From f4af81ef02d9c7529f812e8701c2275aaa776d50 Mon Sep 17 00:00:00 2001
From: jakeaturner <jturner@cosmistack.com>
Date: Tue, 23 Jun 2026 17:40:10 +0000
Subject: [PATCH] fix: lazy-connect + retry for ioredis to avoid blocked
 startup

---
 admin/config/queue.ts | 19 +++++++++++++++++++
 install/entrypoint.sh | 23 +++++++++++++++++++++++
 2 files changed, 42 insertions(+)

diff --git a/admin/config/queue.ts b/admin/config/queue.ts
index 3c6a112..5b33d1e 100644
--- a/admin/config/queue.ts
+++ b/admin/config/queue.ts
@@ -1,4 +1,5 @@
 import env from '#start/env'
+import logger from '@adonisjs/core/services/logger'
 import { Redis } from 'ioredis'
 
 // BullMQ treats a plain `{host, port}` connection object as a recipe: every
@@ -14,6 +15,24 @@ const sharedConnection = new Redis({
   port: env.get('REDIS_PORT') ?? 6379,
   db: env.get('REDIS_DB') ?? 0,
   maxRetriesPerRequest: null,
+  // Don't open the socket at module import time. Importing this file (during
+  // `node ace migration:run`, `db:seed`, `queue:work`, or HTTP boot) otherwise
+  // races Docker's network/DNS lifecycle: on a fresh `up` the `redis` name is
+  // not yet resolvable (EAI_AGAIN), and on a recreate the embedded DNS briefly
+  // serves the previous container's IP (ECONNREFUSED to the stale address).
+  // Lazy-connecting defers the first dial until BullMQ actually needs Redis —
+  // after the entrypoint has confirmed it is reachable — so each (re)connect
+  // re-resolves the current IP instead of hammering a stale one.
+  lazyConnect: true,
+  // Bounded, backing-off retry so a transient outage doesn't busy-loop.
+  retryStrategy: (times) => Math.min(times * 200, 2000),
+})
+
+// Without an `error` listener ioredis logs the raw "[ioredis] Unhandled error
+// event" lines and, on some Node versions, an EventEmitter `error` with no
+// listener can crash the process. Route them through the app logger instead.
+sharedConnection.on('error', (err) => {
+  logger.error({ err }, 'Shared Redis connection error')
 })
 
 const queueConfig = {
diff --git a/install/entrypoint.sh b/install/entrypoint.sh
index 4361903..7375e71 100644
--- a/install/entrypoint.sh
+++ b/install/entrypoint.sh
@@ -7,6 +7,29 @@ echo "Starting entrypoint script..."
 # Ensure required storage directories exist (volume may be freshly mounted)
 mkdir -p /app/storage/logs /app/storage/kb_uploads
 
+# Wait for Redis to be reachable before booting anything that opens a BullMQ
+# connection. `depends_on: condition: service_healthy` only gates a clean
+# `up --recreate`; it is NOT re-checked on `docker compose restart` or a
+# `restart: unless-stopped` bounce, so without this the app can race Docker's
+# DNS (EAI_AGAIN) or dial a restarted Redis container's stale IP (ECONNREFUSED).
+# This isn't load-bearing since legacy installs used a mounted entrypoint script
+# that may override this script, but it's a cost-nothing check for newer installs.
+# The real check is done by the application itself, but this provides a safety net.
+REDIS_HOST="${REDIS_HOST:-redis}"
+REDIS_PORT="${REDIS_PORT:-6379}"
+echo "Waiting for Redis at ${REDIS_HOST}:${REDIS_PORT}..."
+for i in $(seq 1 60); do
+  if node -e "const net=require('net');const s=net.connect(Number(process.env.REDIS_PORT||6379),process.env.REDIS_HOST||'redis');s.on('connect',()=>{s.end();process.exit(0)});s.on('error',()=>process.exit(1));" 2>/dev/null; then
+    echo "Redis is up and running!"
+    break
+  fi
+  if [ "$i" -eq 60 ]; then
+    echo "Timed out waiting for Redis at ${REDIS_HOST}:${REDIS_PORT}" >&2
+    exit 1
+  fi
+  sleep 1
+done
+
 # Run AdonisJS migrations
 echo "Running AdonisJS migrations..."
 node ace migration:run --force