Ariel Weisberg created CASSANDRA-21726:
------------------------------------------
Summary: CEP-45: Fix mutation tracking range read bugs
Key: CASSANDRA-21726
URL: https://issues.apache.org/jira/browse/CASSANDRA-21726
Project: Apache Cassandra
Issue Type: Bug
Components: Consistency/Coordination
Reporter: Ariel Weisberg
Assignee: Ariel Weisberg
* Fix filtered range reads timing out when reconciliation delivers a mutation
for a partition the row filter dropped
* Fix filtered range reads losing partitions that reconciliation made match,
across pages and limits
* Fix filtered range reads returning too few rows, or ending paging early, when
rows the filter rejects count against the limit
* Fix range reads timing out when a limit is reached inside an open range
tombstone
* Fix indexed range reads timing out when reconciliation delivers a key past
the end of the index scan
* Fix tracked reads timing out when a mutation they pull arrives as a duplicate
* Fix range reads missing partitions where a node is a full replica of one
range and a witness of the next
* Fix follow up range reads missing data when the remaining range spans more
than one replica set
* Fix tracked reads on legacy secondary indexes hanging or returning a
partition twice when reconciliation delivers a static row
* Fix single partition queries on a static column index returning no rows when
they restrict a clustering column
* Fix partition key indexes missing a partition whose static row became live in
a later update
* Fix failed or abandoned tracked reads leaking, or releasing twice, the read
OpOrder group that flush and compaction wait on
* Fix partition iterator concatenation not closing inputs it never consumed
* Fix an offsets broadcast for a just dropped keyspace killing the stage that
processes it
* Add \{{TrackedRangeReadTest}}, which checks tracked range reads against an
untracked keyspace holding the same data, replacing the ignored
\{{MutationTrackingRangeReadTest}}
* Run range reads and legacy secondary index queries in the randomized multi
node tests for mutation tracking and witnesses
(\{{MultiNodeTableWalkWithMutationTrackingTest}} and
\{{MultiNodeTableWalkWithWitnessesTest}})
Not fixed here: a range read with a per partition limit can still fail with
\{{Multiple partitions received}} when its response holds the same partition
twice, so those randomized tests use a per partition limit only on single
partition reads. Some witness issues are also not addressed although one
witness bug fix is included.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
