Commit graph

8 commits

Author SHA1 Message Date
Yuneng Jiang
d1b160310f
fix(docker): require a generated master key in the quickstart stack
The committed sk-1234 placeholder is in PUBLICLY_KNOWN_MASTER_KEYS, so once an
image ships fe480533e8 the proxy refuses to boot and the quickstart stops
working. It also meant the documented stack came up on port 4000 with a
credential anyone could guess.

Both keys now come from .env and compose refuses to render without them. The
salt key is generated alongside so it stays stable across restarts, which
keeps stored credentials readable.

Verified: no .env -> compose fails closed naming the missing variable; with a
generated .env the stack is healthy, /v1/models returns 200 for the generated
key, 401 for sk-1234 and 401 unauthenticated, /ui/ serves, and the key still
works after a restart.
2026-09-21 14:27:29 -07:00
Yuneng Jiang
7590012a03
feat(docker): add a quickstart compose file served from the product repo
The docs quickstart pipes a compose file hosted on the docs site straight
into `docker compose -f -`. That puts the content users execute in the docs
repo rather than here, and nothing lands on disk for them to read first.

Move the two-service stack (gateway + Postgres) into docker/ so it ships and
is reviewed alongside the code it starts, and pin the image to main-stable
instead of latest. The docs change to download-then-run follows separately.

Verified: `docker compose up -d` brings the stack healthy, /health/liveliness
returns 200, /v1/models returns 200 with the placeholder key and 401 without,
and /ui/ serves.
2026-09-21 14:12:55 -07:00
Sameer Kankute
7b7f304675
fix(mcp): block arbitrary command execution via stdio transport
Add command allowlist for MCP stdio transport to prevent RCE via
/mcp-rest/test/* endpoints. Restrict test endpoints to PROXY_ADMIN
role. Fix docker/README.md MASTER_KEY -> LITELLM_MASTER_KEY.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-04-08 19:42:37 +05:30
Achilleas Athanasiou Fragkoulis
cb95b1cf92 fix: Add LITELLM_UI_PATH and LITELLM_ASSETS_PATH for read-only filesystem support (#20492)
Fixes #19578

---

When deploying the LiteLLM proxy with `readOnlyRootFilesystem: true` in Kubernetes, UI routes returned `404` because:

- Hardcoded paths:
  - `/var/lib/litellm/ui`
  - `/var/lib/litellm/assets`
- Runtime copy/restructure operations failed on read-only filesystems
- No detection mechanism for pre-restructured UI

---

Add configurable environment variables with intelligent detection, graceful fallbacks, and code quality improvements.

---

- **`LITELLM_UI_PATH`** — Custom UI directory location
  - Default: `/var/lib/litellm/ui` (when `LITELLM_NON_ROOT=true`)
  - Default: packaged UI path (otherwise)
  - Example: `/app/var/litellm/ui` for `emptyDir` volumes

- **`LITELLM_ASSETS_PATH`** — Custom assets directory location
  - Default: `/var/lib/litellm/assets` (when `LITELLM_NON_ROOT=true`)
  - Default: current working directory (otherwise)
  - Example: `/app/var/litellm/assets`

---

UI is detected as **pre-restructured and ready** if any of the following apply:

1. **Primary**: `.litellm_ui_ready` marker file exists (created by Dockerfile)
2. **Fallback**: Pattern-based detection — finds *any* subdirectory containing `index.html`
   (resilient to UI structure changes; no hardcoded route names)
3. **Safety**: Filesystem writability check before operations

---

**`litellm/proxy/proxy_server.py`**

- `_validate_ui_directory()` — Verifies UI has required structure (`index.html`, `_next/`)
- `_is_ui_pre_restructured()` — Pattern-based detection (not hardcoded routes)
- `_try_populate_ui_directory()` — Helper for clean error handling
- Refactored UI path decision tree with numbered cases (1, 2, 3, 4a, 4b)
- Updated UI path logic to use `LITELLM_UI_PATH`
- Added writability checks before copy/restructure operations
- Graceful fallback to packaged UI if operations fail
- Updated `server_root_path` replacement with read-only check
- Simplified assets directory creation (try/except instead of complex parent checks)
- Updated `get_image()` endpoint to use `LITELLM_ASSETS_PATH`
- Added validation for packaged and final UI paths

**`docker/Dockerfile.non_root`**

- Added `touch .litellm_ui_ready` marker after UI restructuring
- Enables automatic detection of pre-built UI in Docker images

**`tests/proxy_unit_tests/test_ui_path_detection.py`**

- Added comprehensive unit tests for new functionality
- Tests env var handling, detection logic, and writability checks

---

**`docs/my-website/docs/proxy/config_settings.md`**

- Added `LITELLM_UI_PATH` and `LITELLM_ASSETS_PATH` to env vars table
- Documented defaults and use cases

**`docs/my-website/docs/proxy/prod.md`**

- Added comprehensive "Read-Only Root Filesystem" section
- Quick fixes for permission errors
- Full Kubernetes setup with `initContainer` + `emptyDir` volumes
- API-only deployment option
- Environment variables reference table
- Notes on migrations, caching, and `server_root_path`

**`docker/README.md`**

- Updated hardened setup notes to mention pre-built UI
- Added details about UI serving from read-only paths

---

- No breaking changes
- Existing deployments continue working without modifications
- New env vars are optional with sensible defaults
- Detection logic supports both old and new builds
- Graceful fallbacks throughout

---

```yaml
apiVersion: apps/v1
kind: Deployment
spec:
  template:
    spec:
      initContainers:
        - name: setup-ui
          image: ghcr.io/berriai/litellm:main-stable
          command: ["sh", "-c", "cp -r /var/lib/litellm/ui/* /app/var/litellm/ui/"]
          volumeMounts:
            - name: ui-volume
              mountPath: /app/var/litellm/ui
      containers:
        - name: litellm
          env:
            - name: LITELLM_UI_PATH
              value: "/app/var/litellm/ui"
            - name: LITELLM_ASSETS_PATH
              value: "/app/var/litellm/assets"
          securityContext:
            readOnlyRootFilesystem: true
          volumeMounts:
            - name: ui-volume
              mountPath: /app/var/litellm/ui
      volumes:
        - name: ui-volume
          emptyDir:
            sizeLimit: 100Mi
2026-02-12 19:39:04 +05:30
Mateo Di Loreto
107ea9043a
[Feature] Download Prisma binaries at build time instead of at runtime for Security Restricted environments (#17695)
* Use config file to enable prometheus metrics

* Revert "Use config file to enable prometheus metrics"

This reverts commit 15ae36e171.

* Improve hardened stack and Prisma offline flow

* Document hardened compose usage

* Remove undesired change in fastapi-sso

* Restore dashboard lockfile

* Remove unecessary tempdirs

* Document hardened/offline Docker validation flow
2025-12-16 21:25:53 +05:30
Kowyo
858f557bce docs: use docker compose instead of docker-compose 2025-09-29 11:59:53 +00:00
Mateo Di Loreto
6e5fe51184
add openssl in apk install in runtime stage in dockerfile.non_root (#13168)
* add openssl in apk install in runtime stage in dockerfile.non_rootdocker-compose logs -f litellm

* Improve Docker-compose.yaml for local debugging

---------

Co-authored-by: Krish Dholakia <krrishdholakia@gmail.com>
2025-07-31 21:52:11 -07:00
Krrish Dholakia
14e501845f fix(proxy_server.py): add support for setting master key via .env 2024-01-03 15:10:25 +05:30