[
https://issues.apache.org/jira/browse/PHOENIX-6967?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17735766#comment-17735766
]
Istvan Toth edited comment on PHOENIX-6967 at 6/22/23 3:55 AM:
---------------------------------------------------------------
My initial idea was to use RoundRobinResultIterator for these cases instead of
ConcatResultIterator.
However, even if I do that, the query still gets serialized.
I'm uploading this as a WIP PR, in case someone wants to pick this up.
was (Author: stoty):
My initial idea iwass to use RoundRobinResultIterator for these cases instead
of ConcatResultIterator.
However, even if I do that, the query still gets serialized.
I'm uploading this as a WIP PR, in case someone wants to pick this up.
> Empty cell generation in PostdDDLMutationPlan is executed serially
> ------------------------------------------------------------------
>
> Key: PHOENIX-6967
> URL: https://issues.apache.org/jira/browse/PHOENIX-6967
> Project: Phoenix
> Issue Type: Bug
> Components: core
> Affects Versions: 5.2.0, 5.1.3
> Reporter: Istvan Toth
> Priority: Major
>
> The empty CQ generation after create table is not run in parallel.
> At first, I can see the reads and writes on happening on each region, but
> after a minute or so, only one region is processed at a time in ascending
> order.
> My guess is that it has something to do with keepalive/chunked results not
> being processed on the client side correctly.
> I also suspect that this is not really specific to PostDDLMutationPlan, and
> all queries where a single chunk takes a lot of time would be affected.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
