[
https://issues.apache.org/jira/browse/CASSANDRA-21251?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Marcus Eriksson updated CASSANDRA-21251:
----------------------------------------
Reviewers: Marcus Eriksson, Marcus Eriksson
Marcus Eriksson, Marcus Eriksson (was: Marcus Eriksson)
Status: Review In Progress (was: Patch Available)
> Transient mismatching partitioner error when querying legacy 2i index
> ---------------------------------------------------------------------
>
> Key: CASSANDRA-21251
> URL: https://issues.apache.org/jira/browse/CASSANDRA-21251
> Project: Apache Cassandra
> Issue Type: Bug
> Components: Feature/2i Index, Transactional Cluster Metadata
> Reporter: Sam Tunnicliffe
> Assignee: Sam Tunnicliffe
> Priority: Normal
> Fix For: 6.x
>
> Attachments: ci_summary.html, result_details.tar.gz
>
>
> When a {{ClusterMetadata}} snapshot is applied, either during catchup from a
> peer or when the CMS is first initialized after upgrade, the
> {{TableMetadata}} of legacy 2i tables are reconstructed, using a new instance
> of {{{}LocalPartitioner{}}}.
> If this occurs concurrently with in-flight 2i queries, it can lead to a race
> which may trigger the assertion in the {{SinglePartitionReadCommand}}
> constructor to fail, resulting in a transient error to the client. There is
> no permanent effect and subsequent requests will succeed as normal.
> Example Stacktrace:
> {code:java}
> Caused by: java.lang.AssertionError
> at
> org.apache.cassandra.db.SinglePartitionReadCommand.<init>(SinglePartitionReadCommand.java:117)
> at
> org.apache.cassandra.db.SinglePartitionReadCommand.create(SinglePartitionReadCommand.java:157)
> at
> org.apache.cassandra.db.SinglePartitionReadCommand.create(SinglePartitionReadCommand.java:196)
> at
> org.apache.cassandra.db.SinglePartitionReadCommand.create(SinglePartitionReadCommand.java:232)
> at
> org.apache.cassandra.db.SinglePartitionReadCommand.create(SinglePartitionReadCommand.java:259)
> at
> org.apache.cassandra.index.internal.CassandraIndexSearcher.queryIndex$original$RNhtC7fd(CassandraIndexSearcher.java:101)
> at
> org.apache.cassandra.index.internal.CassandraIndexSearcher.queryIndex$original$RNhtC7fd$accessor$6QjMuAM9(CassandraIndexSearcher.java)
> at
> org.apache.cassandra.index.internal.CassandraIndexSearcher$auxiliary$cKsJOYZj.call(Unknown
> Source)
> at
> org.apache.cassandra.distributed.test.tcm.IndexTablePartitionerAfterForceSnapshotTest$BBInterceptor.queryIndex(IndexTablePartitionerAfterForceSnapshotTest.java:136)
> at
> org.apache.cassandra.index.internal.CassandraIndexSearcher.queryIndex(CassandraIndexSearcher.java)
> at
> org.apache.cassandra.index.internal.CassandraIndexSearcher.search(CassandraIndexSearcher.java:84)
> at
> org.apache.cassandra.db.ReadCommand.executeLocally(ReadCommand.java:495)
> at
> org.apache.cassandra.db.ReadCommandVerbHandler.doVerb(ReadCommandVerbHandler.java:72)
> ... 10 more
> {code}
> The solution is probably to simply weaken the assertion slightly to verify
> equivalence rather than object equality. We could memoize
> {{LocalPartitioner}} instances like we do with the other partitioner types,
> but that would be way more invasive and provide minimal benefit.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
