What this is

The procedure for bringing up the GrowDirect Docker stacks in the correct order on either machine. There are two stacks: shared infra (always first) and app stacks (after). The mini has an additional Canary Go stack.

When to run

• At the start of any state session on the mini — Docker must be confirmed up before any dispatch work begins (Mini Docker Gate rule)
• Any time curl -s http://127.0.0.1:8003/mcp returns no response or connection refused
• After a machine restart
• After docker compose down was run for any reason

Preconditions

1. Docker Desktop is running (Mac: check the whale icon in the menu bar)
2. You are in the correct repo: git -C ~/GrowDirect rev-parse --show-toplevel should return the GrowDirect path

Canonical steps

Laptop

# 1. Shared infra (postgres, valkey, ollama, pgadmin, memory-bus)
cd ~/GrowDirect/devops && docker compose up -d

# 2. Verify memory bus is reachable
curl -s http://127.0.0.1:8003/mcp | head -1
# Expected: {"jsonrpc":"2.0"... or similar MCP response (not empty, not connection refused)

Mac mini

# 1. Shared infra
cd ~/GrowDirect/devops && docker compose up -d

# 2. Canary Go stack (compose lives in deploy/, not devops/)
cd ~/GrowDirect/CanaryGo && docker compose -f deploy/docker-compose.yml up -d

# 3. Verify memory bus reachable
curl -s http://127.0.0.1:8003/mcp | head -1
# Must return a response. If not: diagnose before proceeding with any dispatch.

Health check (both machines)

docker ps --format "table {{.Names}}\t{{.Status}}" | grep -E "postgres|valkey|ollama|memory_bus"
# All four should show "Up" — none should show "unhealthy" or "Restarting"

Verification

# Memory bus alive
curl -s http://127.0.0.1:8003/mcp | head -1
# Non-empty response = bus is up

# Postgres reachable
docker exec growdirect_postgres psql -U growdirect -d growdirect_memory -c "SELECT 1;" 
# Should return: (1 row)

# Ollama loaded (laptop only)
curl -s http://127.0.0.1:11434/api/tags | python3 -c "import sys,json; m=json.load(sys.stdin)['models']; print(f'{len(m)} models loaded')"
# Should show at least 1 model

Failure modes

Memory bus unhealthy after docker compose up - Symptom: docker ps shows growdirect_memory_bus as (unhealthy) - Cause: memory bus started before postgres was fully ready (startup race) - Fix: docker restart growdirect_memory_bus — the bus will reconnect on restart

Ollama restart kills a running seed - Symptom: seed process fails mid-run with embedding errors after Ollama restart - Cause: docker compose up -d restarted Ollama because its healthcheck changed - Fix: never restart Ollama while a seed is running; check pgrep -f seed_standalone.py first

POSTGRES_PASSWORD missing error during docker compose up - Symptom: error while interpolating services.postgres.environment: variable POSTGRES_PASSWORD is not set - Cause: .env file missing from devops/ - Fix: use docker restart <container> for individual container restarts rather than docker compose up if .env is absent; or restore .env from the project secrets store

docker compose up for Canary Go fails on mini - Symptom: no configuration file provided: not found - Cause: running from wrong directory or wrong -f flag - Fix: compose file is at CanaryGo/deploy/docker-compose.yml — always specify -f deploy/docker-compose.yml from the CanaryGo/ root

Invariants

• Shared infra stack always starts before any app stack.
• Memory bus health check must pass before any dispatch work on the mini.
• Never restart Ollama while seed_standalone.py is running.
• The mini never runs the seeder (seed_standalone.py) — Ollama is not a runtime dependency on the mini.

Related

• runbook-memory-bus-seed — procedure that depends on this stack being up
• platform-alx-vsm — ALX's operating environment; Docker is the runtime
• devops/docker-compose.yml — shared infra compose file
• CanaryGo/deploy/docker-compose.yml — Canary Go stack (mini only)