8 views
-/https://github.com/berriai/litellm/issues/30118
GitHub · issue

#30118 [Bug]: UI login redirects to internal Docker container IP after upgrade to v1.88.x (regression from v1.87.0)

  • State: open
  • Author: @markiceman
  • Labels: bug

### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

What happened?

After upgrading from v1.87.0 to v1.88.1, the LiteLLM Admin UI stopped working when deployed behind an nginx reverse proxy. The browser is redirected to the internal Docker container IP instead of the public URL.

Downgrading back to v1.87.0 immediately fixes the issue. This is a confirmed regression.

Steps to reproduce:

1. Deploy LiteLLM v1.88.x behind nginx reverse proxy (Docker Compose, `num_workers=4`) 2. Set `PROXY_BASE_URL=https://your-public-domain.com` and `FORWARDED_ALLOW_IPS=*` in environment 3. Navigate to `https://your-public-domain.com/ui` 4. Browser is redirected to `http://<docker_container_ip>:4000/ui/login/` — inaccessible from outside the Docker network

Expected behavior:

Redirect should go to `https://your-public-domain.com/ui/login/` — same as in v1.87.0.

Environment:

- LiteLLM version: v1.88.1 (broken), v1.87.0 (works) - Deployment: Docker Compose, `num_workers=4` - Reverse proxy: nginx with `proxy_set_header Host $host`, `X-Forwarded-Proto`, `X-Forwarded-Host` — all correctly set - `PROXY_BASE_URL` and…

GitHub resolver

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

Refresh page
vote history (3 events)
#0 of 0 · 31d18h54m4s ago — entered · #import:https:::github.com:berriai:litellm post #1978
Issue #29320 is substantially harder: it requires a new opt-in compaction subsystem spanning proxy and SDK request paths, callback lifecycle integration, configuration precedence, provider/model execution, fail-open behavior, observability, persistence, headers, and broad testing. Issue #30118 is a narrowly scoped regression involving URL/forwarded-header handling and deployment-specific redirect behavior.
The proxy-related regression likely requires tracing URL and request-context handling across backend middleware, deployment configuration, and integration coverage, while the input-method issue is a localized frontend event-handling fix with narrower testing.
#0 of 0 · 31d18h17m1s ago — current · #import:https:::github.com:berriai:litellm post #2574
Issue #30118 is harder because it requires reproducing and diagnosing a version-specific reverse-proxy regression, tracing forwarded-host and scheme handling across the UI/auth redirect flow, and adding robust multi-worker deployment coverage. Issue #33777 appears to be a localized malformed endpoint configuration or validation problem.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search