Summary

• monitorMSTeamsProvider() returned immediately after starting the Express server, causing the channel orchestrator to treat it as "account exited" and schedule a restart — which led to an infinite restart loop with EADDRINUSE errors
• Wraps the server lifecycle in a long-lived Promise that stays pending while the HTTP server is running, rejects on server error, and resolves only on server close (graceful shutdown)
• Handles abort signal for clean teardown, including the already-aborted edge case

This matches the pattern used by the Feishu channel plugin (extensions/feishu/src/monitor.ts).

Root Cause

The channel orchestrator calls startAccount() which returns a Promise from monitorMSTeamsProvider(). The old code called expressApp.listen() and then immediately returned { app, shutdown }, resolving the Promise. The orchestrator interprets a resolved startAccount Promise as "the account has exited" and schedules a restart. Since the port is still bound from the previous server instance, the restart fails with EADDRINUSE, creating an infinite loop.

Test plan

• CI passes (oxfmt, oxlint, tsgo, vitest)
• Manual verification: MS Teams provider starts once and stays running without restart loop
• Abort signal triggers clean shutdown (server closes, Promise resolves)
• Server error rejects the Promise (propagates to orchestrator for proper error handling)

Fixes #24374

🤖 Generated with Claude Code

Greptile Summary

Fixes the MS Teams channel infinite restart loop (#24374) by wrapping the Express server lifecycle in a long-lived Promise. The old code resolved the startAccount Promise immediately after calling expressApp.listen(), causing the channel orchestrator to interpret this as "account exited" and schedule a restart — while the port was still bound, leading to EADDRINUSE errors in a loop.

The new code:

• Keeps the Promise pending while the HTTP server runs
• Rejects on server error (propagating to the orchestrator for proper backoff-based restart)
• Resolves only on server close (graceful shutdown via abort signal)
• Handles the already-aborted edge case and uses { once: true } on the abort listener

This aligns with the pattern used by the Feishu channel plugin (extensions/feishu/src/monitor.ts).

Confidence Score: 4/5

• This PR is safe to merge — it correctly fixes a critical infinite restart loop with a well-understood pattern.
• The fix is straightforward and follows the same long-lived Promise pattern used by the Feishu channel plugin. The code correctly handles the three key scenarios: normal operation (Promise stays pending), server error (reject), and graceful shutdown via abort signal (close → resolve). The only minor imprecision (both reject and resolve being called on EADDRINUSE) is harmless due to Promise settlement semantics. No new functionality is introduced, and the change is scoped to a single file.
• No files require special attention — the single changed file is well-structured and follows established patterns.

_{Last reviewed commit: 43bbc83}