Fix standalone server not catching SIGINT/SIGTERM

[baltasarblanco]

Fixes #1603

Problem

golem server run never installs a signal handler — the run path just awaits the services' JoinSet until completion. Running natively that is invisible: with no handler installed, the kernel applies SIGINT's default disposition and the process dies. But when the standalone server runs as PID 1 in a container (the Docker repro in the issue), the kernel only delivers a signal to a PID namespace's init process if init has installed a handler for it — default dispositions are never applied (pid_namespaces(7)). So SIGINT (Ctrl+C, kill -INT 1) and SIGTERM (docker stop) are silently discarded, which matches every observation in the issue thread.

Fix

Wrap the join_next() loop in ServerSubcommand::Run in a tokio::select! against a shutdown signal (SIGINT + SIGTERM via tokio::signal::unix on Unix, ctrl_c() elsewhere). On signal: log, join_set.shutdown().await, and return. Aborting the tasks drops the router task, which owns the worker executor's RunDetails; its Drop cancels the graph-wide shutdown token and stops the epoch thread — the same mechanism already documented in cli/golem/src/router.rs ("Dropping its RunDetails cancels the shutdown token that drives background services").

One file changed, no new dependencies (tokio already has the signal feature enabled workspace-wide).

Verification

Reproduced the PID 1 condition without Docker, since the bug's trigger is being the init of a PID namespace: sudo unshare --fork --pid --mount-proc ./target/debug/golem server run -vv.

Before (main):

• Normal process: kill -INT <pid> kills it instantly (kernel default action, no shutdown of any kind).
• As PID 1: the same kill -INT is silently ignored; the server keeps running and can only be stopped with SIGKILL.

After (this branch):

• Both cases shut down cleanly on SIGINT and SIGTERM, logging Received shutdown signal, stopping Golem server, with exit code 0.
• The shutdown token propagates to the background services; depending on timing you can see the scheduler log Shutdown requested, stopping scheduler background loop before the process exits.

cargo build -p golem and cargo make fix pass.

Note: on unpatched versions, docker run --init works around the issue, since tini becomes PID 1 and forwards signals.

[netlify]

✅ Deploy Preview for golemcloud canceled.

Name	Link
🔨 Latest commit	78d4a0d
🔍 Latest deploy log	https://app.netlify.com/projects/golemcloud/deploys/6abcf5908c9b480008aa9b26

[github-actions]

✅ All contributors have signed the CLA.
_{Posted by the CLA Assistant Lite bot.}

[github-actions]

Hi @baltasarblanco, thanks for your interest in contributing!

This project requires that pull request authors are vouched, and you are not in the list of vouched users.

This PR will be closed automatically. See https://github.com/golemcloud/golem/blob/main/CONTRIBUTING.md for more details.

[baltasarblanco]

I have read the CLA Document and I hereby sign the CLA