[IGNITE-3972] Continuous query events could be lost on backup node when primary leaves (atomic cache).

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.7
• Fix Version/s: 1.8
• Component/s: cache
• Assignee: Vladimir Ozerov
• Reporter: Vladimir Ozerov
• Created: 27/Sep/16 10:45
• Updated: 09/Nov/16 08:39
• Resolved: 12/Oct/16 11:49

Description

Consider the following scenario:
1) One node in topology;
2) PARTITIONED cache with 1 backup;
3) Continuous query is set on the cache;

If another node joins the cluster, it will handle some updates. If it leaves the topology, the first node must flush it's own events from backup queue thus ensuring that no events are lost.

But this doesn't happen because GridContinuousProcessor.addBackupNotification method perform lookup only on remote infos map, while handler is local and hence located in local infos map.

Issue Links

• is related to: IGNITE-4015 Continuous query events could be lost on backup node when primary leaves (transactional cache).
• is related to: IGNITE-4047 CQ skips events if autoUnsubscribe flag was set to false
• links to: GitHub Pull Request #1133