Unnecessary data copy on index add

[TomaszGaweda]

For the the situation when the index already exists, then there is no need to fill it again with the same data

Closes #13964

[TomaszGaweda]

@mmedenjak @taburet Can you please add me to CloudBees or paste here what the error was?

[mmedenjak]

Hi @TomaszGaweda , these are the failed tests:

com.hazelcast.client.map.ClientPagingPredicateTest.testIssue3047
com.hazelcast.client.map.ClientPagingPredicateTest.mapPagingPredicateEmployeeObjectWithOrderedIndexSmallTest
com.hazelcast.client.map.ClientPagingPredicateTest.mapPagingPredicateEmployeeObjectWithOrderedIndexLargeTest
com.hazelcast.map.PagingPredicateTest.testPagingWithFilteringAndComparatorAndIndex

I've checked the first one and it seems to fail consistently.

[mmedenjak]

Can you also please follow the instructions to sign and send the Hazelcast Contributor Agreement?

[TomaszGaweda]

@mmedenjak Sure, I will sign it, my boss just wants to check this by lawyer to be sure :) One test is broken, I probably have fix and will commit shortly

[taburet]

@TomaszGaweda the tests are failing because of the so called global indexes, which share their index definitions across partitions, so if a global index definition is present that doesn't mean a corresponding index is populated by all partitions. Apparently, to fix that we need to track a set of indexed partitions.

Also, I have discovered another flaw in the distributed map proxies implementation: looks like we are rebuilding all indexes (configured using an xml config) when a new member joins a cluster even if there are no manual calls to addIndex. See com.hazelcast.map.impl.proxy.MapProxySupport#initialize and com.hazelcast.map.impl.proxy.MapProxySupport#initializeIndexes.

If you don't mind, I will take over the issue and fix both problems.

[TomaszGaweda]

@taburet Sorry for the late response. It would be great if you can take over this issue, I'm not feeling I can easilly fix the second issue you've written. For the issue with tests, I've tried to add counter to Indexes class that checks if all partitions are migrated, seems to work, but I didn't have time to check it more than rerun test

[taburet]

Superseded by #13984, closing.