feat: streamline presence cleanup logic and enhance stale presence deletion

app/realtime.php

@@ -567,38 +567,6 @@ $server->onWorkerStart(function (int $workerId) use ($server, $register, $stats,
     ));
     $register->get('telemetry.workerCounter')->add(1);
 
-    // Orphan sweep on pod restart -> no connections in in-memory realtime connections object.
-    // So db must be in sync and only one coroutine shall trigger this
-    if ($workerId === 0) {
-        go(function (): void {
-            $hostname = \gethostname();
-            if ($hostname === false || $hostname === '') {
-                return;
-            }
-
-            try {
-                $consoleDB = getConsoleDB();
-                $consoleDB->getAuthorization()->skip(fn () => $consoleDB->foreach('projects', function (Document $project) use ($hostname): void {
-                    try {
-                        $dbForProject = getProjectDB($project);
-                        $deletionCount = $dbForProject->deleteDocuments('presenceLogs', [
-                            Query::equal('hostname', [$hostname]),
-                            Query::equal('source', ['realtime']),
-                        ]);
-                        triggerPresenceUsage(-$deletionCount, $project);
-                    } catch (Throwable $th) {
-                        Console::error("Realtime startup orphan sweep failed for project {$project->getId()}: {$th->getMessage()}");
-                        logError($th, 'realtimeOrphanPresenceCleanup', tags: [
-                            'projectId' => $project->getId(),
-                        ]);
-                    }
-                }));
-            } catch (Throwable $th) {
-                Console::error('Realtime startup orphan sweep error: ' . $th->getMessage());
-            }
-        });
-    }
-
     $attempts = 0;
     $start = time();
 

src/Appwrite/Platform/Modules/Presences/HTTP/XList.php

@@ -95,10 +95,7 @@ class XList extends PlatformAction
 
         // should be excluded from the user provided query as user query would be used for caching only
         // otherwise cache will always miss due to the datetime now
-        $expiryFilter = Query::or([
-            Query::isNull('expiresAt'),
-            Query::greaterThan('expiresAt', DateTime::now()),
-        ]);
+        $expiryFilter = Query::greaterThan('expiresAt', DateTime::now());
 
         try {
             if ((int)$ttl > 0) {

src/Appwrite/Platform/Workers/Deletes.php

@@ -220,6 +220,7 @@ class Deletes extends Action
                 $this->deleteExpiredSessions($project, $getProjectDB);
                 $this->deleteExpiredTransactions($project, $getProjectDB);
                 $this->deleteExpiredPresences($project, $getProjectDB, $publisherForUsage);
+                $this->deleteStalePresences($project, $getProjectDB, $publisherForUsage);
                 $this->deleteOldDeployments($queueForDeletes, $project, $getProjectDB);
                 break;
             case DELETE_TYPE_REPORT:
@@ -1750,23 +1751,31 @@ class Deletes extends Action
     {
         $dbForProject = $getProjectDB($project);
 
-        // Drop a presence if either:
-        //   - its expiresAt has passed (HTTP-upserted rows always set this), or
-        //   - it was created more than 30 days ago. The upsert validator caps expiresAt at
-        //     +30 days, so nothing legitimate should outlive that. This branch catches rows
-        //     the expiresAt branch can't — chiefly realtime presences (expiresAt is null,
-        //     lifetime tied to the websocket) whose disconnect-time cleanup never ran.
         $now = DateTime::format(new \DateTime());
+
+        $deleted = $dbForProject->deleteDocuments('presenceLogs', [
+            Query::lessThan('expiresAt', $now),
+        ], onError: function (Throwable $th) {
+            // Swallow errors to avoid breaking the cleanup process
+        });
+
+        if ($deleted > 0) {
+            $usage = (new UsageContext())->addMetric(METRIC_USERS_PRESENCE, -$deleted);
+            $publisherForUsage->enqueue(new Usage(
+                project: $project,
+                metrics: $usage->getMetrics(),
+            ));
+        }
+    }
+
+    private function deleteStalePresences(Document $project, callable $getProjectDB, UsagePublisher $publisherForUsage): void
+    {
+        $dbForProject = $getProjectDB($project);
+
         $oldestAllowed = DateTime::format((new \DateTime())->sub(\DateInterval::createFromDateString('30 days')));
 
         $deleted = $dbForProject->deleteDocuments('presenceLogs', [
-            Query::or([
-                Query::and([
-                    Query::isNotNull('expiresAt'),
-                    Query::lessThan('expiresAt', $now),
-                ]),
-                Query::lessThan('$createdAt', $oldestAllowed),
-            ]),
+            Query::lessThan('$createdAt', $oldestAllowed),
         ], onError: function (Throwable $th) {
             // Swallow errors to avoid breaking the cleanup process
         });

src/Appwrite/Realtime/Message/Handlers/Presence.php

@@ -9,6 +9,7 @@ use Appwrite\Realtime\Message\Dispatcher;
 use Appwrite\Utopia\Database\Documents\User;
 use Closure;
 use Utopia\Database\Database;
+use Utopia\Database\DateTime;
 use Utopia\Database\Document;
 use Utopia\Database\Validator\Authorization;
 use Utopia\Database\Validator\Permissions;
@@ -77,8 +78,7 @@ class Presence extends Action
             'userId' => $user->getId(),
             'source' => 'realtime',
             'status' => $status,
-            // Pod identity, used by the realtime worker's startup sweep to clean up rows
-            // orphaned by a previous incarnation of this hostname (i.e. pod crash before onClose ran).
+            'expiresAt' => DateTime::format((new \DateTime())->modify('+30 days')),
             'hostname' => \gethostname() ?: null,
         ];
         if ($metadata !== null) {