Drop cache

[bengland2]

This change would enable snafu benchmarks to utilize privileged cache-dropper pods as a cherrypy web service with well-known TCP ports. run_snafu.py would invoke drop_cache() before every sample. The ripsaw component of the benchmark would start cache dropper pods and then would wait until they were running, then get their IP addresses and pass them as environment variables into run_snafu.py. It was tested using minikube - run_snafu has to be run within a k8s pod to be able to talk to cache-dropper pod.

[bengland2]

evidence that it works

[rsevilla87]

Hey @bengland2, can you rebase please?

[bengland2]

@rsevilla87 I rebased, what do you think?

[jtaleric]

/rerun all

[bengland2]

am cleaning up the lint errors now, will rerun afterwards. My bad, forgot to run pycodestyle

[comet-perf-ci]

Results for SNAFU CI Test

Test	Result	Runtime
src/fio_wrapper	PASS	00:13:23
src/fs_drift_wrapper	PASS	00:08:41
src/hammerdb	PASS	00:07:00
src/iperf	PASS	00:06:25
src/pgbench_wrapper	PASS	00:05:12
src/smallfile_wrapper	PASS	00:09:24
src/sysbench	PASS	00:03:09
src/uperf_wrapper	PASS	00:27:23
src/vegeta_wrapper	PASS	00:08:21
src/ycsb_wrapper	PASS	00:08:38

[bengland2]

actually the linter found bugs, so worth the effort, retesting with changes, then will re-CI it, then will merge.

[bengland2]

/rerun all

[comet-perf-ci]

Results for SNAFU CI Test

Test	Result	Runtime
src/fio_wrapper	PASS	00:13:56
src/fs_drift_wrapper	PASS	00:08:51
src/hammerdb	PASS	00:07:11
src/iperf	PASS	00:05:54
src/pgbench_wrapper	PASS	00:05:09
src/smallfile_wrapper	PASS	00:09:04
src/sysbench	FAIL	00:03:11
src/uperf_wrapper	PASS	00:26:50
src/vegeta_wrapper	PASS	00:08:09
src/ycsb_wrapper	PASS	00:08:15

[bengland2]

I'm not buying that this particular PR broke only sysbench and nothing else. Merging it. It does run fio.

[jtaleric]

22:21:22 Logs from sysbench-1ecb6eab-sc4xt
22:21:22 ++ kubectl -n my-ripsaw-6c45a80f-3f6b-446e-8cc4-48a4e978585b logs --tail=40 sysbench-1ecb6eab-sc4xt
22:21:22 Error from server (BadRequest): container "sysbench" in pod "sysbench-1ecb6eab-sc4xt" is waiting to start: PodInitializing
22:21:22 +++ error```

I am curious, most of our other CI has retries, but it looks like SNAFU does not?