Some tests take too long time

[jihoonson]

Recently, our travis ci sometimes fails due to the limited job time. I found some unit tests take too long time. Here is the list.

Tests run: 5, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 148.191 sec - in io.druid.query.groupby.epinephelinae.BufferGrouperTest

Tests run: 5, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 233.411 sec - in io.druid.query.groupby.epinephelinae.ByteBufferMinMaxOffsetHeapTest

Tests run: 24, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 315.178 sec - in io.druid.segment.data.CompressedVSizeIndexedV3WriterTest

Tests run: 66, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 1,095.118 sec - in io.druid.segment.data.CompressedVSizeIntsIndexedSupplierTest

Tests run: 2, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 59.984 sec - in io.druid.server.initialization.JettyQosTest

Tests run: 35, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 161.293 sec - in io.druid.query.aggregation.datasketches.theta.oldapi.OldApiSketchAggregationTest

Tests run: 50, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 196.692 sec - in io.druid.query.aggregation.datasketches.theta.SketchAggregationTest

[jon-wei]

Thanks for gathering this info, I'm looking into scaling down the unit tests in ByteBufferMinMaxOffsetHeapTest

[jihoonson]

@jon-wei thanks for looking into it!

[leventov]

@gianm seems like none of Travis build after #4394 succeed, I think there is some problem with this PR

[gianm]

@leventov which failures do you think might be related to that PR? Is https://travis-ci.org/druid-io/druid/jobs/242999630 (#4404) one of them?

[leventov]

@gianm it also fails

[gianm]

Based on @jihoonson's original comment, and also the failure at https://travis-ci.org/druid-io/druid/jobs/242999630, and also running them locally, seems like the problem is likely in CompressedVSizeIndexedV3WriterTest and/or CompressedVSizeIntsIndexedSupplierTest. Travis fails if there's no output for 10 minutes, and the latter test does take more than 10 minutes to run sometimes. I have no idea why or if it is really related to #4394 or not.

[gianm]

I ran CompressedVSizeIntsIndexedSupplierTest on the druid-0.10.0 tag and it took 15s. So something has changed since then.

[gianm]

I ran it again on 976492c (which doesn't include #4394) and it took at least a few minutes (I stopped it before it finished). So I don't think #4394 is related. But something since druid-0.10.0 had some kind of an effect.

[gianm]

@leventov I looked into it, and the culprit is #4252, specifically these two changes.

• https://github.com/druid-io/druid/pull/4252/files#diff-2bed53e52b4912e98a891b1ccff9326eR363
• https://github.com/druid-io/druid/pull/4252/files#diff-fed2dc5ea85516ef128b0a2fdb923c69R192

For some reason, the changed version is much slower. Reverting back to Arrays.asList speeds up the tests significantly.

[gianm]

@leventov #4412

[gianm]

On my machine JettyQosTest takes 6s. I wonder if the substantially longer time you saw is because it spawns a lot of threads. @jihoonson did those numbers come from travis or one of your own machines?

[gianm]

#4414 is meant to help with CompressedVSizeIntsIndexedSupplierTest.