Commands hang after completing — output written, process never exits

[azataiot]

Environment

• sentry CLI 0.38.0, installed via curl https://cli.sentry.dev/install | bash
• macOS (Darwin 25.5.0), arm64, zsh

Summary

Every command I tried (org list, project list, issue view, issue resolve, auth status) finishes its work and writes complete output, but the process never exits. Reproduced both with stdout redirected to a file and under a pseudo-TTY (script); one project list was still alive ~40 minutes later. This breaks scripted usage — anything waiting on the exit code stalls.

Repro

sentry org list --json --log-level debug > out.json 2> debug.log
# out.json is complete, valid JSON ~2s in; process still alive 120s later

Debug log shows the work finishing, then nothing:

[http] ⚙ GET /api/0/users/me/regions/ → 200 (371ms)
[http] ⚙ GET /api/0/organizations/ → 200 (1424ms)

The main thread is parked, not spinning — looks like an open handle keeps the event loop referenced after the command completes. Reminiscent of #782/#833, but for ordinary commands rather than the init wizard.

Workaround

Redirect output to a file, poll until complete, then kill the PID.

[sentry-pierre]

Confirmed from code inspection.

The hang is caused by Sentry SDK integrations that keep the event loop alive after the command finishes its work:

1. ProcessSession — registers an anonymous beforeExit handler with no cleanup mechanism. The test suite literally scrubs these listeners after every test because it knows they hang the Bun runner (test/lib/telemetry.test.ts:47).
2. NodeRuntimeMetrics — runs a metrics-collection interval timer. The code explicitly excludes it from library mode because runtime metrics timer would keep host event loop alive (telemetry.ts:420), yet EXCLUDED_INTEGRATIONS (used in normal CLI mode) leaves it active.
3. sendClientReports / enableLogs — the code comments at telemetry.ts:635 and :639 explicitly warn that these register setInterval + beforeExit handler that keep the event loop alive. They are disabled in library mode but enabled for ordinary commands.

cli.ts initializes the SDK via withTelemetry → initSentry on every run, but never calls Sentry.close() afterward. The custom beforeExit flush handler only fires when the event loop drains, which never happens because the SDK's own timers keep it referenced.

Fix direction: explicitly close the Sentry client and/or unref the leaking timers after command execution in cli.ts (or inside withTelemetry so the cleanup is guaranteed).