litestream-coder（莱特流编码器）

Litestream Coder

Overview

Litestream provides continuous streaming backup for SQLite databases to S3-compatible storage. Essential for Rails 8+ apps using SQLite in production with Solid Queue, Solid Cache, and Solid Cable.

Configuration File

Create config/litestream.yml:

dbs:
  - path: /rails/storage/production.sqlite3
    replicas:
      - type: s3
        bucket: ${BUCKET_NAME}
        path: sqlite/production
        endpoint: ${S3_ENDPOINT}
        region: ${S3_REGION}
        access-key-id: ${LITESTREAM_ACCESS_KEY_ID}
        secret-access-key: ${LITESTREAM_SECRET_ACCESS_KEY}
        force-path-style: true       # Required for S3-compatible storage
        sync-interval: 5s            # How often to sync WAL
        snapshot-interval: 1h        # Full snapshot frequency
        retention: 168h              # 7 days
        retention-check-interval: 1h
        validation-interval: 12h

Retention Strategies by Database Type

Primary database (production.sqlite3):

• sync-interval: 5s - Frequent syncing for data durability
• snapshot-interval: 1h - Hourly snapshots
• retention: 720h - 30 days for recovery

Cache database (production_cache.sqlite3):

• sync-interval: 10s - Less critical, reduce load
• snapshot-interval: 6h - Less frequent
• retention: 24h - 1 day sufficient

Queue database (production_queue.sqlite3):

• sync-interval: 5s - Important for job durability
• snapshot-interval: 1h - Hourly
• retention: 72h - 3 days for debugging

Cable database (production_cable.sqlite3):

• sync-interval: 10s - Ephemeral data
• snapshot-interval: 6h - Infrequent
• retention: 24h - 1 day

Complete Configuration Example

dbs:
  - path: /rails/storage/production.sqlite3
    replicas:
      - type: s3
        bucket: myapp-backups
        path: sqlite/production
        endpoint: https://fsn1.your-objectstorage.com
        region: fsn1
        access-key-id: ${LITESTREAM_ACCESS_KEY_ID}
        secret-access-key: ${LITESTREAM_SECRET_ACCESS_KEY}
        force-path-style: true
        sync-interval: 5s
        snapshot-interval: 1h
        retention: 720h
        retention-check-interval: 1h
        validation-interval: 12h

  - path: /rails/storage/production_cache.sqlite3
    replicas:
      - type: s3
        bucket: myapp-backups
        path: sqlite/cache
        endpoint: https://fsn1.your-objectstorage.com
        region: fsn1
        access-key-id: ${LITESTREAM_ACCESS_KEY_ID}
        secret-access-key: ${LITESTREAM_SECRET_ACCESS_KEY}
        force-path-style: true
        sync-interval: 10s
        snapshot-interval: 6h
        retention: 24h

  - path: /rails/storage/production_queue.sqlite3
    replicas:
      - type: s3
        bucket: myapp-backups
        path: sqlite/queue
        endpoint: https://fsn1.your-objectstorage.com
        region: fsn1
        access-key-id: ${LITESTREAM_ACCESS_KEY_ID}
        secret-access-key: ${LITESTREAM_SECRET_ACCESS_KEY}
        force-path-style: true
        sync-interval: 5s
        snapshot-interval: 1h
        retention: 72h

  - path: /rails/storage/production_cable.sqlite3
    replicas:
      - type: s3
        bucket: myapp-backups
        path: sqlite/cable
        endpoint: https://fsn1.your-objectstorage.com
        region: fsn1
        access-key-id: ${LITESTREAM_ACCESS_KEY_ID}
        secret-access-key: ${LITESTREAM_SECRET_ACCESS_KEY}
        force-path-style: true
        sync-interval: 10s
        snapshot-interval: 6h
        retention: 24h

Kamal 2 Deployment

Add Litestream as an accessory in config/deploy.yml:

accessories:
  litestream:
    image: litestream/litestream:0.3
    host: <SERVER_IP>
    cmd: replicate
    volumes:
      - "myapp_storage:/rails/storage:ro"  # Read-only access
    files:
      - config/litestream.yml:/etc/litestream.yml
    env:
      secret:
        - LITESTREAM_ACCESS_KEY_ID
        - LITESTREAM_SECRET_ACCESS_KEY
    options:
      health-cmd: "litestream databases || exit 1"
      health-interval: 30s
      health-timeout: 5s
      health-retries: 3

Key points:

• Mount storage as read-only (:ro) - Litestream only reads WAL files
• Use same volume name as main app
• Health check verifies Litestream can see databases

Secrets Configuration

In .kamal/secrets:

LITESTREAM_ACCESS_KEY_ID=$(op read "op://myproject/production-s3/access_key_id")
LITESTREAM_SECRET_ACCESS_KEY=$(op read "op://myproject/production-s3/secret_access_key")

Disaster Recovery

Restore Database

# Stop application first
kamal app stop

# Restore from latest backup
docker run --rm \
  -v myapp_storage:/rails/storage \
  -e LITESTREAM_ACCESS_KEY_ID="$(op read 'op://myproject/production-s3/access_key_id')" \
  -e LITESTREAM_SECRET_ACCESS_KEY="$(op read 'op://myproject/production-s3/secret_access_key')" \
  litestream/litestream:0.3 restore \
  -config /etc/litestream.yml \
  /rails/storage/production.sqlite3

# Restart application
kamal app start

Point-in-Time Recovery

docker run --rm \
  -v myapp_storage:/rails/storage \
  -e LITESTREAM_ACCESS_KEY_ID="..." \
  -e LITESTREAM_SECRET_ACCESS_KEY="..." \
  litestream/litestream:0.3 restore \
  -config /etc/litestream.yml \
  -timestamp "2025-01-15T10:30:00Z" \
  /rails/storage/production.sqlite3

Monitoring

Check backup status:

# Via Kamal
kamal accessory exec litestream -- litestream databases

# Direct on server
docker exec myapp-litestream litestream databases

S3-Compatible Storage

Litestream works with any S3-compatible storage:

Critical: Always set force-path-style: true for non-AWS S3-compatible storage.

VACUUM Impact on Litestream

Full VACUUM rewrites the entire database file. Litestream detects this as a complete change and triggers a full snapshot upload (not incremental WAL sync).

Operational considerations:

• VACUUM increases backup storage temporarily (old snapshot + new snapshot retained during retention window)
• Stagger VACUUM across databases to avoid bandwidth spikes: primary at 3:00, cache at 3:15, queue at 3:30
• This is expected and safe — don't disable VACUUM to avoid snapshots
• Monitor with litestream databases after VACUUM to confirm snapshot completed

See resources/database-admin/sqlite.md for VACUUM scheduling strategies.

Best Practices

• Separate databases by retention needs - Main DB needs longer retention than cache
• Monitor backup lag - litestream databases shows replication status
• Test restores regularly - Don't wait for a disaster to verify backups work
• Use read-only mount - Litestream only reads WAL files, :ro prevents accidents
• Stagger VACUUM operations - Avoid simultaneous full snapshots across databases