[
https://issues.apache.org/jira/browse/PHOENIX-2622?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15111724#comment-15111724
]
Terence Yim commented on PHOENIX-2622:
--------------------------------------
Hi [~jamestaylor], is the Tephra server running inside the HBase cluster,
sharing CPU/IO with the region server, or is it outside of the cluster? Also,
is it possible that after the region splits, the HDFS replica that the Tephra
WAL is writing to is fighting for IO resource with the region servers for the
newly assigned region? I am just guessing here, but Tephra itself does not
aware of regions at all.
> Writes to transactional table get slower after regions splits to multiple
> region servers
> ----------------------------------------------------------------------------------------
>
> Key: PHOENIX-2622
> URL: https://issues.apache.org/jira/browse/PHOENIX-2622
> Project: Phoenix
> Issue Type: Bug
> Reporter: Mujtaba Chohan
> Attachments: upsert_rate.png
>
>
> Upsert rate get slower for transactional table as more and more regions gets
> split to different region servers. More details to be added.
> !upsert_rate.png! This chart shows the upsert rate of 100K batches over time
> for transactional vs standard table for a total of 4M rows split in 6 regions.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
