roachtest: tpchvec/perf failed

[cockroach-teamcity]

(roachtest).tpchvec/perf failed on release-20.1@7699136a4861a6f6423e16ab7954fc4ac226bd6a:

The test failed on branch=release-20.1, cloud=gce:
test artifacts and logs in: /home/agent/work/.go/src/github.com/cockroachdb/cockroach/artifacts/20200407-1857166/tpchvec/perf/run_1
	tpchvec.go:644,tpchvec.go:659,test_runner.go:753: [q15] vec ON is slower by 15.33% than vec OFF
		vec ON times: [4.49 4.59 4.61]
		vec OFF times: [3.95 3.98 4.18]

More

Artifacts: /tpchvec/perf

See this test on roachdash
_{powered by pkg/cmd/internal/issues}

[yuzefovich]

Here are the comparisons of the run times of query 15 for the last five days:

09:20:32 tpchvec.go:629: [q15] vec OFF was faster by 10.58%: 4.39s ON vs 3.97s OFF --- WARNING
09:04:26 tpchvec.go:629: [q15] vec OFF was faster by 9.98%: 4.41s ON vs 4.01s OFF --- WARNING
09:13:59 tpchvec.go:629: [q15] vec OFF was faster by 9.93%: 4.65s ON vs 4.23s OFF --- WARNING
09:16:37 tpchvec.go:629: [q15] vec OFF was faster by 12.11%: 4.63s ON vs 4.13s OFF --- WARNING
09:25:08 tpchvec.go:629: [q15] vec OFF was faster by 15.33%: 4.59s ON vs 3.98s OFF --- WARNING

Most of the time of this query spends in index join, and it seems to me that the overhead of columnarizer-materializer causes this slowdown. It doesn't seem like it is a regression due to recent commits, so I'll open up a PR to increase the slowness failure threshold from 15% to 20%.