[
https://issues.apache.org/jira/browse/CASSANDRA-21658?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Michael Semb Wever updated CASSANDRA-21658:
-------------------------------------------
Test and Documentation Plan:
The new ReadCommandUnknownIndexTest (1 test) passes on JDK 17, and it fails
with {{NullPointerException: Cannot read field "name" because "metadata" is
null}} when only the null test is removed.
The test serialises a read command that carries an index query plan, drops the
index, which stands in for a replica that has not seen the index yet, and then
deserialises the command and holds that it carries no index query plan.
ReadMessageTest (3 tests) and SecondaryIndexManagerTest (18 tests) pass.
Status: Patch Available (was: Open)
> A read on a replica that does not know the index yet fails with a
> NullPointerException
> --------------------------------------------------------------------------------------
>
> Key: CASSANDRA-21658
> URL: https://issues.apache.org/jira/browse/CASSANDRA-21658
> Project: Apache Cassandra
> Issue Type: Bug
> Components: Feature/2i Index
> Reporter: Michael Semb Wever
> Priority: Normal
> Fix For: 5.0.x, 6.0.x, 7.x
>
>
> {{ReadCommand.Serializer.deserializeIndexMetadata}} returns null when the
> index the coordinator names is unknown to this replica, and says so:
> {code:java}
> logger.info("Couldn't find a defined index on {}.{} with the id {}. " +
> "If an index was just created, this is likely due to the schema
> not " +
> "being fully propagated. Local read will proceed without using
> the " +
> "index. ...");
> return null;
> {code}
> The caller passes that null straight on:
> {code:java}
> IndexMetadata index = deserializeIndexMetadata(in, version, metadata);
> Index.Group indexGroup =
> Keyspace.openAndGetStore(metadata).indexManager.getIndexGroup(index);
> {code}
> {{SecondaryIndexManager.getIndexGroup}} calls {{getIndex}}, which reads
> {{metadata.name}}, so the local read does not proceed at all: deserialization
> throws a NullPointerException, the message is dropped, and the coordinator
> waits for the read to time out. The window is the schema propagation of a new
> index, so a client sees read timeouts for a short time after {{CREATE INDEX}}.
> The patch tests the index metadata for null, so the read proceeds without the
> index as the log line states, and marks {{deserializeIndexMetadata}}
> {{Nullable}} together with the two {{SecondaryIndexManager}} methods that
> require an index.
> Patch:
> [mck/upstream/read-command-unknown-index/5.0|https://github.com/thelastpickle/cassandra/tree/mck/upstream/read-command-unknown-index/5.0]
> Provenance:
> [69d5b5a300|https://github.com/datastax/cassandra/commit/69d5b5a300d2c82a83713191b4a8ed29c2dd5e04]
> by Piotr Kołaczkowski, [~cassandra-te]. That commit is the same change; this
> patch adds the regression test it lacks.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
