From c8260cb6e2475d8425b9b421208b57292cb236a3 Mon Sep 17 00:00:00 2001
From: f-amine <aminefrira@gmail.com>
Date: Thu, 18 Jun 2026 15:32:30 +0100
Subject: [PATCH] fix(deploy): migrate stays healthy instead of exiting
 (Dokploy)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The migrate service was a one-shot that exited 0 after applying
migrations. Dokploy's Compose orchestration treats an exited init
container as the deploy state and doesn't reliably start services that
depend on it via `service_completed_successfully` — so web/admin never
came up and the only logs/status were migrate's "exited".

Now migrate applies migrations, writes /tmp/migrated, then idles
(`tail -f /dev/null`) so it stays a running, healthy container.
web/admin depend on its healthcheck (`test -f /tmp/migrated`) instead of
on completion. Nothing exits; dependents start once migrations are done.
On migration failure the `&&` chain aborts and `restart: unless-stopped`
retries.

Verified locally: built the migrate image, ran it against Postgres —
migrations applied, marker written, container stays Up, schema created.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 docker-compose.prod.yml   | 18 ++++++++++++------
 docker/Dockerfile.migrate | 16 ++++++++++------
 2 files changed, 22 insertions(+), 12 deletions(-)

diff --git a/docker-compose.prod.yml b/docker-compose.prod.yml
index 994f5d5..98cc64c 100644
--- a/docker-compose.prod.yml
+++ b/docker-compose.prod.yml
@@ -49,20 +49,26 @@ services:
       timeout: 5s
       retries: 5
 
-  # One-shot: applies committed Drizzle migrations (packages/db/src/migrations)
-  # before web/admin start. Generate migrations with `pnpm db:generate` and
-  # commit them — this service is a no-op if none exist yet.
+  # Applies committed Drizzle migrations (packages/db/src/migrations), then
+  # idles so it stays a healthy long-running container. web/admin wait on its
+  # healthcheck. Generate migrations with `pnpm db:generate` and commit them.
   migrate:
     build:
       context: .
       dockerfile: docker/Dockerfile.migrate
-    restart: "no"
+    restart: unless-stopped
     depends_on:
       postgres:
         condition: service_healthy
     env_file: .env
     networks:
       - internal
+    healthcheck:
+      test: ["CMD-SHELL", "test -f /tmp/migrated"]
+      interval: 5s
+      timeout: 3s
+      retries: 30
+      start_period: 5s
 
   web:
     build:
@@ -79,7 +85,7 @@ services:
       redis:
         condition: service_healthy
       migrate:
-        condition: service_completed_successfully
+        condition: service_healthy
     env_file: .env
     # No host port: Traefik (Dokploy) routes to the container over
     # dokploy-network using the port set in the Domain config. Publishing
@@ -133,7 +139,7 @@ services:
       redis:
         condition: service_healthy
       migrate:
-        condition: service_completed_successfully
+        condition: service_healthy
     env_file: .env
     expose:
       - "3002"
diff --git a/docker/Dockerfile.migrate b/docker/Dockerfile.migrate
index 3e679d1..7f0de98 100644
--- a/docker/Dockerfile.migrate
+++ b/docker/Dockerfile.migrate
@@ -1,9 +1,12 @@
 # syntax=docker.io/docker/dockerfile:1
-# One-shot migration runner for docker-compose.prod.yml.
-# Installs only @vibestack/db and its workspace deps, then runs
-# `drizzle-kit migrate` (applies committed SQL from packages/db/src/migrations).
-# Generate migrations locally with `pnpm db:generate` and commit them —
-# `migrate` is a no-op if the folder is empty.
+# Migration runner for docker-compose.prod.yml.
+# Installs only @vibestack/db and its workspace deps, applies committed SQL
+# from packages/db/src/migrations via `drizzle-kit migrate`, writes a marker,
+# then idles so the container stays running (and healthy) instead of exiting.
+# Some orchestrators (Dokploy) treat an exited one-shot container as a failed
+# deploy and never start the services that depend on it; here dependents wait
+# on this container's healthcheck (the marker file), not on it completing.
+# Generate migrations locally with `pnpm db:generate` and commit them.
 
 FROM node:22-alpine
 
@@ -21,4 +24,5 @@ COPY packages/config packages/config
 RUN pnpm --filter @vibestack/db... install
 
 # DATABASE_URL comes from env_file at runtime (docker-compose.prod.yml).
-CMD ["pnpm", "--filter", "@vibestack/db", "run", "db:migrate"]
+# Migrate, mark done for the healthcheck, then idle so the container stays up.
+CMD ["sh", "-c", "pnpm --filter @vibestack/db run db:migrate && touch /tmp/migrated && echo '[migrate] applied; idling' && tail -f /dev/null"]