fix: remove separate health app

[yassin-berriai]

Relevant issues

Linear ticket

Resolves LIT-2908

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/test_litellm/ directory, Adding at least 1 test is a hard requirement - see details
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem
• I have requested a Greptile review by commenting @greptileai and received a Confidence Score of at least 4/5 before requesting a maintainer review

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

CI (LiteLLM team)

CI status guideline:

• 50-55 passing tests: main is stable with minor issues.
• 45-49 passing tests: acceptable but needs attention
• <= 40 passing tests: unstable; be careful with your merges and assess the risk.

• Branch creation CI run
  Link:

• CI run for the last commit
  Link:

• Merge / cherry-pick CI run
  Links:

Screenshots / Proof of Fix

Type

🆕 New Feature
🐛 Bug Fix
🧹 Refactoring
📖 Documentation
🚄 Infrastructure
✅ Test

Changes

Remove SEPARATE_HEALTH_APP and the supervisord scaffolding that existed solely to support it. This ran the proxy's health endpoints as a second FastAPI app on a second port, in a separate process, supervised alongside the main proxy. That defeats the purpose of a liveness probe.

The Kubernetes docs are explicit about what a liveness probe is for:

Liveness probes determine when to restart a container. For example, liveness probes could catch a deadlock, where an application is running, but unable to make progress.

Liveness probes must be configured carefully to ensure that they truly indicate unrecoverable application failure, for example a deadlock.

— Kubernetes docs: Liveness, Readiness, and Startup Probes

A probe is supposed to fire when the application itself is wedged — event loop blocked, deadlocked under load, GIL-bound, etc. A probe served from a different process can't observe any of that. It happily returns 200 while the thing it's supposed to be checking is dead, which gives Kubernetes false confidence not to restart a deadlocked proxy.

Colin Breck names exactly this failure mode:

For example, if the service hosts two web servers — one for the service routes and one for status routes, like readiness and liveness probes, or metrics collection — the service can be slow or unresponsive, while the liveness probe route returns just fine.

— Colin Breck: Kubernetes Liveness and Readiness Probes — How to Avoid Shooting Yourself in the Foot

Updated docs: BerriAI/litellm-docs#99

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[greptile-apps]

Greptile Summary

This PR removes the SEPARATE_HEALTH_APP feature — a supervisord-based pattern that ran health endpoints in a second FastAPI process on a separate port — replacing it with the straightforward single-process model where liveness/readiness probes are served by the same event loop they are meant to observe.

• All three Dockerfiles drop supervisor from their apk installs and stop copying supervisord.conf; prod_entrypoint.sh removes the branch that exec'd supervisord when SEPARATE_HEALTH_APP=1.
• docker/supervisord.conf and litellm/proxy/health_endpoints/health_app_factory.py are fully deleted; TestHealthAppFactory tests are removed alongside them and replaced with a new TestRunServerDbSetup class that exercises prisma DB setup logic via mocks.
• Helm chart removes separateHealthApp/separateHealthPort values and hardens probe port references to always use "http".

Confidence Score: 5/5

Safe to merge — the removal is complete and consistent across all Dockerfiles, the Helm chart, the entrypoint, and the test suite.

Every artifact tied to the separate-health-app pattern (supervisord config, factory module, Dockerfile installs, Helm conditionals, entrypoint branch) is removed in lock-step. The replacement test class uses only mocks and makes no real network calls. The deliberate breaking change for existing SEPARATE_HEALTH_APP=1 users has already been reviewed and accepted by a senior developer in a prior thread.

No files require special attention.

Important Files Changed

Filename	Overview
docker/prod_entrypoint.sh	Removes the SEPARATE_HEALTH_APP=1 branch that exec'd supervisord; also fixes the missing trailing newline.
docker/supervisord.conf	Entire file deleted; the supervisord configuration was the only artifact that tied the separate health process to the main process.
litellm/proxy/health_endpoints/health_app_factory.py	Entire file deleted — the factory that spun up the second FastAPI app is no longer needed.
deploy/charts/litellm-helm/templates/deployment.yaml	Removes separateHealthApp/separateHealthPort env injection and port definition; hardcodes liveness/readiness/startup probe port to "http".
deploy/charts/litellm-helm/values.yaml	Removes separateHealthApp and separateHealthPort keys from the default values file.
tests/test_litellm/proxy/test_proxy_cli.py	Removes TestHealthAppFactory (tests for the deleted factory) and adds TestRunServerDbSetup which tests prisma DB setup behaviour via mocks — no real network calls.
Dockerfile	Removes supervisor from apk install and drops COPY docker/supervisord.conf — clean removal of the separate-health-app scaffolding.
docker/Dockerfile.database	Mirrors main Dockerfile: removes supervisor from apk install and drops COPY docker/supervisord.conf.
docker/Dockerfile.non_root	Removes supervisor from apk install and removes the multi-stage copy of supervisord.conf from the builder layer.
litellm/proxy/proxy_cli.py	Removes the comment block advertising the separate-health-app uvicorn invocation pattern.

_{Reviews (2): Last reviewed commit: "fix: remove separate health app" | Re-trigger Greptile}

[greptile-apps]

Breaking change for existing SEPARATE_HEALTH_APP users

Any deployment that currently runs with SEPARATE_HEALTH_APP=1 set as an environment variable was relying on supervisord launching a dedicated health-check process on a separate port. After this change, that env var is silently ignored and the regular single-process litellm server starts instead. Users who chose the separate health app specifically to keep probes alive under main-process load (the primary motivation for the feature) will lose that isolation with no migration path or deprecation notice.

Rule Used: What: avoid backwards-incompatible changes without... (source)

[yassin-berriai]

This flag should never have been supported, it causes k8s deployments to silently fail since health checks wont check the actual health of litellm

[yassin-berriai]

@greptileai