fix(instance.controller): emit remove.instance even when logout fails (zombie cleanup) (#2520)

* fix(instance.controller): emit remove.instance even when logout fails

When a Baileys WebSocket dies but the in-memory `waInstances[name]`
entry still exists (a "zombie" instance), `deleteInstance()` calls
`await this.logout()` which throws "Connection Closed". The throw
causes the outer try/catch to skip the `eventEmitter.emit('remove.instance')`
call — which is the only mechanism that purges the zombie from
`waInstances`.

Result: zombies persist in memory until the entire `evo2_api`
container is restarted, affecting ALL instances on the host (not
just the broken one). Operators have no per-instance recovery path
in v2.3.x — their only option is `docker restart`, which forces
every connected user to re-scan the QR code.

Fix: wrap the inner `logout()` call in its own try/catch. Log a
warning when it fails but continue to the cleanup emit. The
in-memory entry must be removed regardless of whether logout
completed cleanly — `remove.instance` is the canonical way to
purge a stuck instance, and DB/cache cleanup happens in the same
event handler.

This makes `DELETE /instance/:name` idempotent against zombies: a
caller can always recover a single instance without nuking the
whole host.

Refs:
- #693  (instance/restart closes the session)
- #1286 (Connection Closed in v2.2.3)
- #2026 (Sync lost after reboot)
- #2027 (Loss of synchronization on reboot)

Tested in production at Rigarr (14 instances, ~25k msgs/day) by
overlaying this patch on v2.3.7 via Docker. Before: any zombie
forced a full container restart. After: per-instance cleanup
works cleanly while other vendors stay connected.

Signed-off-by: Bruno Cavalcante Sgarbi <bcsgarbi@gmail.com>

* review: address Sourcery feedback — neutral language + log error object

Per #2520 review:

1. Drop vendor-specific markers in code comment and log message
   (was '[ZOMBIE-CLEANUP]' and 'RIGARR PATCH'). Comment now describes
   the bug in upstream-friendly terms.

2. Pass the full error object to logger.warn instead of toString(),
   following the existing convention in monitor.service.ts
   ('no.connection' handler) where structured object logging is used
   to preserve diagnostic detail.

No behavior change.

---------

Signed-off-by: Bruno Cavalcante Sgarbi <bcsgarbi@gmail.com>
Co-authored-by: Bruno Sgarbi <bcsgarbi@gmail.com>

Author: inovacaoecrescimento

src/api/controllers/instance.controller.ts

@@ -458,7 +458,21 @@ export class InstanceController {
       if (this.configService.get<Chatwoot>('CHATWOOT').ENABLED) waInstances?.clearCacheChatwoot();
 
       if (instance.state === 'connecting' || instance.state === 'open') {
-        await this.logout({ instanceName });
+        try {
+          await this.logout({ instanceName });
+        } catch (error) {
+          // logout can throw "Connection Closed" when the underlying Baileys
+          // socket is already dead but waInstances[name] still exists. We
+          // must continue to the remove.instance emit below — that is the
+          // only path that purges the in-memory entry and runs cleaningUp().
+          // Without this catch, the stale entry persists until the entire
+          // process restarts.
+          this.logger.warn({
+            message: 'logout failed during deleteInstance — proceeding with cleanup',
+            instanceName,
+            error,
+          });
+        }
       }
 
       try {