[CASSANDRA-14587] Deduplicate sstables shared by multiple snapshots when computing true disk space used

• Type: Improvement
• Status: Open
• Priority: Low
• Resolution: Unresolved
• Component/s: Tool/nodetool

Environment

Debian 8

Cassandra 3.11.2

• Impacts: Docs
• Assignee: Unassigned
• Reporter: Elliott Sims
• Created: 24/Jul/18 19:12
• Updated: 08/Feb/22 13:26

Description

Running 'nodetool listsnapshots' seems to overcount "TrueDiskSpaceUsed" under some circumstances.  Specifically when there's a large number of snapshots.  I suspect that it's not deduplicating space used when multiple snapshots share sstables that are not part of the current table.

Results of "nodetool listsnapshots":
Total TrueDiskSpaceUsed: 396.11 MiB
Results of "du -hcs" on the table's directory:
18M    total

This is 50+ snapshots (every minute) run with "-t <datestamp> -sf --column-family <tablename> <keyspace>"
The results of a "du -hcs -L <directory" come out pretty close to the "TrueDiskSpaceUsed"

I have only tested against 3.11.2, but have no reason to believe it's unique to that version or even 3.x.

Issue Links

• relates to: CASSANDRA-17267 Snapshot true size is miscalculated