7 views
-/https://github.com/berriai/litellm/issues/27173
GitHub · issue

#27173 [Bug]: Helm 1.1.0 standalone DB secret drifts from postgresql.existingSecret and causes CrashLoop

  • State: open
  • Author: @ben-wangz

## Check for existing issues

- [x] I have searched existing issues and this appears related to (but not resolved by): - #13544 - #21716

## What happened?

When deploying `litellm-helm` `1.1.0` via ArgoCD with bundled Postgres (`db.deployStandalone=true`) and `postgresql.auth.existingSecret` set, LiteLLM still uses/creates `litellm-dbcredentials` in a way that drifts from the intended secret.

In our test, `litellm-credentials` had the correct password, but `litellm-dbcredentials` ended up with a different/default value (e.g. `NoTaGrEaTpAsSwOrD`). LiteLLM then failed Prisma DB auth (`P1000`) and stayed in `CrashLoopBackOff`.

## What I expected

- If `db.dbCredentialsSecretName` and `postgresql.auth.existingSecret` are both set to the same Secret, all DB auth values should remain consistent. - The chart should not render or preserve conflicting DB credentials that make LiteLLM fail against the bundled Postgres.

## Environment

- Chart: `litellm-helm:1.1.0` - Deployment method: ArgoCD `Application` - Cluster: k3s - LiteLLM image: `ghcr.io/berriai/litellm:main-v1.83.14-stable` - Postgres image: `bitnamilegacy/postgresql`

## Steps to Reproduce

1. Create `litellm-credentials` s…

GitHub resolver

Import GitHub neighbors on demand. Results are saved as system ingests.

Refresh page
vote history (3 events)
#0 of 0 · 31d19h16m41s ago — entered · #import:https:::github.com:berriai:litellm post #1697
Cross-component deployment configuration, secret precedence, template behavior, and regression testing create substantially more integration risk; the other is a localized parameter-propagation fix with narrower test scope.
The left issue is harder because it requires diagnosing and correcting provider-specific request translation across multiple tool types, with compatibility and regression risks in the proxy path. The right issue is more localized to Helm secret templating and chart validation, with a narrower implementation surface.
#0 of 0 · 31d19h4m58s ago — current · #import:https:::github.com:berriai:litellm post #1879
The left issue requires tracing and correcting stateful cross-turn translation behavior across Gemini response parts and the Claude-compatible messages path, with provider-specific semantics and regression coverage. The right issue is more localized to Helm chart secret precedence/rendering and deployment validation, making it lower risk and narrower in scope.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search