#152 deploy.sh OOM-kills the image build, and ClickHouse with it: six services build the same image in parallel

[Anonymous]

Originally created by: grynn-in

What happened

Deploying konsol main on 12 Sep 2026 to the local stack (colima: 4 CPU, 8 GiB) failed at Step 2/5. bench build --app konsol → yarn build exited 137 (Killed).

The VM kernel log (colima ssh -- sudo dmesg) shows a global OOM that killed the node (vite) build processes and clickhouse-server (pid 1857171). ClickHouse's RestartCount is 3, so earlier kills happened too; restart: unless-stopped brought it back silently every time.

Root cause

deploy.sh:277:

docker compose --profile setup --profile backup build

With those profiles, six services have a build: section, and all six build the same context and Dockerfile with no shared image:: backup, configurator, dbt_init, frappe_backend, frappe_scheduler, frappe_worker → docker/frappe/Dockerfile.

When the konsol source changes, COPY konsol /tmp/konsol (Step 11/22) misses the cache for every one of them. Compose builds them concurrently, so six RUN bench get-app /tmp/konsol steps each run vite build at about 700 MB RSS, on top of the running stack. The failed log shows five Running in … containers side by side at Step 12/22.

What deploy.sh does NOT do: build the image once and reuse it.

Suggested fix

1. Immediate, verified. Serialise the build: export COMPOSE_PARALLEL_LIMIT=1 before line 277, or docker compose --parallel 1 … build (Compose 5.1.4 supports both). A redeploy with COMPOSE_PARALLEL_LIMIT=1 ran one vite build in total, because services 2–6 came from the first build's layer cache. It exited 0, and ClickHouse was not restarted (RestartCount stayed 3).
2. Structural. Build once: give frappe_backend an image: konsolidat-frappe:latest, and point the other five services at that image: with no build: section. That removes both the duplicated work and the memory spike.

Repro

On an 8 GiB colima VM with the stack running: change any file in the konsol checkout, then run KONSOL_REPO=… KONSOL_BRANCH=main ./deploy.sh. Watch for Killed / exit 137 during Step 2, and Out of memory: Killed process … (clickhouse-serv) in the VM's dmesg.

🤖 Generated with Claude Code

[Anonymous]

Ticket changed by: grynn-in

• status: open --> closed