Andrew Wong has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/16319 )

Change subject: KUDU-3180: prioritize larger mem-stores in time-based flusing
......................................................................


Patch Set 1: Code-Review+1

(2 comments)

http://gerrit.cloudera.org:8080/#/c/16319/1/src/kudu/tablet/tablet_replica_mm_ops.cc
File src/kudu/tablet/tablet_replica_mm_ops.cc:

http://gerrit.cloudera.org:8080/#/c/16319/1/src/kudu/tablet/tablet_replica_mm_ops.cc@78
PS1, Line 78: "Number of milliseconds after which a non-empty MRS/DMS will 
reach "
            :              "'full perf improvement' in time-based flushing.");
nit: "full perf improvement" may lead users to believe that a higher score 
cannot be given to these ops. How about something like:

"Number of milliseconds after which the time-based performance improvement 
score of a non-empty MRS/DMS flush op will reach its maximum value. The score 
may further increase as the MRS/DMS grows in size."


http://gerrit.cloudera.org:8080/#/c/16319/1/src/kudu/tablet/tablet_replica_mm_ops.cc@120
PS1, Line 120: std::max(1.0, extra_mb)
> We want to distinguish between mem-stores that above the threshold and that
Got it. Thanks for the clarification!

At first I had concerns about whether this would starve flushing for older 
mem-stores, but even if it did, this change doesn't make it much worse, given 
we can restore something close to the original behavior by increasing the flush 
threshold by 1.



--
To view, visit http://gerrit.cloudera.org:8080/16319
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I0a826643709a4990e40b0a49f89f4ea34f14163b
Gerrit-Change-Number: 16319
Gerrit-PatchSet: 1
Gerrit-Owner: Yifan Zhang <[email protected]>
Gerrit-Reviewer: Andrew Wong <[email protected]>
Gerrit-Reviewer: Kudu Jenkins (120)
Gerrit-Reviewer: Yifan Zhang <[email protected]>
Gerrit-Comment-Date: Wed, 12 Aug 2020 22:00:56 +0000
Gerrit-HasComments: Yes

Reply via email to