[CASSANDRA-17619] Expired snapshots of dropped tables are not removed after node restart

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 4.1-alpha1, 4.1, 5.0-alpha1, 5.0
• Component/s: Local/Snapshots
• Bug Category: Correctness - API / Semantic Implementation
• Severity: Normal
• Complexity: Normal
• Discovered By: User Report
• Platform: All
• Since Version: 4.0

Source Control Link

https://github.com/apache/cassandra/commit/a604c0d4defb50ffdf7f66c3ac372eb14f39ba3b

Test and Documentation Plan

unit tests

• Assignee: Paulo Motta
• Reporter: Paulo Motta

Authors

Paulo Motta

Reviewers

Stefan Miklosovic

• Created: 10/May/22 15:22
• Updated: 12/Sep/23 13:02
• Resolved: 29/May/22 19:45

Description

Expired TTL snapshots from a dropped table are only removed if the node is not restarted after the table is dropped.

The reason for this is because SnapshotManager is using the old implementation of listsnapshots to load snapshots in memory during node startup, before CASSANDRA-16843 is fixed so it doesn't keep track of snapshots of dropped tables.

We should make SnapshotManager use the new implementation to load snapshots of dropped tables in memory, what ensures they will be removed when expired.

Issue Links

• causes: CASSANDRA-17769 Fix org.apache.cassandra.simulator.test.ShortPaxosSimulationTest.simulationTest