[CASSANDRA-14084] Disks can be imbalanced during replace of same address when using JBOD

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 3.11.2, 4.0-alpha1, 4.0
• Severity: Normal
• Assignee: Paulo Motta
• Reporter: Paulo Motta

Authors

Paulo Motta

Reviewers

Marcus Eriksson

• Created: 30/Nov/17 17:00
• Updated: 15/May/20 08:06
• Resolved: 11/Dec/17 20:24

Description

While investigating CASSANDRA-14083, I noticed that we use the pending ranges to calculate the disk boundaries when the node is bootstrapping.

The problem is that when the node is replacing a node with the same address, it sets itself as normal locally (for other unrelated reasons), so the local ranges will be null and consequently the disk boundaries will be null. This will cause the sstables to be randomly spread across disks potentially causing imbalance.

Issue Links

• relates to: CASSANDRA-12344 Forward writes to replacement node with same address during replace
• relates to: CASSANDRA-14083 Avoid invalidating disk boundaries unnecessarily