fix: only delete expired agents on success (#11940)

coadler · web-flow · commit 2fd1a726aa42 · 2024-01-30T14:11:45.000-06:00
diff --git a/coderd/tailnet.go b/coderd/tailnet.go
@@ -206,12 +206,13 @@ func (s *ServerTailnet) doExpireOldAgents(cutoff time.Duration) {
 		// If no one has connected since the cutoff and there are no active
 		// connections, remove the agent.
 		if time.Since(lastConnection) > cutoff && len(s.agentTickets[agentID]) == 0 {
-			deletedCount++
-			delete(s.agentConnectionTimes, agentID)
 			err := agentConn.UnsubscribeAgent(agentID)
 			if err != nil {
 				s.logger.Error(ctx, "unsubscribe expired agent", slog.Error(err), slog.F("agent_id", agentID))
+				continue
 			}
+			deletedCount++
+			delete(s.agentConnectionTimes, agentID)
 		}
 	}
 	s.nodesMu.Unlock()

Original file line number	Diff line number	Diff line change
`@@ -206,12 +206,13 @@ func (s *ServerTailnet) doExpireOldAgents(cutoff time.Duration) {`
`206`	`206`	`// If no one has connected since the cutoff and there are no active`
`207`	`207`	`// connections, remove the agent.`
`208`	`208`	`if time.Since(lastConnection) > cutoff && len(s.agentTickets[agentID]) == 0 {`
`209`		`- deletedCount++`
`210`		`- delete(s.agentConnectionTimes, agentID)`
`211`	`209`	`err := agentConn.UnsubscribeAgent(agentID)`
`212`	`210`	`if err != nil {`
`213`	`211`	`s.logger.Error(ctx, "unsubscribe expired agent", slog.Error(err), slog.F("agent_id", agentID))`
	`212`	`+ continue`
`214`	`213`	`}`
	`214`	`+ deletedCount++`
	`215`	`+ delete(s.agentConnectionTimes, agentID)`
`215`	`216`	`}`
`216`	`217`	`}`
`217`	`218`	`s.nodesMu.Unlock()`