fix(helm): mount gateway config.yaml on backend Deployment

[yassin-berriai]

Summary

• The componentized chart's backend Deployment never mounted {fullname}-gateway-config or set CONFIG_FILE_PATH, so anything declared in gateway.config.proxy_config.litellm_settings (callbacks, cache, guardrails, ...) only loaded on the gateway process — even though the backend runs the same proxy code and needs the same callbacks to fire for management-API and proxy-level OTel spans.
• Reuse the existing gateway.config.create toggle and the existing ConfigMap: when true, the backend now also gets CONFIG_FILE_PATH=/app/config/config.yaml, the gateway-config volume + volumeMount, and a checksum/config pod annotation so ConfigMap edits roll backend pods alongside gateway pods.
• No values.yaml change. No new toggle. When gateway.config.create=false, the backend Deployment is byte-identical to before.

Why reuse the gateway ConfigMap?

The litellm proxy reads a single config.yaml; splitting per-component would let the two drift and break shared settings (callbacks, cache, guardrails) on whichever side forgets to set them. Reusing keeps a single source of truth and matches runtime reality.

Test plan

• helm template with gateway.config.create=true shows the backend Deployment now contains CONFIG_FILE_PATH, the gateway-config volumeMount + volume, and a checksum/config annotation pointing at {release}-litellm-gateway-config.
• helm template with gateway.config.create=false produces a backend Deployment with no CONFIG_FILE_PATH, no volumeMounts, no volumes, and no checksum/config annotation (identical to main).
• Deploy to a cluster, edit litellm-gateway-config to add a callback, confirm backend pods roll and the new callback is registered in backend logs.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you all sign our Contributor License Agreement before we can accept your contribution.
0 out of 2 committers have signed the CLA.

❌ claude
❌ yassin-berriai
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

Greptile Summary

This PR fixes the backend Deployment in the componentized Helm chart so it mounts and reads the same gateway-config ConfigMap that the gateway process already uses, ensuring shared settings (callbacks, cache, guardrails) apply to both processes.

• When gateway.config.create=true, the backend Deployment now receives CONFIG_FILE_PATH=/app/config/config.yaml, a gateway-config volumeMount/volume referencing the existing ConfigMap, and a checksum/config pod annotation that rolls backend pods whenever the ConfigMap changes — mirroring the gateway Deployment exactly.
• When gateway.config.create=false, the implementation is functionally equivalent to the original, though a bare annotations: key is now always emitted in the pod spec regardless of whether backend.podAnnotations is populated.

Confidence Score: 4/5

The change is safe to merge; the new mount, env var, and checksum annotation are only active when the user explicitly opted in via gateway.config.create=true, and the implementation faithfully mirrors the gateway Deployment.

The core logic — volume, volumeMount, env var, and checksum annotation — is correct and matches the gateway Deployment exactly. The one cosmetic discrepancy is a bare annotations: key that is now always emitted even when there is nothing to annotate, which disagrees with the PR description's byte-identical claim but causes no runtime problem.

helm/litellm/templates/backend/deployment.yaml — the unconditional annotations block

Important Files Changed

Filename	Overview
helm/litellm/templates/backend/deployment.yaml	Adds gateway ConfigMap mount, CONFIG_FILE_PATH env, and checksum/config annotation to backend Deployment when gateway.config.create=true; unconditionally emits an annotations: block that was previously absent when podAnnotations is empty

_{Reviews (1): Last reviewed commit: "fix(helm): mount gateway config.yaml on ..." | Re-trigger Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[mateo-berri]

Superceded by #29605. Thanks