You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cassandra.apache.org by "Sam Tunnicliffe (Jira)" <ji...@apache.org> on 2024/07/05 17:51:00 UTC
[jira] [Comment Edited] (CASSANDRA-19755) Coordinator read latency metrics are inflated for some queries
[ https://issues.apache.org/jira/browse/CASSANDRA-19755?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17863392#comment-17863392 ]
Sam Tunnicliffe edited comment on CASSANDRA-19755 at 7/5/24 5:50 PM:
---------------------------------------------------------------------
The issue is with sharing the same start time ({{RequestTime.startedAtNanos()}}) for the timing of each individual read query, which was introduced by CASSANDRA-19534.
This results in accumulation so that the latency reported for query _n_ includes the latency of _(n-1)_ + _(n-2)_ + _(n-3)_ etc.
The fix is trivial and doesn't affect the other aspects of CASSANDRA-19534.
|4.1 patch:|[https://github.com/apache/cassandra/pull/3404]|
|5.0 patch:|[https://github.com/apache/cassandra/pull/3405]|
|trunk patch:|[https://github.com/apache/cassandra/pull/3406]|
There are 3 failures in the 4.1 CI results, but all of them are present without this patch. Two are already tracked (CASSANDRA-17298, CASSANDRA-19702), but I couldn't find an open JIRA for the {{ReadRepairTest}} dtest failure.
The jvm17 dtest timeouts on trunk are most likely down to CASSANDRA-19239, for which a fix is coming soon.
was (Author: beobal):
The issue is with sharing the same start time ({{RequestTime.startedAtNanos()}}) for the timing of each individual read query, which was introduced by CASSANDRA-19534.
This results in accumulation so that the latency reported for query _n_ includes the latency of _(n-1)_ + _(n-2)_ + _(n-3)_ etc.
The fix is trivial and doesn't affect the other aspects of CASSANDRA-19534.
|4.1 patch:|[https://github.com/apache/cassandra/pull/3404]|
|5.0 patch:|[https://github.com/apache/cassandra/pull/3405]|
|trunk patch:|[https://github.com/apache/cassandra/pull/3406]|
There are 3 failures in the 4.1 CI results, but all of them are present without this patch. Two are already tracked (CASSANDRA-17928, CASSANDRA-19702), but I couldn't find an open JIRA for the {{ReadRepairTest}} dtest failure.
The jvm17 dtest timeouts on trunk are most likely down to CASSANDRA-19239, for which a fix is coming soon.
> Coordinator read latency metrics are inflated for some queries
> --------------------------------------------------------------
>
> Key: CASSANDRA-19755
> URL: https://issues.apache.org/jira/browse/CASSANDRA-19755
> Project: Cassandra
> Issue Type: Bug
> Components: Observability/Metrics
> Reporter: Sam Tunnicliffe
> Assignee: Sam Tunnicliffe
> Priority: Normal
> Fix For: 4.1.x, 5.0-rc
>
> Attachments: ci_summary-4.1.html, ci_summary-5.0.html, ci_summary-trunk.html
>
>
> When a partition read is decomposed on the coordinator into multiple single partition read queries, the latency metric captured in StorageProxy can be artificially increased.
> This primarily affects reads where paging and aggregates are used or where an IN clause selects multiple partition keys.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@cassandra.apache.org
For additional commands, e-mail: commits-help@cassandra.apache.org