Add SIGTERM handling at tracker signal boundary

[josecelano]

Parent

Part of #1488.

Goal

Handle SIGTERM in src/main.rs alongside the existing SIGINT handler so kill <pid>, docker stop, systemctl stop, and Kubernetes pod termination start the existing coordinated graceful-shutdown sequence.

Background

main.rs currently waits only for tokio::signal::ctrl_c(). SIGTERM, the default signal sent by kill, container runtimes, and service managers, bypasses main.rs; consequently, it does not invoke jobs.cancel() or jobs.wait_for_all(...). Server libraries currently retain independent legacy signal subscriptions. That temporary duplicate reaction is expected and must remain untouched in this slice.

Scope

• Register a Unix-only Tokio SIGTERM listener with tokio::signal::unix::signal(SignalKind::terminate()).
• Use one tokio::select! to determine whether SIGINT or SIGTERM requested shutdown.
• Log Torrust tracker shutting down (SIGINT) ... or Torrust tracker shutting down (SIGTERM) ..., then run the existing sequence exactly once: jobs.cancel(), jobs.wait_for_all(Duration::from_secs(10)).await, and completion logging.
• Preserve tokio::signal::ctrl_c() support on Windows without adding Windows service-control-manager work.
• Preserve the startup-error path.
• Move and maintain the local issue specification and raw verification evidence under docs/issues/open/2132-add-sigterm-to-main/.

Out of scope

• JobManager ownership, JoinSet, concurrency, deadline, or exit-result changes (Adopt JoinSet supervision in JobManager #1586 and SI-20).
• Periodic-job token migrations (SI-4 and SI-5).
• Legacy global_shutdown_signal() or server API removal (SI-2 and later).
• Axum drain, UDP lifecycle, readiness, SIGHUP, profiling, and Windows SCM work.

Acceptance criteria

• kill <tracker-binary-pid> starts the coordinated shutdown sequence and logs the SIGTERM source.
• kill -TERM <tracker-binary-pid> has the same outcome.
• SIGINT still starts the sequence and logs SIGINT, not SIGTERM.
• SIGTERM runs cancellation and named JobManager waiting after the source log.
• Unix-only signal registration is conditionally compiled; Windows retains Ctrl+C support only.
• Verification targets ./target/release/torrust-tracker directly, not cargo run.
• Verification permits the current sequential legacy per-job shutdown behavior; SI-20 owns final deadline policy.
• cargo test and linter all pass.

Verification

Build the release binary, launch ./target/release/torrust-tracker directly in the background while recording its PID and logs, wait for a deterministic startup indication, and run separate default-SIGTERM, explicit-SIGTERM, and SIGINT checks. Record complete raw terminal and log output in the local verification evidence. Do not use timeout 20s cargo run as signal-boundary evidence. SIGKILL is only a baseline/control, not a normal stop mechanism.

References

• Overhaul: Tracker Shutdown #1488
• docs/issues/open/2132-add-sigterm-to-main/ISSUE.md
• docs/issues/open/2132-add-sigterm-to-main/verification.md
• docs/features/shutdown-process/questions.md (Q1, Q5, Q7, and Q9)
• docs/adrs/20260902074438_adopt_supervised_cancellation_tree_for_shutdown.md