[
https://issues.apache.org/jira/browse/CASSANDRA-13738?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16112230#comment-16112230
]
Jay Zhuang commented on CASSANDRA-13738:
----------------------------------------
I think the problem has been there for awhile. It only happens when [the
IndexSummary is
rebuilt|https://github.com/apache/cassandra/blob/trunk/src/java/org/apache/cassandra/io/sstable/IndexSummaryRedistribution.java#L129],
which is triggered by [large read traffic load
changing|https://github.com/apache/cassandra/blob/trunk/src/java/org/apache/cassandra/io/sstable/IndexSummaryRedistribution.java#L202].
I'm able to reproduce it with an unittest. Here is the patch, please review:
| branch | utest |
| [13738-2.2|https://github.com/cooldoger/cassandra/tree/13738-2.2] |
[circleci#53|https://circleci.com/gh/cooldoger/cassandra/53] |
| [13738-3.0|https://github.com/cooldoger/cassandra/tree/13738-3.0] |
[circleci#52|https://circleci.com/gh/cooldoger/cassandra/52] |
| [13738-3.11|https://github.com/cooldoger/cassandra/tree/13738-3.11] |
[circleci#51|https://circleci.com/gh/cooldoger/cassandra/51] |
| [13738-trunk|https://github.com/cooldoger/cassandra/tree/trunk] |
[circleci#50|https://circleci.com/gh/cooldoger/cassandra/50] |
Seems branch {{2.1}} don't have this issue.
> Load is over calculated after each IndexSummaryRedistribution
> -------------------------------------------------------------
>
> Key: CASSANDRA-13738
> URL: https://issues.apache.org/jira/browse/CASSANDRA-13738
> Project: Cassandra
> Issue Type: Bug
> Components: Core
> Reporter: Jay Zhuang
> Assignee: Jay Zhuang
> Fix For: 3.0.x, 3.11.x, 4.x
>
> Attachments: sizeIssue.png
>
>
> For example, here is one of our cluster with about 500GB per node, but
> {{nodetool status}} shows far more load than it actually is and keeps
> increasing, restarting the process will reset the load, but keeps increasing
> afterwards:
> {noformat}
> Status=Up/Down
> |/ State=Normal/Leaving/Joining/Moving
> -- Address Load Tokens Owns (effective) Host ID
> Rack
> UN IP1***** 13.52 TB 256 100.0%
> c4c31e0a-3f01-49f7-8a22-33043737975d rac1
> UN IP2***** 14.25 TB 256 100.0%
> efec4980-ec9e-4424-8a21-ce7ddaf80aa0 rac1
> UN IP3***** 13.52 TB 256 100.0%
> 7dbcfdfc-9c07-4b1a-a4b9-970b715ebed8 rac1
> UN IP4***** 22.13 TB 256 100.0%
> 8879e6c4-93e3-4cc5-b957-f999c6b9b563 rac1
> UN IP5***** 18.02 TB 256 100.0%
> 4a1eaf22-4a83-4736-9e1c-12f898d685fa rac1
> UN IP6***** 11.68 TB 256 100.0%
> d633c591-28af-42cc-bc5e-47d1c8bcf50f rac1
> {noformat}
> !sizeIssue.png|test!
> The root cause is if the SSTable index summary is redistributed (typically
> executes hourly), the updated SSTable size is added again.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
