docker-compose.yml

services:
  kafka:
    image: apache/kafka
    # The official image's own launch script defaults KAFKA_HEAP_OPTS to -Xmx1G -Xms1G when unset -
    # undocumented on the image's own page, and never sized for this container's actual budget,
    # same gap postgres/mongo close below via mem_limit + explicit tuning. mem_limit leaves ~1GB
    # of the 3g ceiling outside the heap for metaspace/direct buffers/page cache (Kafka leans on
    # OS page cache for log segment reads/writes, unlike a JVM-heap-only workload).
    environment:
      CLUSTER_ID: cluster
      KAFKA_NODE_ID: 0
      KAFKA_PROCESS_ROLES: broker,controller
      KAFKA_LISTENERS: CLIENT://:9092,CONTROLLER://localhost:9093
      KAFKA_ADVERTISED_LISTENERS: CLIENT://localhost:9092
      KAFKA_CONTROLLER_LISTENER_NAMES: CONTROLLER
      KAFKA_INTER_BROKER_LISTENER_NAME: CLIENT
      KAFKA_LISTENER_SECURITY_PROTOCOL_MAP: CONTROLLER:PLAINTEXT,CLIENT:PLAINTEXT
      KAFKA_CONTROLLER_QUORUM_VOTERS: 0@localhost:9093
      KAFKA_OFFSETS_TOPIC_REPLICATION_FACTOR: 1
      KAFKA_TRANSACTION_STATE_LOG_REPLICATION_FACTOR: 1
      KAFKA_TRANSACTION_STATE_LOG_MIN_ISR: 1
      KAFKA_HEAP_OPTS: "-Xmx2G -Xms2G"
    mem_limit: 3g
    ports:
      - 9092:9092

  postgres:
    build:
      context: .
      dockerfile: Dockerfile.postgres
    # shared_buffers/effective_cache_size/work_mem/maintenance_work_mem sized off mem_limit below
    # (25%/75%/16MB/128MB), per Docker's own Postgres tuning guide - the built-in default
    # (shared_buffers=128MB) is deliberately conservative and was never sized for this container's
    # actual memory ceiling. mem_limit is set explicitly on both postgres and mongo so the two are
    # bounded to the same memory budget instead of Mongo's WiredTiger auto-sizing off the whole VM
    # while Postgres stays pinned to its 128MB default.
    # max_connections=200 gives headroom above no-mongodb-demo's HikariCP pool (maximum-pool-size:
    # 100) - the default 100 left zero room for anything else once Hikari's pool filled it.
    command: >
      postgres -c wal_level=logical -c timezone=UTC
      -c shared_buffers=512MB -c effective_cache_size=1536MB
      -c work_mem=16MB -c maintenance_work_mem=128MB
      -c max_connections=200
    environment:
      POSTGRES_DB: demo
      POSTGRES_USER: demo
      POSTGRES_PASSWORD: demo
      LANG: en_US.utf8
    mem_limit: 2g
    ports:
      - 5432:5432

  redis:
    image: redis:alpine
    ports:
      - 6379:6379

  rabbitmq:
    image: rabbitmq:latest
    ports:
      - 5672:5672
      - 15672:15672

  mongo:
    image: mongo:7
    # Same 2g ceiling as postgres above, for a comparable memory budget between the two. mongo:7
    # reads cgroups v2 limits correctly (fixed in MongoDB SERVER-60412, well before this release),
    # so WiredTiger auto-sizes its cache to ~50% of (this limit - 1GB) without needing a manual
    # --wiredTigerCacheSizeGB override.
    mem_limit: 2g
    ports:
      - 27017:27017