the same lock for
stop() and spawn so no replacement can be started after shutdown began.
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(proxy): refuse to start pgbouncer on a loopback port another process already owns
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(proxy): count pgbouncer ready only once its own unix socket answers, not any listener on the port
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(proxy): refuse pgbouncer older than 1.19, whose unix socket cannot vouch for the tcp port
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* feat(helm,terraform): expose the in-container pgbouncer pool for the componentized gateway
Add database.connectionPool to helm/litellm and gateway_connection_pool_* to
terraform/litellm/aws so the componentized gateway can receive the
LITELLM_PGBOUNCER_* env the classic image already honours. Both reject the
pool under IAM or Entra token auth at render/plan time: the pooler holds one
static database password for the life of the pod or task.
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* feat(gateway): launch the componentized gateway image through a pgbouncer-aware supervisor (#40592)
The componentized gateway image started uvicorn directly, so the in-container
PgBouncer never ran for it: every worker opened its own Prisma pool to the
database. It also passed no keep-alive timeout, so behind a load balancer with
a 60s idle timeout uvicorn's 5s default closed idle connections first and the
balancer returned 502s on scale-out
gateway.launch assembles DATABASE_URL, starts PgBouncer once per pod when
LITELLM_PGBOUNCER_ENABLED is set, hands the workers the loopback URL and then
runs uvicorn on gateway.main:app with KEEPALIVE_TIMEOUT as --timeout-keep-alive.
The image builds PgBouncer 1.25.2 from a checksummed tarball, copies the
compiled Rust extension into the /app source tree it imports from (it was only
in site-packages, which PYTHONPATH=/app shadows) and asserts the native bridge
loads. The app user is added to stats_users so operators can read the PgBouncer
console with the application credentials
The supervisor returns the pooled URL instead of writing into the mapping it
was handed, a database user whose name PgBouncer would split into several
stats_users entries is refused before the config is written, and the launcher
tests drive main() with an injected serve callable
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* docs(terraform): describe the gateway.launch pooler entrypoint in the aws module README
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(proxy): run pgbouncer exit hooks only in the parent and copy the CA into the runtime dir
Gunicorn workers inherit the parent's atexit table, so a recycled worker (max_requests) stopped the shared pooler and removed its runtime dir, then hung in the inherited Popen lock. The hooks now no-op unless os.getpid() is the process that started PgBouncer
A verified TLS upstream named the operator's CA bundle directly, which is often a 0600 root-owned file that nobody (the user PgBouncer drops to) cannot read, so every server connection failed with "failed to load CA". The bundle is copied into the runtime dir next to the ini and chowned with it
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* feat(terraform): run the gateway through gateway.launch and add the gcp connection-pool variables
Cloud Run and ECS overrode the image command with uvicorn gateway.main:app, which skips the supervisor that starts the in-container PgBouncer, so LITELLM_PGBOUNCER_ENABLED was inert on both stacks. Both now exec python -m gateway.launch (under ddtrace-run when USE_DDTRACE is set), and the gcp module gains gateway_connection_pool_enabled / gateway_pool_max_db_connections / gateway_pool_max_client_conn wired to the gateway service only
The test_launch password_env fixture now restores DATABASE_URL even when it was unset: monkeypatch.delenv records nothing for an absent var, so main() left postgresql://...@db.internal in the xdist worker's environ and the key-rotation e2e test in the same proxy-infra shard stopped skipping and tried to reach db.internal
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(pgbouncer): keep channel_binding and gssencmode off the loopback URL
Prisma would demand TLS channel binding from a pooler that only speaks
plain TCP on 127.0.0.1. Also pass the request the marketplace test
started needing after #40518 landed on top of #40496
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---------
Co-authored-by: yassin <yassin@berri.ai>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
A loopback Redis answers many commands inside the 1 ms socket timeout, so nothing guaranteed
the failure path ran. The test now holds the proxy's Redis in CLIENT PAUSE WRITE for its
duration, so every write the proxy sends, the spend counter increment included, outlives the
timeout, and lifts the pause in teardown. Reads stay live so the control connection can do that.
Enable the prometheus callback in the gateway config and assert from /metrics that the proxy
counted at least the breaker's five timeouts and that, during each case, it saw fresh timeouts,
a breaker transition, or an open breaker rejecting every call. The open breaker is the state a
customer's worker sits in, and cost tracking fails on every request either way.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014ZDULyJPp17ZFiJenRxs2T
It is a functional e2e test, not a load test, so give it its own workflow instead of a job
inside the load anomaly run. It keeps the Saturday 12:00 UTC cadence and manual dispatch, and
boots the timeout-config proxy with Postgres and Valkey services exactly as before.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014ZDULyJPp17ZFiJenRxs2T
Add tests/e2e/router/test_redis_timeout_e2e.py against a proxy booted from
tests/e2e/gateway/redis_timeout_ci_config.yml: a real Redis with socket_timeout 0.001, so every
command times out and the circuit breaker opens, plus a primary deployment that always fails and
falls back to a healthy one, so every request carries retry breadcrumbs into cost tracking. The
test drives twenty chat requests through the proxy and asserts each answers within ten seconds,
the last third is no slower than the first, /health/liveliness stays fast, and every request
still reaches the spend log.
Gate it behind the redis_timeout marker and E2E_REDIS_TIMEOUT, exclude it from the per-PR
e2e-changed selector, register the reliability.circuit_breaker.redis_timeout.stays_responsive
cell, and run it as its own job in the weekly load anomaly workflow with a Postgres and Valkey
service. Against a v1.100.0 proxy the run wedges the worker: requests time out and liveliness
stops answering (LIT-6780).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014ZDULyJPp17ZFiJenRxs2T
* feat(skills): semantic search over the LiteLLM-hosted skill registry
Adds GET /v1/skills?query= (custom_llm_provider=litellm_proxy) and a
skill_search MCP virtual tool, ranking the caller's accessible skills by
semantic similarity, mirroring the A2A agent registry search (LIT-6309).
Also fixes a pre-existing bug where create_skill() dropped description and
instructions for the litellm_proxy provider, which left every LiteLLM-hosted
skill with no searchable text.
* fix(mcp): coerce skill_search top_k instead of raising 500 on malformed input
The MCP-REST skill_search dispatch validated raw tool arguments through a
pydantic model directly, so a non-numeric top_k raised a ValidationError
that the endpoint's catch-all turned into an HTTP 500. Mirrors the
agent_search branch's tolerant coerce_top_k handling instead.
* fix(skills): enforce key limits on search embeddings and bound the semantic index
Semantic search embeddings now run the same pre_call_hook the /embeddings
route runs, so key rate limits, budgets and guardrails apply before the
embedding model is called. The shared SemanticTextIndex caps cached vectors
and evicts the least recently searched entries, and each skill's embedded
text is capped so one skill cannot inflate the embedding batch
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(skills): surface proxy 429s from search embeddings instead of a 503
ProxyRateLimitError is also an OpenAIError, so the search engine was folding
a key rate limit into skill_search_unavailable. Proxy HTTPExceptions now
propagate so the caller gets the same 429 the /embeddings route returns
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(skills): import assert_never from typing_extensions for Python 3.10
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* fix(skills): embed the request as the pre-call hooks returned it, not the original text
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* refactor(skills): keep the litellm_proxy provider check for GET /v1/skills?query= inside llms/
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* test(skills): move the GET /v1/skills?query= endpoint tests under tests/test_litellm/proxy
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---------
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* ci: limit Rust workflows to Rust directory changes
* ci: run Rust checks when their workflow changes
* ci: report Rust wheels only for successful Rust changes
* ci: keep Rust wheel reports in the workflow summary
* ci: group Rust lint and validation jobs
* ci: keep Rust job names distinct from required lint and test checks
* ci: drop the unused Python setup from the Rust lint job
---------
Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Its job id was `test`, the same check-run name test-mcp.yml publishes and
guard-internal-staging requires, so a commit touching the sweep carried two
check runs called `test`. Renaming it to `sweep-tests` leaves the required
context with exactly one job behind it.
A ruleset's required status check names a check run and GitHub matches it by
that name alone, so two jobs publishing the same name leave the gate unable to
say which job proved it. The new code-quality check reads every workflow,
expands matrix values and local reusable-workflow calls the way Actions does,
and fails when one name has more than one job behind it.
The selector picked up two suites that can never pass in this stack, so
editing either one turned the check permanently red: the presidio masking
suite calls pytest.fail without an analyzer and anonymizer that up.sh
never starts, and the pipecat audio suite skips itself at import time
unless the NLTK punkt_tab data is present, which nothing installs.
tests/e2e/coverage_registry/test_collector.py had the same problem for a
different reason. Its nested pytest.main autoloads pytest-retry from the
ci group the workflow installs and dies with "INTERNALERROR: no option
named 'filtered_exceptions'", so the collect-only pass now disables that
plugin. The plugin's entry point is pytest-retry, not retry, so the same
one-word fix lands on mutmut's pytest_add_cli_args, where "-p no:retry"
was disabling nothing.
Two smaller holes in the harness: a canary argument the shell never
expanded used to select nothing and let the gate pass green, and a secret
that cannot be represented in both bash and dotenv was rejected without
naming the key.
A harness-only change (proxy_client.py, conftest.py, pytest.ini, the gateway
config, .github/e2e-stack, or the workflow) selected nothing, so the stack was
never exercised by the change that touched it. select_tests.py keeps the
changed-file rule and adds the access_control suite whenever a harness file
changes. The run step now reports the pytest exit code before the evidence
check, prints pytest's summary line per pass so the rerun count is visible,
and assert_tests_ran.py names each failed or errored test as classname::name
A one-character value in the provider secret bundle was masked too, which
turned every 1 in the run log into ***, including the pass numbers and the
gateway addresses, so the only public diagnostics were unreadable
The changed-tests workflow overrode the suite's `--reruns 1` with `--reruns 0`, so a
transport blip failed a pass that pytest.ini already scopes to network errors and
5xx responses. Pass 2 of run 33692484803 also went red 15s after a model write with
"no healthy deployments": the barrier only polled /v1/models through nginx, which
proves one gateway converged, and the next request rolled the other. The stack now
exports LITELLM_PROXY_REPLICA_URLS, the barrier polls every replica with the full
budget before settling, and up.sh refuses to boot without DD_API_KEY, since the
gateway config enables the datadog callback on every run
* fix(ci): grant pull_requests write for release wheel reporter
The reporter posts a PR comment via github.rest.issues.createComment.
GitHub requires both issues=write and pull_requests=write to comment on
a PR issue, as returned in x-accepted-github-permissions. The workflow
had pull-requests: read, so the POST failed with 403 'Resource not
accessible by integration'. Bumping to pull-requests: write fixes the
create path; the read-only pulls.get call still works. Same-repo scope
is preserved by the existing head_repository.full_name check.
Co-authored-by: Krrish Dholakia <krrish-berri-2@users.noreply.github.com>
* fix(ci): scope release wheel reporter permissions to pull requests
---------
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Krrish Dholakia <krrish-berri-2@users.noreply.github.com>
Co-authored-by: Yujong Lee <yujong@berri.ai>
