docker-compose.yml

# Compose project name.
name: private-channel

# Build-arg anchor for Solana toolchain consumers.
# Values come from versions.env — invoke compose with
# `--env-file versions.env --env-file .env` (later files win on conflict).
x-solana-build-args: &solana-build-args
  SOLANA_VERSION: ${SOLANA_VERSION}
  PNPM_VERSION: ${PNPM_VERSION}

x-validator-build-args: &validator-build-args
  SOLANA_VERSION: ${SOLANA_VERSION}
  YELLOWSTONE_TAG: ${YELLOWSTONE_TAG}

# Default healthcheck timings — 10s interval keeps probe load light, 3s timeout
# fails fast on a wedged service, 5 retries gives ~50s of hysteresis to absorb
# transients. start_period overrides per service below where boot is slow.
x-default-healthcheck: &default-healthcheck
  interval: 10s
  timeout: 3s
  retries: 5
  start_period: 30s

# Default logging policy — rotates each service's stdout/stderr at 100MB per file
# with 5 files retained (=~500MB ceiling per container). Lives in compose so the
# policy travels with the repo to any deploy target, instead of relying on the
# host's daemon.json.
x-default-logging: &default-logging
  driver: json-file
  options:
    max-size: "100m"
    max-file: "5"

# Shared image+build for every Rust service that compiles from the main Dockerfile.
# All such services share one byte-identical image; per-service tags would be
# 9 copies of the same artifact differing only in compose metadata. Each service
# overrides only `command:` (and per-service env). Pin in CI via
# PRIVATE_CHANNEL_VERSION; defaults to `latest` for local dev. Validator, prometheus,
# grafana and admin-ui use their own Dockerfiles and keep their own build blocks.
x-private-channel-app: &private-channel-app
  image: private-channel-app:${PRIVATE_CHANNEL_VERSION:-latest}
  build:
    context: .
    dockerfile: Dockerfile
    args: *solana-build-args

services:
  # PostgreSQL Primary (for write node)
  postgres-primary:
    image: postgres:16-alpine
    container_name: private-channel-postgres-primary
    restart: unless-stopped
    logging: *default-logging
    environment:
      - POSTGRES_DB=${POSTGRES_DB}
      - POSTGRES_USER=${POSTGRES_USER}
      - POSTGRES_PASSWORD=${POSTGRES_PASSWORD}
      - POSTGRES_REPLICATION_MODE=master
      - POSTGRES_REPLICATION_USER=${POSTGRES_REPLICATION_USER}
      - POSTGRES_REPLICATION_PASSWORD=${POSTGRES_REPLICATION_PASSWORD}
      # Enable replication - allow connections from Docker network
      - POSTGRES_INITDB_ARGS=--auth-host=scram-sha-256
      - POSTGRES_HOST_AUTH_METHOD=scram-sha-256
    ports:
      - "5432:5432"
    volumes:
      - postgres-primary-data:/var/lib/postgresql/data
      - postgres-primary-wal-archive:/wal_archive
      - ./init-primary.sql:/docker-entrypoint-initdb.d/01-init.sql:ro
      - ./init-primary-hba.sh:/docker-entrypoint-initdb.d/02-hba.sh:ro
    # `synchronous_commit` defaults to `on` (full crash durability). The bench
    # profile in bench-tps/.env explicitly opts into `off` for throughput;
    # operators who want the same trade-off in production must do so deliberately.
    command: |
      postgres
      -c wal_level=replica
      -c hot_standby=on
      -c max_wal_senders=10
      -c max_replication_slots=10
      -c hot_standby_feedback=on
      -c wal_log_hints=on
      -c archive_mode=on
      -c archive_command='cp %p /wal_archive/%f.tmp && mv /wal_archive/%f.tmp /wal_archive/%f'
      -c listen_addresses='*'
      -c shared_buffers=${PRIVATE_CHANNEL_PG_SHARED_BUFFERS:-4GB}
      -c synchronous_commit=${PRIVATE_CHANNEL_PG_SYNCHRONOUS_COMMIT:-on}
      -c wal_compression=${PRIVATE_CHANNEL_PG_WAL_COMPRESSION:-lz4}
      -c max_wal_size=${PRIVATE_CHANNEL_PG_MAX_WAL_SIZE:-4GB}
    networks:
      - private-channel-network
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U ${POSTGRES_USER}"]
      interval: 10s
      timeout: 5s
      retries: 5

  # Write Node (formerly sequencer)
  write-node:
    <<: *private-channel-app
    container_name: private-channel-write-node
    restart: unless-stopped
    logging: *default-logging
    ulimits:
      nofile:
        soft: 65536
        hard: 65536
    depends_on:
      postgres-primary:
        condition: service_healthy
    environment:
      - RUST_LOG=info
      - PRIVATE_CHANNEL_PORT=${PRIVATE_CHANNEL_WRITE_PORT}
      - PRIVATE_CHANNEL_ACCOUNTSDB_CONNECTION_URL=postgres://${POSTGRES_USER}:${POSTGRES_PASSWORD}@postgres-primary:5432/${POSTGRES_DB}
      - PRIVATE_CHANNEL_ENABLE_READ=false
      - PRIVATE_CHANNEL_SIGVERIFY_QUEUE_SIZE=${PRIVATE_CHANNEL_SIGVERIFY_QUEUE_SIZE}
      - PRIVATE_CHANNEL_SIGVERIFY_WORKERS=${PRIVATE_CHANNEL_SIGVERIFY_WORKERS}
      - PRIVATE_CHANNEL_MAX_CONNECTIONS=${PRIVATE_CHANNEL_WRITE_MAX_CONNECTIONS}
      - PRIVATE_CHANNEL_PG_MAX_CONNECTIONS=${PRIVATE_CHANNEL_PG_MAX_CONNECTIONS:-32}
      - PRIVATE_CHANNEL_MAX_TX_PER_BATCH=${PRIVATE_CHANNEL_MAX_TX_PER_BATCH}
      - PRIVATE_CHANNEL_BATCH_DEADLINE_MS=${PRIVATE_CHANNEL_BATCH_DEADLINE_MS:-10}
      - PRIVATE_CHANNEL_BATCH_CHANNEL_CAPACITY=${PRIVATE_CHANNEL_BATCH_CHANNEL_CAPACITY:-16}
      - PRIVATE_CHANNEL_MAX_SVM_WORKERS=${PRIVATE_CHANNEL_MAX_SVM_WORKERS:-8}
      - PRIVATE_CHANNEL_LOG_LEVEL=info
      - PRIVATE_CHANNEL_JSON_LOGS=false
      - PRIVATE_CHANNEL_MODE=write
      - PRIVATE_CHANNEL_ADMIN_KEYS=${PRIVATE_CHANNEL_ADMIN_KEYS}
      - PRIVATE_CHANNEL_METRICS=${PRIVATE_CHANNEL_METRICS:-false}
      - PRIVATE_CHANNEL_METRICS_PORT=${PRIVATE_CHANNEL_METRICS_PORT:-9090}
    ports:
      - "${PRIVATE_CHANNEL_WRITE_PORT}:${PRIVATE_CHANNEL_WRITE_PORT}"
    command: ["/usr/local/bin/private-channel-node"]
    networks:
      - private-channel-network
    # Real probe — /health dispatches getEpochSchedule through the in-process RPC
    # module and returns 503 if the call fails. Fails when the RPC pipeline is dead.
    healthcheck:
      <<: *default-healthcheck
      test: ["CMD-SHELL", "curl -fsS http://localhost:${PRIVATE_CHANNEL_WRITE_PORT}/health || exit 1"]

  # PostgreSQL Replica (for read node)
  postgres-replica:
    image: postgres:16-alpine
    container_name: private-channel-postgres-replica
    restart: unless-stopped
    logging: *default-logging
    depends_on:
      postgres-primary:
        condition: service_healthy
    environment:
      - POSTGRES_DB=${POSTGRES_DB}
      - POSTGRES_USER=${POSTGRES_USER}
      - POSTGRES_PASSWORD=${POSTGRES_PASSWORD}
      - POSTGRES_REPLICATION_MODE=slave
      - POSTGRES_MASTER_HOST=postgres-primary
      - POSTGRES_MASTER_PORT=5432
      - POSTGRES_REPLICATION_USER=${POSTGRES_REPLICATION_USER}
      - POSTGRES_REPLICATION_PASSWORD=${POSTGRES_REPLICATION_PASSWORD}
      - PGUSER=${POSTGRES_USER}
    ports:
      - "5433:5432"
    volumes:
      - postgres-replica-data:/var/lib/postgresql/data
    user: postgres
    command: |
      bash -c "
        until PGPASSWORD=${POSTGRES_PASSWORD} psql -h postgres-primary -U ${POSTGRES_USER} -d ${POSTGRES_DB} -c 'SELECT 1' > /dev/null 2>&1; do
          echo 'Waiting for primary to be ready...'
          sleep 2
        done

        if [ ! -f /var/lib/postgresql/data/PG_VERSION ]; then
          echo 'Setting up replica...'
          rm -rf /var/lib/postgresql/data/*
          PGPASSWORD=${POSTGRES_REPLICATION_PASSWORD} pg_basebackup -h postgres-primary -D /var/lib/postgresql/data -U ${POSTGRES_REPLICATION_USER} -v -P -R -S replica_slot

          chmod 700 /var/lib/postgresql/data
          touch /var/lib/postgresql/data/standby.signal
        fi

        postgres
      "
    networks:
      - private-channel-network
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U ${POSTGRES_USER}"]
      interval: 10s
      timeout: 5s
      retries: 5

  # Read Node
  read-node:
    <<: *private-channel-app
    container_name: private-channel-read-node
    restart: unless-stopped
    logging: *default-logging
    ulimits:
      nofile:
        soft: 65536
        hard: 65536
    depends_on:
      postgres-replica:
        condition: service_healthy
      write-node:
        condition: service_started
    environment:
      - RUST_LOG=info
      - PRIVATE_CHANNEL_PORT=${PRIVATE_CHANNEL_READ_PORT}
      - PRIVATE_CHANNEL_ACCOUNTSDB_CONNECTION_URL=postgres://${POSTGRES_USER}:${POSTGRES_PASSWORD}@postgres-replica:5432/${POSTGRES_DB}
      - PRIVATE_CHANNEL_MAX_CONNECTIONS=${PRIVATE_CHANNEL_READ_MAX_CONNECTIONS}
      - PRIVATE_CHANNEL_PG_MAX_CONNECTIONS=${PRIVATE_CHANNEL_PG_MAX_CONNECTIONS:-32}
      - PRIVATE_CHANNEL_ENABLE_WRITE=false
      - PRIVATE_CHANNEL_ENABLE_READ=true
      - PRIVATE_CHANNEL_LOG_LEVEL=info
      - PRIVATE_CHANNEL_JSON_LOGS=false
      - PRIVATE_CHANNEL_MODE=read
    ports:
      - "${PRIVATE_CHANNEL_READ_PORT}:${PRIVATE_CHANNEL_READ_PORT}"
    command: ["/usr/local/bin/private-channel-node"]
    networks:
      - private-channel-network
    # Real probe — same /health implementation as write-node, against the replica pool.
    healthcheck:
      <<: *default-healthcheck
      test: ["CMD-SHELL", "curl -fsS http://localhost:${PRIVATE_CHANNEL_READ_PORT}/health || exit 1"]

  # Gateway (Routes requests between write and read nodes)
  gateway:
    <<: *private-channel-app
    container_name: private-channel-gateway
    restart: unless-stopped
    logging: *default-logging
    ulimits:
      nofile:
        soft: 65536
        hard: 65536
    depends_on:
      - write-node
      - read-node
    environment:
      - RUST_LOG=info
      - GATEWAY_PORT=${GATEWAY_PORT}
      - GATEWAY_WRITE_URL=${GATEWAY_WRITE_URL}
      - GATEWAY_READ_URL=${GATEWAY_READ_URL}
      - METRICS_PORT=9101
      # Optional — omit or leave blank to disable RBAC enforcement.
      - JWT_SECRET=${JWT_SECRET:-}
      - AUTH_DATABASE_URL=postgres://${POSTGRES_USER}:${POSTGRES_PASSWORD}@postgres-primary:5432/${POSTGRES_DB}
      - AUTH_DATABASE_MAX_CONNECTIONS=${AUTH_DATABASE_MAX_CONNECTIONS:-10}
    ports:
      - "${GATEWAY_PORT}:${GATEWAY_PORT}"
    command: ["/usr/local/bin/gateway"]
    networks:
      - private-channel-network
    # Liveness only — /health returns static 200. Gateway deliberately
    # doesn't probe its upstreams; their own probes report their state, and
    # restarting the gateway when an upstream is down doesn't fix anything.
    # /ready (separate endpoint, probes both backends, cached 2s) exists for
    # out-of-band monitoring; compose stays on /health.
    healthcheck:
      <<: *default-healthcheck
      test: ["CMD-SHELL", "curl -fsS http://localhost:${GATEWAY_PORT}/health || exit 1"]

  # Auth Service (optional — run with: docker compose --profile auth up)
  auth:
    profiles: [auth]
    <<: *private-channel-app
    container_name: private-channel-auth
    restart: unless-stopped
    logging: *default-logging
    depends_on:
      postgres-primary:
        condition: service_healthy
    environment:
      - RUST_LOG=info
      - AUTH_PORT=${AUTH_PORT}
      - AUTH_DATABASE_URL=postgres://${POSTGRES_USER}:${POSTGRES_PASSWORD}@postgres-primary:5432/${POSTGRES_DB}
      - JWT_SECRET=${JWT_SECRET}
      - CORS_ALLOWED_ORIGIN=${CORS_ALLOWED_ORIGIN:-*}
      - AUTH_DATABASE_MAX_CONNECTIONS=${AUTH_DATABASE_MAX_CONNECTIONS:-10}
    ports:
      - "${AUTH_PORT}:${AUTH_PORT}"
    command: ["/usr/local/bin/auth"]
    networks:
      - private-channel-network
    # /health reads a pool-status flag updated by handlers after each DB call.
    # Communication errors flip it unhealthy; logical errors (UNIQUE etc.)
    # leave it healthy. The probe itself doesn't take a connection from the
    # pool, so probe load can't contribute to pool exhaustion.
    healthcheck:
      <<: *default-healthcheck
      test: ["CMD-SHELL", "curl -fsS http://localhost:${AUTH_PORT}/health || exit 1"]

  # WebSocket Transaction Streamer
  streamer:
    <<: *private-channel-app
    container_name: private-channel-streamer
    restart: unless-stopped
    logging: *default-logging
    depends_on:
      postgres-replica:
        condition: service_healthy
      write-node:
        condition: service_healthy
      postgres-indexer:
        condition: service_healthy
    environment:
      - RUST_LOG=info
      - STREAMER_PORT=${STREAMER_PORT:-8902}
      - STREAMER_ACCOUNTSDB_CONNECTION_URL=postgres://${POSTGRES_USER}:${POSTGRES_PASSWORD}@postgres-replica:5432/${POSTGRES_DB}
      - STREAMER_DATABASE_URL=postgres://${POSTGRES_USER}:${POSTGRES_PASSWORD}@postgres-indexer:5432/${POSTGRES_INDEXER_DB}
      - STREAMER_POLL_INTERVAL_MS=200
      - STREAMER_LOG_LEVEL=info
    ports:
      - "${STREAMER_PORT:-8902}:${STREAMER_PORT:-8902}"
    command: ["/usr/local/bin/streamer"]
    networks:
      - private-channel-network
    # /health checks each running poll loop's heartbeat — flips 503 if any
    # has gone stale (>30s). Empty queries still bump the heartbeat, so a
    # quiet DB stays healthy; only a truly stuck loop trips the probe.
    healthcheck:
      <<: *default-healthcheck
      test: ["CMD-SHELL", "curl -fsS http://localhost:${STREAMER_PORT:-8902}/health || exit 1"]

  # Solana Validator with Yellowstone gRPC Plugin
  validator:
    build:
      context: .
      dockerfile: validator.Dockerfile
      args: *validator-build-args
    container_name: private-channel-validator
    restart: unless-stopped
    logging: *default-logging
    deploy:
      replicas: 1
    # Agave 3.1 asserts io_uring_supported(); Docker's default seccomp profile
    # blocks io_uring_setup, so the validator panics at startup without this.
    security_opt:
      - seccomp=unconfined
    ports:
      - "18899:8899"
      - "18900:8900"
      - "10000:10000"
      - "18999:8999"
    volumes:
      - validator-data:/validator-data
      - ./yellowstone-config.json:/config/yellowstone-config.json:ro
      - ./target/deploy:/programs:ro
    environment:
      - RUST_LOG=info
      - ESCROW_PROGRAM_ID=${ESCROW_PROGRAM_ID}
      - WITHDRAW_PROGRAM_ID=${WITHDRAW_PROGRAM_ID}
    # VALIDATOR_RESET_FLAG defaults to --reset (today's behavior: wipe ledger
    # at every container start). Set it blank to preserve /validator-data
    # across restarts so on-chain state stays consistent with Postgres rows.
    # --limit-ledger-size caps the rocksdb blockstore at ~1M shreds (~1 GB);
    # without this the test-validator default of 10000 shreds wraps too fast
    # for the indexer to keep up. AccountsDB has no explicit cap but is
    # naturally bounded by Solana Private Channels's PDA/ATA count; `make docker-clean` is
    # the floor if /validator-data ever needs reclaiming.
    command: >
      solana-test-validator
      ${VALIDATOR_RESET_FLAG---reset}
      --limit-ledger-size 1000000
      --bpf-program ${ESCROW_PROGRAM_ID} /programs/private_channel_escrow_program.so
      --bpf-program ${WITHDRAW_PROGRAM_ID} /programs/private_channel_withdraw_program.so
      --clone-upgradeable-program TokenkegQfeZyiNwAJbNbGKPFXCWuBvf9Ss623VQ5DA
      --clone-upgradeable-program TokenzQdBNbLqP5VEhdkAS6EPFLC1PHnBqCXEpPxuEb
      --clone ATokenGPvbdGVxr1b2hvZbsiqW5xWH25efTNsLJA8knL
      --url https://api.mainnet-beta.solana.com
      --geyser-plugin-config /config/yellowstone-config.json
    networks:
      - private-channel-network
    healthcheck:
      test: ["CMD-SHELL", "solana cluster-version || exit 1"]
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 30s

  # PostgreSQL for Indexer
  postgres-indexer:
    image: postgres:16-alpine
    container_name: private-channel-postgres-indexer
    restart: unless-stopped
    logging: *default-logging
    environment:
      - POSTGRES_DB=${POSTGRES_INDEXER_DB}
      - POSTGRES_USER=${POSTGRES_USER}
      - POSTGRES_PASSWORD=${POSTGRES_PASSWORD}
    ports:
      - "5434:5432"
    volumes:
      - postgres-indexer-data:/var/lib/postgresql/data
      - postgres-indexer-wal-archive:/wal_archive
    command: |
      postgres
      -c wal_level=replica
      -c archive_mode=on
      -c archive_command='cp %p /wal_archive/%f.tmp && mv /wal_archive/%f.tmp /wal_archive/%f'
      -c listen_addresses='*'
    networks:
      - private-channel-network
    healthcheck:
      test:
        [
          "CMD-SHELL",
          "pg_isready -U ${POSTGRES_USER} -d ${POSTGRES_INDEXER_DB}",
        ]
      interval: 10s
      timeout: 5s
      retries: 5

  # Indexer Service - Solana (Yellowstone)
  indexer-solana:
    <<: *private-channel-app
    container_name: private-channel-indexer-solana
    restart: unless-stopped
    logging: *default-logging
    # tini as PID 1 so SIGTERM reaches the binary through the /bin/sh -c wrapper.
    init: true
    deploy:
      replicas: 1
    depends_on:
      postgres-indexer:
        condition: service_healthy
      validator:
        condition: service_healthy
    environment:
      - RUST_LOG=info,private_channel_indexer=debug
      - DATABASE_URL=postgres://${POSTGRES_USER}:${POSTGRES_PASSWORD}@postgres-indexer:5432/${POSTGRES_INDEXER_DB}
      - COMMON_RPC_URL=${LOCAL_VALIDATOR_RPC_URL}
      - INDEXER_YELLOWSTONE_ENDPOINT=http://validator:10000
      - METRICS_PORT=9100
      - ALERT_WEBHOOK_URL=${ALERT_WEBHOOK_URL:-http://localhost:9/no-op}
      # Injected by bench-tps/scripts/run.sh, the full-stack orchestration script that
      # derives the escrow instance PDA and writes it to .env before starting services.
      # When absent (e.g. manual `docker compose up`), the TOML config's
      # escrow_instance_id is used — update that value to match your deployment.
      - COMMON_ESCROW_INSTANCE_ID
    volumes:
      - ./indexer/config:/config:ro
    command:
      - /bin/sh
      - -c
      - |
        /usr/local/bin/indexer \
          --config /config/local/indexer-solana.toml \
          -v \
          indexer
    networks:
      - private-channel-network
    # /health on the metrics port reads HealthState — backed by the indexer's
    # slot-lag gauge and a heartbeat that bumps on each SlotComplete. With
    # require_continuous_progress=true, a dead Yellowstone stream flips the
    # probe even when slot-lag is 0.
    healthcheck:
      <<: *default-healthcheck
      test: ["CMD-SHELL", "curl -fsS http://localhost:9100/health || exit 1"]

  # Indexer Service - Solana Private Channels (RPC Polling)
  indexer-private-channel:
    <<: *private-channel-app
    container_name: private-channel-indexer-private-channel
    restart: unless-stopped
    logging: *default-logging
    init: true
    deploy:
      replicas: 1
    depends_on:
      postgres-indexer:
        condition: service_healthy
      # Watches Solana Private Channels write-node for WithdrawFunds instructions
      gateway:
        condition: service_started
    environment:
      - RUST_LOG=info,private_channel_indexer=debug
      - DATABASE_URL=postgres://${POSTGRES_USER}:${POSTGRES_PASSWORD}@postgres-indexer:5432/${POSTGRES_INDEXER_DB}
      # rpc_url = Solana Private Channels gateway — polls for WithdrawFunds (burn) instructions on Solana Private Channels.
      - COMMON_RPC_URL=http://gateway:${GATEWAY_PORT}
      - METRICS_PORT=9100
      - ALERT_WEBHOOK_URL=${ALERT_WEBHOOK_URL:-http://localhost:9/no-op}
    volumes:
      - ./indexer/config:/config:ro
    command:
      - /bin/sh
      - -c
      - |
        /usr/local/bin/indexer \
          --config /config/local/indexer-private-channel.toml \
          -v \
          indexer
    networks:
      - private-channel-network
    # See indexer-solana — same heartbeat-backed /health.
    healthcheck:
      <<: *default-healthcheck
      test: ["CMD-SHELL", "curl -fsS http://localhost:9100/health || exit 1"]

  # Operator Service - Solana
  operator-solana:
    <<: *private-channel-app
    container_name: private-channel-operator-solana
    restart: unless-stopped
    logging: *default-logging
    init: true
    depends_on:
      postgres-indexer:
        condition: service_healthy
      write-node:
        condition: service_healthy
    environment:
      - RUST_LOG=info,private_channel_indexer=debug
      - DATABASE_URL=postgres://${POSTGRES_USER}:${POSTGRES_PASSWORD}@postgres-indexer:5432/${POSTGRES_INDEXER_DB}
      # rpc_url = Solana Private Channels gateway (routes reads→read-node, writes→write-node)
      - COMMON_RPC_URL=http://gateway:${GATEWAY_PORT}
      # source_rpc_url = Solana Solana validator 
      - COMMON_SOURCE_RPC_URL=${LOCAL_VALIDATOR_RPC_URL}
      # Injected by bench-tps/scripts/run.sh, the full-stack orchestration script that
      # derives the escrow instance PDA and writes it to .env before starting services.
      # When absent (e.g. manual `docker compose up`), the TOML config's
      # escrow_instance_id is used — update that value to match your deployment.
      - COMMON_ESCROW_INSTANCE_ID
      - ADMIN_SIGNER=memory
      - ADMIN_PRIVATE_KEY=${ADMIN_PRIVATE_KEY}
      - OPERATOR_SIGNER=memory
      - OPERATOR_PRIVATE_KEY=${ADMIN_PRIVATE_KEY}
      - METRICS_PORT=9102
      - ALERT_WEBHOOK_URL=${ALERT_WEBHOOK_URL:-http://localhost:9/no-op}
    ports:
      # Exposed to host so bench-tps can scrape private_channel_operator_mints_sent_total
      # from outside Docker for e2e deposit measurement (BENCH_INDEXER_METRICS_URL).
      - "9102:9102"
    volumes:
      - ./indexer/config:/config:ro
    command:
      - /bin/sh
      - -c
      - |
        /usr/local/bin/indexer \
          --config /config/local/operator-solana.toml \
          -v \
          operator
    networks:
      - private-channel-network
    # /health on the metrics port reads HealthState — backlog comes from the
    # operator's pending-tx count, progress bumps as txs are forwarded to the
    # processor. Idle (pending=0) stays healthy.
    healthcheck:
      <<: *default-healthcheck
      test: ["CMD-SHELL", "curl -fsS http://localhost:9102/health || exit 1"]

  # Operator Service - Solana Private Channels
  operator-private-channel:
    <<: *private-channel-app
    container_name: private-channel-operator-private-channel
    restart: unless-stopped
    logging: *default-logging
    init: true
    depends_on:
      postgres-indexer:
        condition: service_healthy
    environment:
      - RUST_LOG=info,private_channel_indexer=debug
      - DATABASE_URL=postgres://${POSTGRES_USER}:${POSTGRES_PASSWORD}@postgres-indexer:5432/${POSTGRES_INDEXER_DB}
      # rpc_url = Solana Solana validator (where ReleaseFunds transactions are sent)
      - COMMON_RPC_URL=${LOCAL_VALIDATOR_RPC_URL}
      # source_rpc_url = Solana Private Channels gateway (source of withdrawal/burn state)
      - COMMON_SOURCE_RPC_URL=http://gateway:${GATEWAY_PORT}
      - ADMIN_SIGNER=memory
      - ADMIN_PRIVATE_KEY=${ADMIN_PRIVATE_KEY}
      - OPERATOR_SIGNER=memory
      - OPERATOR_PRIVATE_KEY=${ADMIN_PRIVATE_KEY}
      # Patched into .env by run.sh before `docker compose up` runs.
      # Passed through from the host shell — run.sh exports it via
      # `set -a; source .env; set +a` so this picks it up automatically.
      - COMMON_ESCROW_INSTANCE_ID
      - METRICS_PORT=9103
      - ALERT_WEBHOOK_URL=${ALERT_WEBHOOK_URL:-http://localhost:9/no-op}
    ports:
      # Exposed to host so bench-tps can scrape private_channel_operator_mints_sent_total
      # from outside Docker for e2e withdraw measurement (BENCH_WITHDRAW_OPERATOR_METRICS_URL).
      - "9103:9103"
    volumes:
      - ./indexer/config:/config:ro
    command:
      - /bin/sh
      - -c
      - |
        /usr/local/bin/indexer \
          --config /config/local/operator-private-channel.toml \
          -v \
          operator
    networks:
      - private-channel-network
    # /health on the metrics port reads HealthState — backlog comes from the
    # operator's pending-tx count, progress bumps as txs are forwarded to the
    # processor. Idle (pending=0) stays healthy.
    healthcheck:
      <<: *default-healthcheck
      test: ["CMD-SHELL", "curl -fsS http://localhost:9103/health || exit 1"]

  # Base backup sidecar for postgres-primary
  pg-backup-primary:
    image: postgres:16-alpine
    container_name: private-channel-pg-backup-primary
    restart: unless-stopped
    logging: *default-logging
    depends_on:
      postgres-primary:
        condition: service_healthy
    # TODO: pg_basebackup fails because POSTGRES_USER ("private_channel") lacks the
    # REPLICATION attribute and isn't in pg_hba.conf for replication. Switch
    # to POSTGRES_REPLICATION_USER/_PASSWORD (already used by the streaming
    # replica). Pre-existing gap; backups currently never complete.
    environment:
      - PGHOST=postgres-primary
      - PGUSER=${POSTGRES_USER}
      - PGPASSWORD=${POSTGRES_PASSWORD}
      - PG_BACKUP_INTERVAL_HOURS=${PG_BACKUP_INTERVAL_HOURS:-6}
      - PG_BACKUP_RETENTION_COUNT=${PG_BACKUP_RETENTION_COUNT:-3}
    volumes:
      - postgres-primary-basebackups:/backups
      - postgres-primary-wal-archive:/wal_archive
      - ./scripts/pg-backup.sh:/usr/local/bin/pg-backup.sh:ro
    command: ["/usr/local/bin/pg-backup.sh"]
    networks:
      - private-channel-network

  # Base backup sidecar for postgres-indexer
  pg-backup-indexer:
    image: postgres:16-alpine
    container_name: private-channel-pg-backup-indexer
    restart: unless-stopped
    logging: *default-logging
    depends_on:
      postgres-indexer:
        condition: service_healthy
    # TODO: same pg_hba/role mismatch as pg-backup-primary above.
    environment:
      - PGHOST=postgres-indexer
      - PGUSER=${POSTGRES_USER}
      - PGPASSWORD=${POSTGRES_PASSWORD}
      - PG_BACKUP_INTERVAL_HOURS=${PG_BACKUP_INTERVAL_HOURS:-6}
      - PG_BACKUP_RETENTION_COUNT=${PG_BACKUP_RETENTION_COUNT:-3}
    volumes:
      - postgres-indexer-basebackups:/backups
      - postgres-indexer-wal-archive:/wal_archive
      - ./scripts/pg-backup.sh:/usr/local/bin/pg-backup.sh:ro
    command: ["/usr/local/bin/pg-backup.sh"]
    networks:
      - private-channel-network

  # Blackbox Exporter for HTTP health probing
  blackbox-exporter:
    image: prom/blackbox-exporter:${BLACKBOX_VERSION}
    container_name: private-channel-blackbox-exporter
    restart: unless-stopped
    logging: *default-logging
    networks:
      - private-channel-network
    healthcheck:
      test: ["CMD-SHELL", "wget --spider -q http://localhost:9115/-/healthy || exit 1"]
      interval: 30s
      timeout: 3s
      retries: 5
      start_period: 10s

  # cAdvisor for container metrics
  cadvisor:
    image: gcr.io/cadvisor/cadvisor:v0.49.1
    container_name: private-channel-cadvisor
    restart: unless-stopped
    logging: *default-logging
    ports:
      - "8080:8080"
    volumes:
      - /:/rootfs:ro
      - /var/run:/var/run:ro
      - /sys:/sys:ro
      - /var/lib/docker/:/var/lib/docker:ro
      - /dev/disk/:/dev/disk:ro
    privileged: true
    networks:
      - private-channel-network
    # No healthcheck — the cadvisor image is distroless (no shell, no wget/curl).
    # Restart-on-crash via `restart: unless-stopped` is the only liveness signal.
    # An out-of-process probe (Prometheus scrape failure on the cadvisor target)
    # is the right monitoring path here.

  # Prometheus for metrics collection
  prometheus:
    build:
      context: .
      dockerfile: Dockerfile.prometheus
      args:
        PROMETHEUS_VERSION: ${PROMETHEUS_VERSION}
    container_name: private-channel-prometheus
    restart: unless-stopped
    logging: *default-logging
    ports:
      - "9090:9090"
    volumes:
      - prometheus-data:/prometheus
    environment:
      - HOST_SUFFIX=
      - BENCH_METRICS_TARGET=${BENCH_METRICS_TARGET:-host.docker.internal:9101}
      - PRIVATE_CHANNEL_METRICS_PORT=${PRIVATE_CHANNEL_METRICS_PORT:-9090}
    extra_hosts:
      - "host.docker.internal:host-gateway"
    networks:
      - private-channel-network
    healthcheck:
      test: ["CMD-SHELL", "wget --spider -q http://localhost:9090/-/healthy || exit 1"]
      interval: 30s
      timeout: 3s
      retries: 5
      start_period: 15s

  # Grafana for visualization. Uses Dockerfile.grafana so the image bakes in
  # provisioning + dashboards (matters for standalone deployments where bind
  # mounts aren't available). Locally the bind mounts below shadow the baked
  # content so edits hot-reload without rebuilding.
  grafana:
    build:
      context: .
      dockerfile: Dockerfile.grafana
      args:
        GRAFANA_VERSION: ${GRAFANA_VERSION}
    container_name: private-channel-grafana
    restart: unless-stopped
    logging: *default-logging
    ports:
      - "37429:3000"
    volumes:
      - grafana-data:/var/lib/grafana
      - ./monitoring/provisioning:/etc/grafana/provisioning:ro
      - ./monitoring/dashboards:/var/lib/grafana/dashboards:ro
    environment:
      - GF_SECURITY_ADMIN_PASSWORD=${GF_ADMIN_PASSWORD}
      - GF_USERS_ALLOW_SIGN_UP=false
      - POSTGRES_INDEXER_DB=${POSTGRES_INDEXER_DB}
      - POSTGRES_INDEXER_HOST=${POSTGRES_INDEXER_HOST:-postgres-indexer}
      - POSTGRES_USER=${POSTGRES_USER}
      - POSTGRES_PASSWORD=${POSTGRES_PASSWORD}
      - PROMETHEUS_URL=${PROMETHEUS_URL:-http://prometheus:9090}
      - ALERT_WEBHOOK_URL=${ALERT_WEBHOOK_URL:-http://localhost:9/no-op}
    networks:
      - private-channel-network
    healthcheck:
      test: ["CMD-SHELL", "wget --spider -q http://localhost:3000/api/health || exit 1"]
      interval: 30s
      timeout: 3s
      retries: 5
      start_period: 20s

volumes:
  postgres-primary-data:
  postgres-replica-data:
  postgres-indexer-data:
  postgres-primary-wal-archive:
  postgres-indexer-wal-archive:
  postgres-primary-basebackups:
  postgres-indexer-basebackups:
  validator-data:
  prometheus-data:
  grafana-data:

networks:
  private-channel-network:
    driver: bridge