URL:
https://github.com/SSSD/sssd/pull/947
Title: #947: tests: fix race conditions in integration tests
pbrezina commented:
"""
tl,dr: I'm not categorically against estimation of
REAL_ENUMERATION_TIMEOUT as 2*ENUMERATION_TIMEOUT.
But IMO this has nothing to do with Nyquist frequency and this is still arbitrary (just
"good enough with a very good margin") value.
You did not give me any argument to convince me otherwise. There is no point spending time
on this as long as the tests works so let's agree to disagree. I can remove this name
from comments and code if you feel strongly about it.
> I think the `/2` sleep is to wait until the first enumeration is
finished.
To be sure that there is really no entry in ldap before test adds it? Makes sense.
But:
1. why then it is /2? As discussed above that's not enough even with this patch.
And it was clearly too small for this purpose before, so intention of this timeout is
still not clear to me.
2. do we really need this in the beginning of every test? I am not sure if sssd
restarted in every test or once for the whole module?
I really do not know. @spbnick Git blame says you are the original author of these tests.
Do you remember why you used `time.sleep(INTERACTIVE_TIMEOUT/2)` at the beginning of each
enumeration test in `test_enumeration.py`?
It is 5*(INTERACTIVE_TIMEOUT/2) = 20 seconds for each run (or twice
if we remove /2). Doesn't sound like too much but things add up. And CI runs are
really slow...
I don't care that it will be five minutes slower if it runs correctly.
"""
See the full comment at
https://github.com/SSSD/sssd/pull/947#issuecomment-559462549