[jira] [Created] (IGNITE-5073) Race between partition exchange process and client cache operations

["Semen Boikov (JIRA)" <ji...@apache.org>]

Semen Boikov created IGNITE-5073:
------------------------------------

             Summary: Race between partition exchange process and client cache operations
                 Key: IGNITE-5073
                 URL: https://issues.apache.org/jira/browse/IGNITE-5073
             Project: Ignite
          Issue Type: Bug
          Components: cache
            Reporter: Semen Boikov
            Assignee: Semen Boikov
            Priority: Critical
             Fix For: 2.1


Added test reproducing issue IgniteCacheClientMultiNodeUpdateTopologyLockTest:
- 3 servers (node1, node2, node3), 1 client
- client starts pessimistic tx
- client locks key1 on node2
- new node joins, exchanges starts, on node3 there are no ongoging cache operations and node3 sends GridDhtPartitionsSingleMessage to coordinator
- client locks key2 on node3
- client commits tx, when tx started on node2 finishes then node2 will send GridDhtPartitionsSingleMessage to coordinator and exchange will be completed before tx on node3 finished

One potential fix for this issue is change exchange protocol to use two steps.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)