----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/55440/#review161987 -----------------------------------------------------------
geode-lucene/src/main/java/org/apache/geode/cache/lucene/internal/LucenePrimaryBucketListener.java (line 39) <https://reviews.apache.org/r/55440/#comment233247> yes, this will result in the initial problem. There may be a possibility that the chunk bucket has become primary but the file bucket was in the process of becoming primary when this listener was invoked, resulting in a bucket not found exception. To avoid exceptions during rebalancing we are giving the system another chance to create the Index repo. - nabarun nag On Jan. 17, 2017, 10:09 p.m., nabarun nag wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/55440/ > ----------------------------------------------------------- > > (Updated Jan. 17, 2017, 10:09 p.m.) > > > Review request for geode, Barry Oglesby, Jason Huynh, Dan Smith, and xiaojian > zhou. > > > Repository: geode > > > Description > ------- > > Issue(Steps are using GFSH as mentioned in the ticket): > 1. Indexes are created, and documents are inserted into a persistent region. > 2. Server is restarted. > 3. Immediately displaying the lucene index stats shows no documents are > indexed > 4. Run a lucene query, which gets executed successfully. > 5. Now, displaying index stats with "list lucene index --with-stats" shows > the actual number of documents indexed. > > > Reason: > Number of documents indexed is taken from the IndexWriter, which is created > lazily when a lucene query is executed. Hence when a persistent server is > restarted or a rebalance occurs we lose the stats for number of documents > indexed until the IndexWriter is created when a lucene query is executed. > > Solution: > We attached a listener to each chunk bucket so that when a bucket becomes > primary, it initiates the process to create the IndexReposity for the bucket > which creates the IndexWriter, thus we get stat for number of documents > immediately after a rebalance or a restart occurs. > When a bucket becomes secondary, it removes its document count from the stats. > > > NOTE: IndexWriter can be created only when we have both chuck and file > primary bucket present. There can be a possibility that chuck bucket became > primary but the file bucket was still in the process of becoming primary > hence the IndexRepository creation will result in BucketNotFoundException. If > this happens we log a warning and defer the index repository to be created > lazily during query excution. Rather than shutting the system down with an > exception we give it second chance. > > > Diffs > ----- > > > geode-lucene/src/main/java/org/apache/geode/cache/lucene/internal/LuceneIndexForPartitionedRegion.java > c760726 > > geode-lucene/src/main/java/org/apache/geode/cache/lucene/internal/LucenePrimaryBucketListener.java > PRE-CREATION > > geode-lucene/src/main/java/org/apache/geode/cache/lucene/internal/repository/IndexRepositoryImpl.java > 572b5ea > > geode-lucene/src/test/java/org/apache/geode/cache/lucene/internal/LuceneIndexForPartitionedRegionTest.java > 5c0c0a0 > > geode-lucene/src/test/java/org/apache/geode/cache/lucene/internal/repository/IndexRepositoryImplJUnitTest.java > dd52258 > > Diff: https://reviews.apache.org/r/55440/diff/ > > > Testing > ------- > > 1. Precheckin > 2. Running gfsh commands to create a persistent server and restarting it and > checking the stats are correct. > 3. Running gfsh command to create two servers and shutting / restarting them > alternatively and checking if the stats remain correct after the primary > buckets are moved around. > > > Thanks, > > nabarun nag > >
