Hi,
we had such cases in the past like bug 1817721 for bionic and maybe bug 1892130 is about the same as well. There were more but I didn't want to search for all of them - what I checked is that there are no open ones clearly pointing out the recent further drop in already flaky subtests.
In particular the tests "tests-in-lxd" and "systemd-fsckd" were known to be flaky before, but got even worse.
Here stats of the last 40 runs, it might be a coincidences that this is after 246-2ubuntu1 landed. Could as well be any other change
groovy
amd64
tests-in-lxd (F 42% S 0% B 10% => P 45%/) FFFFBFFFFFFFB....FF.B.....F.....F...FBF
build-login (F 0% S 0% B 10% => P 87%/) ....B.......B.......B................B.
unit-config (F 0% S 0% B 10% => P 87%/) ....B.......B.......B................B.
networkd-testpy (F 0% S 0% B 10% => P 87%/) ....B.......B.......B................B.
boot-and-services (F 0% S 0% B 10% => P 87%/) ....B.......B.......B................B.
boot-smoke (F 0% S 0% B 10% => P 87%/) ....B.......B.......B................B.
logind (F 0% S 0% B 10% => P 87%/) ....B.......B.......B................B.
storage (F 0% S 0% B 10% => P 87%/) ....B.......B.......B................B.
upstream (F 35% S 0% B 10% => P 52%/) ..FFB.FFF.FFB....FF.B.....F.F..F....FBF
udev (F 0% S 0% B 10% => P 87%/) ....B.......B.......B................B.
systemd-fsckd (F 37% S 0% B 10% => P 50%/) FFFFBFFFFFFFB.FF...FB.....F..........B.
root-unittests (F 0% S 0% B 10% => P 87%/) ....B.......B.......B................B.
ppc64el
tests-in-lxd (F 25% S 0% B 0% => P 75%/) FFFF....FF............FFF.....F.........
systemd-fsckd (F 35% S 0% B 0% => P 65%/) FFFFFFF...FF........F....FF.F..F........
root-unittests (F 2% S 0% B 0% => P 97%/) ..............................F.........
s390x
tests-in-lxd (F 52% S 0% B 0% => P 47%/) FFFFFFF.FFFFFFF.FF.........FFFF...F.....
timedated (F 2% S 0% B 0% => P 97%/) ...........F............................
upstream (F 17% S 0% B 0% => P 82%/) .....F......F.F.............FFF...F.....
systemd-fsckd (F 32% S 0% B 0% => P 67%/) FFFFFFF..FF..F.................FF..F....
root-unittests (F 10% S 0% B 0% => P 90%/) ............................FFF...F.....
arm64
tests-in-lxd (F 40% S 0% B 2% => P 57%/) FFFFF.B...FFF.FF..F..F.........FFF.F....
logind (F 2% S 0% B 2% => P 95%/) ......B...................F.............
upstream (F 22% S 0% B 2% => P 75%/) ...F.FB.....F.F.............F..FFF.F....
root-unittests (F 12% S 0% B 2% => P 85%/) ......B.F...........F.F........F...F....
(I'm sure LP will make this unreadable, but is is nice in monospace)
Whatever the root cause is - the success rate of these has reduced so much that the (even formerly questionable) practice of retry-until-success won't work anymore.
I have run the two tests in a local VM and systemd-fsckd works there while tests-in-lxd seems to trip over the old flaky fellow being "boot-and-services".
We had the discussion in the past, but I think I need to again bring up the suggestion to skip "tests-in-lxd" and "systemd-fsckd" until they are on reasonable success rates.
Hi,
we had such cases in the past like bug 1817721 for bionic and maybe bug 1892130 is about the same as well. There were more but I didn't want to search for all of them - what I checked is that there are no open ones clearly pointing out the recent further drop in already flaky subtests.
In particular the tests "tests-in-lxd" and "systemd-fsckd" were known to be flaky before, but got even worse.
Here stats of the last 40 runs, it might be a coincidences that this is after 246-2ubuntu1 landed. Could as well be any other change
groovy ...FF.B. ....F.. ...F... FBF .....B. ......B. ....... ....... .B. .....B. ......B. ....... ....... .B. .....B. ......B. ....... ....... .B. .....B. ......B. ....... ....... .B. .....B. ......B. ....... ....... .B. .....B. ......B. ....... ....... .B. .....B. ......B. ....... ....... .B. FFB.... FF.B... ..F.F.. F....FBF .....B. ......B. ....... ....... .B. FF...FB. ....F.. ....... .B. .....B. ......B. ....... ....... .B. .FF.... ....... .FFF... ..F.... ..... ..FF... .....F. ...FF.F. .F..... ... ....... ....... ....... ..F.... ..... FFFFFFF. FF..... ....FFFF. ..F.... . ....F.. ....... ....... ....... ..... .....F. F...... ....... FFF...F. .... .FF..F. ....... ....... ..FF..F. ... ....... ....... ....... FFF...F. .... ..FFF.FF. .F..F.. ....... FFF.F.. .. ....... ....... ....F.. ....... .... ....F.F. ....... .....F. .FFF.F. ... F...... .....F. F...... ..F...F. ...
amd64
tests-in-lxd (F 42% S 0% B 10% => P 45%/) FFFFBFFFFFFFB.
build-login (F 0% S 0% B 10% => P 87%/) ....B..
unit-config (F 0% S 0% B 10% => P 87%/) ....B..
networkd-testpy (F 0% S 0% B 10% => P 87%/) ....B..
boot-and-services (F 0% S 0% B 10% => P 87%/) ....B..
boot-smoke (F 0% S 0% B 10% => P 87%/) ....B..
logind (F 0% S 0% B 10% => P 87%/) ....B..
storage (F 0% S 0% B 10% => P 87%/) ....B..
upstream (F 35% S 0% B 10% => P 52%/) ..FFB.FFF.
udev (F 0% S 0% B 10% => P 87%/) ....B..
systemd-fsckd (F 37% S 0% B 10% => P 50%/) FFFFBFFFFFFFB.
root-unittests (F 0% S 0% B 10% => P 87%/) ....B..
ppc64el
tests-in-lxd (F 25% S 0% B 0% => P 75%/) FFFF...
systemd-fsckd (F 35% S 0% B 0% => P 65%/) FFFFFFF.
root-unittests (F 2% S 0% B 0% => P 97%/) .......
s390x
tests-in-lxd (F 52% S 0% B 0% => P 47%/) FFFFFFF.
timedated (F 2% S 0% B 0% => P 97%/) .......
upstream (F 17% S 0% B 0% => P 82%/) .....F.
systemd-fsckd (F 32% S 0% B 0% => P 67%/) FFFFFFF.
root-unittests (F 10% S 0% B 0% => P 90%/) .......
arm64
tests-in-lxd (F 40% S 0% B 2% => P 57%/) FFFFF.B.
logind (F 2% S 0% B 2% => P 95%/) ......B.
upstream (F 22% S 0% B 2% => P 75%/) ...F.FB.
root-unittests (F 12% S 0% B 2% => P 85%/) ......B.
(I'm sure LP will make this unreadable, but is is nice in monospace)
Whatever the root cause is - the success rate of these has reduced so much that the (even formerly questionable) practice of retry-until-success won't work anymore.
I have run the two tests in a local VM and systemd-fsckd works there while tests-in-lxd seems to trip over the old flaky fellow being "boot-and- services" .
We had the discussion in the past, but I think I need to again bring up the suggestion to skip "tests-in-lxd" and "systemd-fsckd" until they are on reasonable success rates.