Deploy workers

You’ll deploy a worker (and optionally a beat) as long-running processes next to your web app: on gisila-panel as portless process apps, or under any process manager.

Prerequisites

A package with bin/worker.dart (and bin/beat.dart) as in Run workers and Periodic tasks.
A Redis the workers can reach, with its URL.

1. Compile to a native binary

dart compile exe bin/worker.dart -o build/worker
./build/worker --namespace=shop --queues=default --concurrency=8

A compiled binary starts fast and needs no Dart SDK on the host.

2. Deploy on gisila-panel

On the panel, a worker is an app with no port:

Create an app with the dart runtime and expose mode process (no port, no nginx, no domain; the unit is kept alive and restarted if it exits).
Set the build command so the panel compiles your worker entrypoint instead of the default bin/server.dart:
Terminal window
```
mkdir -p build && dart pub get && dart compile exe bin/worker.dart -o build/app
```
Set the start command to the binary with its flags, for example app --namespace=shop --queues=default,emails --concurrency=8.
Add an environment variable REDIS_URL (for example redis://:password@10.0.0.5:6379/0). The CLIs read it, which keeps the password out of the process list.
Deploy. Repeat with bin/beat.dart for a beat app if you have periodic tasks.

Run the web app as a separate web app from the same repository, with the same REDIS_URL and namespace. See the panel docs for creating apps, env vars and deployments.

3. Or use another process manager

Any supervisor that restarts the process and sends SIGTERM on stop works. A systemd unit:

[Service]
ExecStart=/srv/shop/worker --namespace=shop --queues=default --concurrency=8
Environment=REDIS_URL=redis://:secret@127.0.0.1:6379/0
Restart=always
TimeoutStopSec=60

In containers, run one process per container and scale the replica count.

4. Get shutdown right

On SIGTERM, WorkerCli stops reserving and waits up to --drain-timeout (30 s) for running tasks. Configure the process manager’s stop timeout longer than that. Tasks still running when the process dies are redelivered after the broker’s visibility timeout (Redis: 5 minutes by default).

During a rolling deploy, old and new workers run side by side. A task name that only the new code knows is put back with a delay by old workers rather than failed, so it reaches a new worker.

5. Run beat safely

Beat built by BeatCli or on GisilaQueue.redis uses a Redis lock, so running it on two hosts (or overlapping during a deploy) doesn’t double-fire. Keep --lock-ttl (30 s) short enough that a standby takes over quickly.

Verify it works

REDIS_URL=... dart run gisila_queue:gisila_queue inspect --namespace=shop --queues=default,emails,dead_letter

Queue depths should stay near zero under normal load and dead_letter should not grow. Check the app’s runtime logs on the panel for worker started: queues=....