[IGNITE-7898] IgniteCachePartitionLossPolicySelfTest is flaky on TC

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.4
• Fix Version/s: 2.5
• Component/s: cache
• Labels: MakeTeamcityGreenAgain
• Assignee: Alexey Goncharuk
• Reporter: Pavel Kovalenko
• Created: 07/Mar/18 11:53
• Updated: 03/Apr/18 16:02
• Resolved: 14/Mar/18 13:52

Description

Affected tests:
testReadOnlyAll
testReadWriteSafe

Exception:

junit.framework.AssertionFailedError: Failed to find expected lost partition [exp=0, lost=[]]
    at org.apache.ignite.internal.processors.cache.distributed.IgniteCachePartitionLossPolicySelfTest.verifyCacheOps(IgniteCachePartitionLossPolicySelfTest.java:219)
    at org.apache.ignite.internal.processors.cache.distributed.IgniteCachePartitionLossPolicySelfTest.checkLostPartition(IgniteCachePartitionLossPolicySelfTest.java:166)
    at org.apache.ignite.internal.processors.cache.distributed.IgniteCachePartitionLossPolicySelfTest.testReadWriteSafe(IgniteCachePartitionLossPolicySelfTest.java:114)

The problem of failure:
After we prepare topology and shutdown the node containing lost partition we start to check it immediately on all nodes (cache.lostPartitions() method). Sometimes we invoke this method on client node where last PME is not even started and getting empty list of lost partitions because we haven't received it yet on PME.

Possible solution:
Wait for PME finishing on all nodes (including client) before start to check for lost partitions.

Issue Links

• relates to: IGNITE-5148 Unstable test IgniteCachePartitionLossPolicySelfTest testReadWriteSafe() seems to be flaky in 2.0
• links to: GitHub Pull Request #3626