[ 
https://issues.apache.org/jira/browse/CASSANDRA-21041?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18041029#comment-18041029
 ] 

Branimir Lambov commented on CASSANDRA-21041:
---------------------------------------------

What you are describing is the output of {{{}Overlaps.constructOverlapSets{}}}, 
with just the lengths of the resulting sets instead the sstable names, isn't 
it? Sounds like a version of the above at full precision instead of for the 
number of shards of the compaction strategy (or a number of shards passed as an 
argument).

This would also work, though a little harder to read. We still need to be able 
to split this into levels to get a better feeling of what compaction is doing, 
at least for some of the strategies.

 

 

> SSTable overlap statistics
> --------------------------
>
>                 Key: CASSANDRA-21041
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-21041
>             Project: Apache Cassandra
>          Issue Type: Improvement
>          Components: Local/Compaction, Local/Compaction/UCS
>            Reporter: Branimir Lambov
>            Assignee: guo Maxwell
>            Priority: Normal
>             Fix For: 6.x
>
>         Attachments: overlap-graph.png
>
>
> When trying to evaluate the effect of compaction and its settings, perhaps 
> the most relevant information is the amount of overlap between sstables, i.e. 
> the number of sstables that have to be queried to serve any request for a 
> specific partition. We do not currently present this information for any 
> compaction strategy.
> For some strategies this can be organized by level, and specifically for UCS, 
> by shard.
> [This patch in the DataStax 
> repository|https://github.com/datastax/cassandra/commit/c7d6f082e9eef8f9121ef271495e9a2469bde671]
>  implements such a statistic for UCS, which outputs overlap per level as
> {code}
> Max overlap map for density.blobs:
>   all: 26, per shard: [11, 11, 17, 17, 14, 14, 20, 20, 17, 17, 23, 23, 20, 
> 20, 26, 26]
>   unrepaired:0: 18, per shard: [6, 10, 14, 18]
>   unrepaired:1: 9, per shard: [6, 9, 5, 8, 4, 7, 3, 6]
>   unrepaired:2: 3, per shard: [3, 3, 3, 3, 3, 3, 2, 2, 3, 3, 3, 3, 3, 3, 2, 
> 2, 3, 3, 3, 3, 3, 3, 2, 2, 3, 3, 3, 3, 3, 3, 2, 2]
> {code}.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to