Ashwani Raina has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/24328 )

Change subject: KUDU-3429 Make rows per block flag based (1/n)
......................................................................


Patch Set 2:

> To my best understanding, one of the potential causes of compaction
 > is that multiple changes to the same rowset still counts as one,
 > right?
 > So if I make 10 rows with a few 64k strings inside them, then start
 > to update them many-many times, then even setting this value to 10
 > or 20 does not save me to OOM issues. So instead of just
 > configuring this, shouldn't we have some safeguard against very
 > frequent updates?

Yes, just by lowering this value is not enough. It may help in cases where 
fewer and infrequent updates are spread across many rows, reducing to 10 rows 
per block would result in lowered memory requirements to load the 
mutations,blocks of those 10 rows as against 100 rows.
However, if there are frequent large number of updates for even reduced number 
of rows with a few 64K strings, having low value as 10, as you said, won't help 
much.

Bounding memory for such cases is the final goal of KUDU-3429. Since, that is a 
big complex change, I am breaking that down into smaller and comprehensible 
patches under the same upstream ticket and this is one of those. Expect more 
patches under the same upstream ticket.


--
To view, visit http://gerrit.cloudera.org:8080/24328
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I65f1e0d91e718b097bcc6e3979f246a13a618edf
Gerrit-Change-Number: 24328
Gerrit-PatchSet: 2
Gerrit-Owner: Ashwani Raina <[email protected]>
Gerrit-Reviewer: Alexey Serbin <[email protected]>
Gerrit-Reviewer: Ashwani Raina <[email protected]>
Gerrit-Reviewer: Kudu Jenkins (120)
Gerrit-Reviewer: Marton Greber <[email protected]>
Gerrit-Reviewer: Zoltan Martonka <[email protected]>
Gerrit-Comment-Date: Mon, 01 Jun 2026 13:01:50 +0000
Gerrit-HasComments: No

Reply via email to