[CASSANDRA-15332] When repair is running with tracing, if a CorruptSSTableException is thrown while building Merkle Trees the DiskFailurePolicy does not get applied

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 4.0-alpha3, 4.0
• Component/s: Consistency/Repair, Observability/Tracing
• Labels: pull-request-available
• Bug Category: Degradation - Other Exception
• Severity: Normal
• Complexity: Low Hanging Fruit
• Discovered By: Unit Test
• Platform: All
• Since Version: 2.1 beta1

Source Control Link

https://github.com/apache/cassandra/commit/c98a31bef0d3071bc8ebfd358584aadf9e787fb8

Test and Documentation Plan

branch	diff	tests	pr
branch	diff	tests	pr

Changes:

1) JVMStabilityInspector checks for corruption or FSError and delegates to FileUtils for handling

2) FileUtils now implements the die behavior

3) JVMStabilityInspector now walks suppressed exceptions as well as the caused exception (edge case not in streaming where the second exception is fserror or corruption, so just added since saw suppression was common)

4) dtest updated to keep track of the number of kill attempts (though does not kill, mostly so tests can reuse the cluster)

Test case added which causes a corruption exception to be thrown from a SSTable during validation and shows that the handling makes its way to JVMStabilities kill instance method.

• Assignee: David Capwell
• Reporter: David Capwell

Authors

David Capwell

Reviewers

Jordan West, Marcus Eriksson

• Created: 21/Sep/19 01:31
• Updated: 15/May/20 08:39
• Resolved: 15/Nov/19 17:18

Description

When a repair is in the validation phase and is building MerkleTrees, if a corrupt SSTable exception is thrown the disk failure policy does not get applied

Issue Links

• links to: GitHub Pull Request #368
• links to: PR