Files
KohakuHub/docker-compose.example.yml
T
narugo1992andClaude Opus 4.7 aff9fd47ef perf(cache): add Valkey-based L2 cache infrastructure
Introduces the prerequisite cache layer tracked in #73. No business code
yet consumes the helpers — this is plumbing only, gated by
KOHAKU_HUB_CACHE_ENABLED (default: false). Subsequent issues will adopt
specific cache patterns on top of this foundation.

Why now: every hot read endpoint currently makes 1–3 LakeFS REST calls
plus several Postgres queries per request, with the only existing cache
being a per-process cachetools.TTLCache in fallback/cache.py — useless
across the default 4-worker uvicorn deployment.

Design highlights (full design in docs/development/cache.md):

- Pure cache, no source-of-truth state. Silent-degradation contract:
  every cache call is wrapped in try/except and falls back to L3 when
  Valkey is unreachable. CI runs a dedicated cache-disabled job to
  regression-guard this.
- L1 (per-worker cachetools) restricted to immutable / content-addressed
  data only — multi-worker uvicorn has no portable cross-worker
  invalidation channel, and constraining L1 to "key contains its own
  version" sidesteps that entirely.
- L2 (Valkey) holds everything. Helpers ship with TTL jitter (±15%
  default), two-level singleflight (asyncio.Lock + Valkey SET NX EX),
  negative cache, generation counters, and per-namespace metrics.
- Persistence: RDB on, AOF off, persistent volume. Mode-A
  (lakefs:commit, lakefs:stat, lakefs:list — commit_id-keyed) survives
  restart safely. Mode-B (mutable) namespaces are flushed on every
  Valkey restart by a run_id-based bootstrap coordinator that
  serializes the flush across workers.

Includes:

- src/kohakuhub/cache.py — the helper module (319 stmts, 83% coverage
  via the new test module).
- src/kohakuhub/api/admin/routers/cache.py — admin endpoints exposing
  hit/miss/error counters, Valkey memory state, and bootstrap-flush
  metadata.
- test/kohakuhub/test_cache.py — 34 tests against a real Valkey,
  covering: round-trips, TTL jitter spread, SCAN-based prefix delete
  over >SCAN_BATCH_SIZE keys, two-level singleflight (100 concurrent
  calls fold to 1 fetch), bootstrap flush selectivity (Mode-A survives,
  Mode-B is wiped, exactly), two-worker bootstrap coordination, silent
  degradation when Valkey is disabled OR unreachable, generation
  counters, negative cache, the Mode-B prefix list shape contract.
- docker-compose.example.yml — adds the valkey service with RDB +
  LFU + bind-mounted hub-meta/valkey-data, mirroring the persistence
  pattern of the other stateful services.
- scripts/dev/up_infra.sh / down_infra.sh / reset_local_data.sh —
  Valkey container plumbing for local dev (host port 26379).
- .github/workflows/fullstack-tests.yml — adds valkey to the existing
  matrix services and adds a separate single-Python job
  (backend-tests-cache-disabled) running with KOHAKU_HUB_CACHE_ENABLED=false
  as the silent-degradation contract regression guard.
- docs/development/cache.md — the design doc referenced by the cache
  module's docstrings.

Refs: #73

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-30 19:21:04 +08:00

196 lines
7.8 KiB
YAML

# docker-compose.yml - Example configuration for KohakuHub
# Copy this to docker-compose.yml and customize for your deployment
services:
hub-ui:
image: nginx:alpine
container_name: hub-ui
restart: always
ports:
- "28080:80" # Public web interface
volumes:
- ./src/kohaku-hub-ui/dist:/usr/share/nginx/html
- ./src/kohaku-hub-admin/dist:/usr/share/nginx/html-admin
- ./docker/nginx/default.conf:/etc/nginx/conf.d/default.conf
depends_on:
- hub-api
hub-api:
build: .
container_name: hub-api
restart: always
ports:
- "48888:48888" # Internal API port (optional, for debugging)
depends_on:
- postgres
- lakefs
- minio
- valkey
environment:
## ===== CRITICAL: Endpoint Configuration (MUST CHANGE) =====
## These determine how users access your KohakuHub instance
- KOHAKU_HUB_BASE_URL=http://127.0.0.1:28080 # Change to your public URL (e.g., https://hub.example.com)
- KOHAKU_HUB_S3_PUBLIC_ENDPOINT=http://127.0.0.1:29001 # Change to your S3 public URL
## ===== CRITICAL: Security Configuration (MUST CHANGE) =====
- KOHAKU_HUB_SESSION_SECRET=change-this-to-random-string-in-production
- KOHAKU_HUB_ADMIN_SECRET_TOKEN=change-this-to-random-admin-token-in-production
- KOHAKU_HUB_DATABASE_KEY=change-this-to-random-encryption-key-in-production # For external tokens (generate with: openssl rand -hex 32)
## ===== Performance Configuration =====
- KOHAKU_HUB_WORKERS=4 # Number of worker processes (1-8, recommend: CPU cores)
## ===== Database Configuration =====
- KOHAKU_HUB_DB_BACKEND=postgres
- KOHAKU_HUB_DATABASE_URL=postgresql://<user>:<password>@postgres:5432/kohakuhub
- KOHAKU_HUB_AUTO_MIGRATE=true # Auto-confirm database migrations (required for Docker)
## ===== S3 Storage Configuration =====
- KOHAKU_HUB_S3_ENDPOINT=http://minio:9000
- KOHAKU_HUB_S3_ACCESS_KEY=minioadmin
- KOHAKU_HUB_S3_SECRET_KEY=minioadmin
- KOHAKU_HUB_S3_BUCKET=hub-storage
- KOHAKU_HUB_S3_REGION=us-east-1 # S3 region (us-east-1 for MinIO, auto for R2, or specific AWS region)
# - KOHAKU_HUB_S3_SIGNATURE_VERSION=s3v4 # Uncomment for R2/AWS S3 (leave commented for MinIO default)
## ===== LakeFS Configuration =====
- KOHAKU_HUB_LAKEFS_ENDPOINT=http://lakefs:28000
- KOHAKU_HUB_LAKEFS_REPO_NAMESPACE=hf
# LakeFS credentials auto-generated on first start
## ===== Application Configuration =====
- KOHAKU_HUB_SITE_NAME=KohakuHub
- KOHAKU_HUB_LFS_THRESHOLD_BYTES=1000000
- KOHAKU_HUB_LFS_MULTIPART_THRESHOLD_BYTES=100_000_000 # 100MB - use multipart for files larger than this
- KOHAKU_HUB_LFS_MULTIPART_CHUNK_SIZE_BYTES=50_000_000 # 50MB - size of each part (min 5MB except last)
- KOHAKU_HUB_LFS_KEEP_VERSIONS=5
- KOHAKU_HUB_LFS_AUTO_GC=true
## ===== Auth & SMTP Configuration =====
- KOHAKU_HUB_REQUIRE_EMAIL_VERIFICATION=false
- KOHAKU_HUB_INVITATION_ONLY=false # Set to true to require invitation for registration
- KOHAKU_HUB_SESSION_EXPIRE_HOURS=168
- KOHAKU_HUB_TOKEN_EXPIRE_DAYS=365
- KOHAKU_HUB_ADMIN_ENABLED=true
# SMTP (Optional - for email verification)
- KOHAKU_HUB_SMTP_ENABLED=false
- KOHAKU_HUB_SMTP_HOST=smtp.gmail.com
- KOHAKU_HUB_SMTP_PORT=587
- KOHAKU_HUB_SMTP_USERNAME=
- KOHAKU_HUB_SMTP_PASSWORD=
- KOHAKU_HUB_SMTP_FROM=noreply@kohakuhub.local
- KOHAKU_HUB_SMTP_TLS=true
## ===== Storage Quota Configuration (Optional) =====
- KOHAKU_HUB_DEFAULT_USER_PRIVATE_QUOTA_BYTES=10_000_000
- KOHAKU_HUB_DEFAULT_USER_PUBLIC_QUOTA_BYTES=100_000_000
- KOHAKU_HUB_DEFAULT_ORG_PRIVATE_QUOTA_BYTES=10_000_000
- KOHAKU_HUB_DEFAULT_ORG_PUBLIC_QUOTA_BYTES=100_000_000
## ===== Fallback Configuration (Optional) =====
# - KOHAKU_HUB_FALLBACK_ENABLED=true
# - KOHAKU_HUB_FALLBACK_REQUIRE_AUTH=false # Set true to require authentication for fallback access
# - KOHAKU_HUB_FALLBACK_SOURCES=[{"url":"https://huggingface.co","name":"HuggingFace","source_type":"huggingface","priority":1}]
## ===== L2 Cache (Valkey) =====
# See docs/development/cache.md for the design (multi-tier, write-through
# invalidation, boot-time flush of mutable namespaces).
- KOHAKU_HUB_CACHE_ENABLED=true
- KOHAKU_HUB_CACHE_URL=redis://valkey:6379/0
# - KOHAKU_HUB_CACHE_NAMESPACE=kh # Override if multiple deployments share Valkey
volumes:
- ./hub-meta/hub-api:/hub-api-creds
minio:
image: quay.io/minio/minio:latest
container_name: minio
command: server /data --console-address ":29000"
environment:
- MINIO_ROOT_USER=minioadmin
- MINIO_ROOT_PASSWORD=minioadmin
# Required so the SPA can do cross-origin Range reads on presigned
# /resolve/ targets for the pure-client safetensors/parquet preview
# (deepghs/KohakuHub#27). Tighten to a specific origin list in prod.
- MINIO_API_CORS_ALLOW_ORIGIN=*
ports:
- "29001:9000" # S3 API
- "29000:29000" # Web Console
volumes:
- ./hub-storage/minio-data:/data
- ./hub-meta/minio-data:/root/.minio
lakefs:
build:
context: ./docker/lakefs
container_name: lakefs
environment:
- LAKEFS_DATABASE_TYPE=local
- LAKEFS_DATABASE_LOCAL_PATH=/var/lakefs/data/metadata.db
- LAKEFS_BLOCKSTORE_TYPE=s3
- LAKEFS_BLOCKSTORE_S3_ENDPOINT=http://minio:9000
- LAKEFS_BLOCKSTORE_S3_BUCKET=hub-storage
- LAKEFS_BLOCKSTORE_S3_FORCE_PATH_STYLE=true
- LAKEFS_BLOCKSTORE_S3_CREDENTIALS_ACCESS_KEY_ID=minioadmin
- LAKEFS_BLOCKSTORE_S3_CREDENTIALS_SECRET_ACCESS_KEY=minioadmin
- LAKEFS_BLOCKSTORE_S3_REGION=us-east-1 # S3 region (us-east-1 for MinIO, auto for R2, or specific AWS region)
- LAKEFS_AUTH_ENCRYPT_SECRET_KEY=change-me-in-production
- LAKEFS_LOGGING_FORMAT=text
- LAKEFS_LISTEN_ADDRESS=0.0.0.0:28000
ports:
- "28000:28000" # LakeFS admin UI (optional)
user: "${UID}:${GID}"
depends_on:
- minio
volumes:
- ./hub-meta/lakefs-data:/var/lakefs/data
- ./hub-meta/lakefs-cache:/lakefs/data/cache
postgres:
image: postgres:15
container_name: postgres
restart: always
environment:
- POSTGRES_USER=<user>
- POSTGRES_PASSWORD=<password>
- POSTGRES_DB=kohakuhub
ports:
- "25432:5432" # Optional: for external access
volumes:
- ./hub-meta/postgres-data:/var/lib/postgresql/data
# L2 cache. Pure cache: losing it never compromises correctness. The
# volume + RDB persistence is intentional — it preserves the warm Mode-A
# working set (LakeFS commit_id-keyed entries) across restarts so that
# Valkey upgrades / patches don't generate a synchronized LakeFS REST
# spike. Mode-B (mutable) namespaces are flushed on every Valkey restart
# by the API's bootstrap coordinator, regardless of RDB contents — see
# MODE_B_PREFIXES in src/kohakuhub/cache.py.
valkey:
image: valkey/valkey:8-alpine
container_name: valkey
restart: always
# Internal-only by default. Add a `ports:` entry only if you need
# to attach valkey-cli from the host for debugging.
command: >-
valkey-server
--maxmemory 512mb
--maxmemory-policy allkeys-lfu
--save 300 100
--appendonly no
user: "${UID}:${GID}"
volumes:
# Same hub-meta convention as the other stateful services so all
# local persistence is rooted under one tree. Cache contents are
# recoverable from L3 — the directory is here for warm-restart
# benefits, not for backup.
- ./hub-meta/valkey-data:/data
healthcheck:
test: ["CMD", "valkey-cli", "ping"]
interval: 10s
timeout: 3s
retries: 5
networks:
default:
name: hub-net