[Bug]: non_root image missing /app/enterprise — "Managed files hook not found" on POST /v1/files (fix exists in main, needs backport to stable)

[alexhuk3]

Check for existing issues

• I have searched the existing issues and checked that my issue is not a duplicate.

What happened?

On the hardened image ghcr.io/berriai/litellm-non_root (reproduced on main-v1.81.3-stable and main-v1.81.9-stable), every managed-files upload fails with HTTP 500
{"error":{"message":"Managed files hook not found",...}}. The identical request against the regular image ghcr.io/berriai/litellm:v1.81.3-stable — same config.yaml, same
database, same request — succeeds. Reads (GET /v1/files) work on both images; only the write path is broken.

Root cause (verified): litellm/proxy/hooks/__init__.py registers enterprise hooks via from enterprise.enterprise_hooks import ENTERPRISE_PROXY_HOOKS inside try/except ImportError that silently swallows the failure. The enterprise package is a top-level repo directory that is importable only because proxy_cli.py does
sys.path.append(os.getcwd()) with WORKDIR /app:

• regular Dockerfile runtime stage: COPY . . → /app/enterprise exists → hook registers;
• docker/Dockerfile.non_root runtime stage copies only selected paths and never copies /app/enterprise → silent ImportError → managed_files missing from PROXY_HOOKS → 500
  on every write.

Inside a running non_root container:

$ python3 -c "import enterprise"   # ModuleNotFoundError
$ python3 -c "from litellm.proxy.hooks import PROXY_HOOKS; print(sorted(PROXY_HOOKS))"
['cache_control_check', 'litellm_skills', 'max_budget_limiter', 'parallel_request_limiter', 'responses_id_security']

The fix already exists in main: docker/Dockerfile.non_root line ~141 has COPY --from=builder /app/enterprise /app/enterprise (present in source tags ≥ v1.90.0). But the
newest published non_root stable image (main-v1.81.9-stable) predates it, so all published non_root stable images ship broken. Request: backport that COPY to the stable line
/ rebuild stable non_root images. Also consider logging a warning instead of silently swallowing the ImportError — it would make this a 5-minute diagnosis.

Expected: POST /v1/files with target_model_names behaves identically on both image variants.

Related but distinct: #32782 (missing litellm_enterprise pip package in another context).

User Flow

Before a (hypothetical) fix: an automation engineer uploading a video for Gemini analysis through a gateway running the hardened non_root image gets a 500 on every upload

1. They send POST https://litellm-domain/v1/files as multipart with file=@video.mp4, purpose=user_data, target_model_names=gemini-2.5-flash
2. The response is HTTP 500 with body {"error":{"message":"Managed files hook not found","type":"None","param":"None","code":"500"}}
3. They send GET https://litellm-domain/v1/files and get 200 with a normal (empty) file list, so reads clearly work
4. They retry the exact same POST against a gateway running the regular (root) image with the same configuration and get 200 with a unified file id — so their request is valid and
   only the hardened image rejects it

After a (hypothetical) fix: the same upload on the hardened non_root image succeeds like on the regular image

1. They send the same POST https://litellm-domain/v1/files as multipart with file=@video.mp4, purpose=user_data, target_model_names=gemini-2.5-flash
2. The response is HTTP 200 with a unified file id in the body
3. GET https://litellm-domain/v1/files returns 200 and now lists the uploaded file
4. A follow-up chat completion referencing that file id runs against gemini-2.5-flash and returns normally

Proof the bug occurs

Config / setup the proxy ran with:

# config.yaml
general_settings:
  master_key: sk-REDACTED
  database_url: os.environ/DATABASE_URL   # local postgres, e.g. postgresql://litellm:***@localhost:5432/litellm
model_list:
  - model_name: gemini-2.5-flash
    litellm_params:
      model: gemini/gemini-2.5-flash
      api_key: os.environ/GEMINI_API_KEY

Version or commit: images ghcr.io/berriai/litellm-non_root:main-v1.81.9-stable (broken) vs ghcr.io/berriai/litellm:v1.81.3-stable (works)

Commands and their full output:

# 1) broken: non_root image
docker run --rm -p 4000:4000 --network host \
  -e DATABASE_URL=postgresql://litellm:REDACTED@localhost:5432/litellm \
  -e GEMINI_API_KEY=REDACTED \
  -v $PWD/config.yaml:/app/config.yaml \
  ghcr.io/berriai/litellm-non_root:main-v1.81.9-stable --config /app/config.yaml --port 4000
  
curl -sS -X POST http://localhost:4000/v1/files \
  -H "Authorization: Bearer sk-REDACTED" \
  -F purpose=user_data -F target_model_names=gemini-2.5-flash -F file=@video.mp4
# → <вставь реальный ответ: HTTP 500 Managed files hook not found>

# 2) supporting evidence — the module is physically absent in the image:
docker exec <container> python3 -c "import enterprise"
# → ModuleNotFoundError: No module named 'enterprise'

# 3) works: regular image, same config, same DB, same curl
docker run ... ghcr.io/berriai/litellm:v1.81.3-stable --config /app/config.yaml --port 4000
curl  (same as above)
# → <вставь реальный ответ: 200 + file id>

What part of LiteLLM is this about?

Other

What LiteLLM version are you on ?

v1.81.3-stable

Twitter / LinkedIn details

No response

[csarushan1729]

Hi @alexhuk3 , I'd like to take this one. I'll backport the COPY --from=builder /app/enterprise /app/enterprise line to docker/Dockerfile.non_root (missing since before v1.81.9-stable, present in main ≥v1.90.0) and add a log warning for the silent ImportError in litellm/proxy/hooks/__init__.py. Will verify by rebuilding the non_root image and confirming POST /v1/files succeeds before opening the PR.