[IGNITE-11204] Merged partitions exchange future ignores NODE_LEFT/FAILED events for merged exchanges

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 2.8
• Assignee: Alexey Goncharuk
• Reporter: Alexey Goncharuk
• Created: 05/Feb/19 12:29
• Updated: 11/Apr/19 17:17
• Resolved: 06/Feb/19 11:57

Description

In GridDhtPartitionsExchangeFuture#onNodeLeft we have the following code:

if (!srvNodes.remove(node)
    return;

The issue is that the srvNodes collection is created when partition exchange future is initialized. After the exchange future is merged, we will wait for more nodes to respond. However, since those nodes never added to srvNodes, the event will never be processed and the exchange future will hang.

Issue Links

• relates to: IGNITE-11732 Multi-merged partitions exchange future may hang if node left event is received last
• links to: GitHub Pull Request #6028