Description
When I run the test cases of 1.10.0-RC2, the 'TestSentryClientMetrics.Basic' is a little bit strange. Sometimes it works, but sometime it doesn't. Today, I took a close look at the output log and found some useful info:
// code placeholder
I0701 16:37:24.925388 33240 thread.cc:675] Ended thread 33240 - thread pool:Sentry [worker]
I0701 16:37:24.925501 33015 thread.cc:624] Started thread 33436 - thread pool:Sentry [worker]
I0701 16:37:25.322556 33015 mini_sentry.cc:164] Pausing Sentry
W0701 16:37:27.331832 33436 sentry_client.cc:134] Time spent starting Sentry client: real 1.999s user 0.000s sys 0.000s
W0701 16:37:27.331894 33436 client.h:352] Failed to connect to Sentry (127.32.61.193:59755): Timed out: failed to open Sentry connection: THRIFT_EAGAIN (timed out)
I0701 16:37:27.331986 33015 mini_sentry.cc:172] Resuming Sentry
/mnt/ddb/2/helif/apache/kudu/src/kudu/master/sentry_authz_provider-test.cc:1415: Failure
Expected: (2000000) < (hist->histogram()->MaxValue()), actual: 2000000 vs 1999002
I0701 16:37:27.332604 33015 mini_sentry.cc:155] Stopping Sentry
Then I looked through the file 'sentry_authz_provider-test.cc', it seems the timeout value is too short:
Perhaps, we can increase this value (default 60 seconds) to 4 or 5 seconds to avoid the failures, though Alexey Serbin(not sure) and I have this problem.