[CASSANDRA-13121] repair progress message breaks legacy JMX support

• Type: Bug
• Status: Resolved
• Priority: Low
• Resolution: Fixed
• Fix Version/s: 3.0.17, 3.11.5
• Component/s: Legacy/Streaming and Messaging
• Labels: jmx lhf notifications repair
• Severity: Low
• Since Version: 2.2.8
• Assignee: Patrick Bannister
• Reporter: Scott Bale

Authors

Patrick Bannister

Reviewers

Michael Semb Wever

• Created: 13/Jan/17 18:03
• Updated: 01/Aug/21 12:26
• Resolved: 14/May/18 09:21

Description

The error progress message in RepairRunnable is not compliant with the LegacyJMXProgressSupport class, which uses a regex to match on the text of a progress event. Therefore, actual failures slip through as successes if using legacy JMX for repairs.

In RepairRunnable

    protected void fireErrorAndComplete(String tag, int progressCount, int totalProgress, String message)
    {
        fireProgressEvent(tag, new ProgressEvent(ProgressEventType.ERROR, progressCount, totalProgress, message));
        fireProgressEvent(tag, new ProgressEvent(ProgressEventType.COMPLETE, progressCount, totalProgress, String.format("Repair command #%d finished with error", cmd)));
    }

Note the "Repair command #%d finished with error"
See https://github.com/apache/cassandra/blob/trunk/src/java/org/apache/cassandra/repair/RepairRunnable.java#L109

In LegacyJMXProgressSupport:

    protected static final Pattern SESSION_FAILED_MATCHER = Pattern.compile("Repair session .* for range .* failed with error .*");
    protected static final Pattern SESSION_SUCCESS_MATCHER = Pattern.compile("Repair session .* for range .* finished");

See https://github.com/apache/cassandra/blob/trunk/src/java/org/apache/cassandra/utils/progress/jmx/LegacyJMXProgressSupport.java#L38

Legacy JMX support was introduced for CASSANDRA-11430 (version 2.2.6) and the bug was introduced as part of CASSANDRA-12279 (version 2.2.8).

Issue Links

• is broken by: CASSANDRA-12279 nodetool repair hangs on non-existant table
• is related to: CASSANDRA-11530 Remove deprecated repair method in 4.0
• relates to: CASSANDRA-11430 Add legacy notifications backward-support on deprecated repair methods
• relates to: CASSANDRA-8901 Generalize progress reporting between tools and a server
• links to: GitHub Pull Request #22