render.yaml

# yaml-language-server: $schema=https://render.com/schema/render.yaml.json

previews:
  generation: off

projects:
  - name: firecrawl
    environments:
      - name: production
        services:
          - type: web
            name: firecrawl-api
            runtime: docker
            plan: pro
            region: oregon
            # Thin Dockerfile wrapper over ghcr.io/firecrawl/firecrawl:latest.
            # Pin the upstream tag by changing the FROM line in ./Dockerfile.
            dockerfilePath: ./Dockerfile
            dockerContext: .
            healthCheckPath: /v0/health/liveness
            envVars:
              - key: PORT
                value: "3002"
              - key: HOST
                value: 0.0.0.0
              - key: USE_DB_AUTHENTICATION
                value: "false"
              - key: LOGGING_LEVEL
                value: info
              # Reduced from upstream defaults to fit the 4 GB pro plan.
              # Bump these (and the plan) for higher throughput.
              - key: NUQ_WORKER_COUNT
                value: "2"
              - key: NUM_WORKERS_PER_QUEUE
                value: "4"
              - key: CRAWL_CONCURRENT_REQUESTS
                value: "5"
              - key: MAX_CONCURRENT_JOBS
                value: "3"
              - key: BROWSER_POOL_SIZE
                value: "3"
              - key: MAX_CPU
                value: "0.8"
              - key: MAX_RAM
                value: "0.8"
              # Wired automatically to sibling services over the private network.
              # Render assigns each pserv a suffixed slug (e.g.
              # firecrawl-postgres-ab12); fromService.host returns the real
              # slug. The entrypoint.sh in the wrapper image hardcodes the
              # well-known ports (5672 AMQP, 3000 Playwright) because Render's
              # pserv port detection picks 15672 (mgmt UI) on the rabbitmq
              # image, which would make the api speak AMQP to the HTTP port.
              - key: PLAYWRIGHT_HOST
                fromService:
                  type: pserv
                  name: firecrawl-playwright
                  property: host
              - key: NUQ_RABBITMQ_HOST
                fromService:
                  type: pserv
                  name: firecrawl-rabbitmq
                  property: host
              - key: REDIS_URL
                fromService:
                  type: keyvalue
                  name: firecrawl-redis
                  property: connectionString
              - key: REDIS_RATE_LIMIT_URL
                fromService:
                  type: keyvalue
                  name: firecrawl-redis
                  property: connectionString
              - key: POSTGRES_HOST
                fromService:
                  type: pserv
                  name: firecrawl-postgres
                  property: host
              - key: POSTGRES_PORT
                value: "5432"
              - key: POSTGRES_DB
                fromService:
                  type: pserv
                  name: firecrawl-postgres
                  envVarKey: POSTGRES_DB
              - key: POSTGRES_USER
                fromService:
                  type: pserv
                  name: firecrawl-postgres
                  envVarKey: POSTGRES_USER
              - key: POSTGRES_PASSWORD
                fromService:
                  type: pserv
                  name: firecrawl-postgres
                  envVarKey: POSTGRES_PASSWORD
              # Protects the Bull admin UI at /admin/$BULL_AUTH_KEY/queues.
              # Render generates this once; do not rotate without redeploying.
              - key: BULL_AUTH_KEY
                generateValue: true
              # Optional: paste an OpenAI key in the dashboard to enable JSON
              # extraction and the /extract endpoint. Leave blank to disable.
              - key: OPENAI_API_KEY
                sync: false
              # Optional outbound proxy. Leave blank if you don't use one.
              - key: PROXY_SERVER
                sync: false
              - key: PROXY_USERNAME
                sync: false
              - key: PROXY_PASSWORD
                sync: false
              # Optional SearXNG endpoint to replace Google search.
              - key: SEARXNG_ENDPOINT
                sync: false

          - type: pserv
            name: firecrawl-playwright
            runtime: image
            plan: standard
            region: oregon
            image:
              url: ghcr.io/firecrawl/playwright-service:latest
            envVars:
              - key: PORT
                value: "3000"
              - key: MAX_CONCURRENT_PAGES
                value: "5"
              - key: BLOCK_MEDIA
                value: "true"
              - key: PROXY_SERVER
                sync: false
              - key: PROXY_USERNAME
                sync: false
              - key: PROXY_PASSWORD
                sync: false

          - type: pserv
            name: firecrawl-postgres
            runtime: image
            plan: standard
            region: oregon
            image:
              url: ghcr.io/firecrawl/nuq-postgres:latest
            disk:
              name: postgres-data
              mountPath: /var/lib/postgresql/data
              sizeGB: 10
            envVars:
              # POSTGRES_DB stays "postgres" because the nuq-postgres image
              # pins cron.database_name = 'postgres' in postgresql.conf.
              # Changing this without rebuilding the image breaks pg_cron.
              - key: POSTGRES_DB
                value: postgres
              - key: POSTGRES_USER
                value: firecrawl
              - key: POSTGRES_PASSWORD
                generateValue: true
              # Postgres refuses to initdb into a mountpoint that already has
              # files (e.g. lost+found on a fresh disk). Use a subdir.
              - key: PGDATA
                value: /var/lib/postgresql/data/pgdata

          - type: pserv
            name: firecrawl-rabbitmq
            runtime: image
            plan: starter
            region: oregon
            image:
              url: docker.io/library/rabbitmq:3-management
            # No disk on purpose. RabbitMQ here is only the pg_notify
            # transport for real-time worker notifications; the
            # authoritative queue state lives in firecrawl-postgres
            # (the nuq.* tables). Losing rabbit's in-memory state on
            # restart just makes workers re-poll Postgres on the next
            # tick. Mounting a Render disk at /var/lib/rabbitmq breaks
            # the Erlang cookie file's owner-only permission check on
            # subsequent boots, so we keep rabbit ephemeral.

          - type: keyvalue
            name: firecrawl-redis
            plan: starter
            region: oregon
            maxmemoryPolicy: noeviction
            ipAllowList: []