Alexey Serbin has posted comments on this change. ( 
http://gerrit.cloudera.org:8080/16312 )

Change subject: [tserver] add test to reproduce KUDU-1587 conditions
......................................................................


Patch Set 1:

(5 comments)

http://gerrit.cloudera.org:8080/#/c/16312/1/src/kudu/tserver/tablet_server-test.cc
File src/kudu/tserver/tablet_server-test.cc:

http://gerrit.cloudera.org:8080/#/c/16312/1/src/kudu/tserver/tablet_server-test.cc@4368
PS1, Line 4368: num_data_dirs
> Would it also make sense to reduce the number of apply threads to 1? Would
Well, I'd rather keep it closer to "real" configurations.  That also helps to 
avoid extra changes introducing max pool size for the apply queue.  Or you 
think it's better to introduce control knobs for the apply queue max_size 
anyways (maybe, in a separate changelist)?


http://gerrit.cloudera.org:8080/#/c/16312/1/src/kudu/tserver/tablet_server-test.cc@4387
PS1, Line 4387: re the
> nit: incomplete?
Done


http://gerrit.cloudera.org:8080/#/c/16312/1/src/kudu/tserver/tablet_server-test.cc@4446
PS1, Line 4446:     EXPECT_LT(h->MaxValue(),
              :               kLatencyMs * 1000 * kNumCalls / base::NumCPUs());
              :     EXPECT_LT(h->ValueAtPercentile(99),
              :               kLatencyMs * 1000 * kNumCalls / (2 * 
base::NumCPUs()));
              :     EXPECT_LT(h->ValueAtPercentile(50),
              :               kLatencyMs * 1000 * kNumCalls / (4 * 
base::NumCPUs()));
> Making sure I understand these assertions: if we had a single core, we'd ex
Well, these are rather simple heuristics.  I'm not sure we want to draw some 
exact lines here based on the parameters for CoDel.

I added an extra comment to clarify on this.


http://gerrit.cloudera.org:8080/#/c/16312/1/src/kudu/tserver/tablet_server-test.cc@4463
PS1, Line 4463:
              :     EXPECT_LT(h->MaxValue(), kNumCalls / 2);
              :     EXPECT_LT(h->MeanValue(), kNumCalls / 4);
> Same here, right? These multipliers will be adjusted depending on the imple
These are simple heuristics as well. They depend on the injected latency and 
the 'overloaded' threshold for the apply queue.

I added this as a comment.

BTW, the nice feature of the CoDel algorithm is that it has almost no 
configuration parameters.


http://gerrit.cloudera.org:8080/#/c/16312/1/src/kudu/tserver/tablet_server-test.cc@4468
PS1, Line 4468:   // Some calls should be rejected by the apply queue, so the 
maximum queue
              :   // time should be much lower than xxx.
> nit: move this above L4436? Or is this referring to the RPC queue?
Done



--
To view, visit http://gerrit.cloudera.org:8080/16312
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-MessageType: comment
Gerrit-Change-Id: I515a1b26152680ee9b9361afcf84fec39b8f962d
Gerrit-Change-Number: 16312
Gerrit-PatchSet: 1
Gerrit-Owner: Alexey Serbin <[email protected]>
Gerrit-Reviewer: Alexey Serbin <[email protected]>
Gerrit-Reviewer: Andrew Wong <[email protected]>
Gerrit-Reviewer: Bankim Bhavsar <[email protected]>
Gerrit-Reviewer: Kudu Jenkins (120)
Gerrit-Comment-Date: Sat, 15 Aug 2020 00:15:27 +0000
Gerrit-HasComments: Yes

Reply via email to