Open j-rivero opened 3 years ago
The test is flaky on Citadel:
But it hasn't failed in Dome yet:
On #174, I'm adding more information to the test so we can understand better how it's failing.
But it hasn't failed in Dome yet:
Now that there are more builds, it's possible to see that the test is flaky on Dome too. And on Edifice:
I suspect this may be related to the issue that #221 is trying to fix, which seems to be caused by a difference in devices.
See a summary of the lastest builds:
Jenkins node | Passes | Failures |
---|---|---|
drogon | 0 | 5 |
optimus | 5 | 1 |
r2d2 | 4 | 0 |
The test seems to pass consistently on r2d2 and fail on drogon.
Note that the only failure on optimus happened on April 1st, and all successes happened before March 17th. I believe that optimus recently got an upgrade that may have changed its settings in a way that causes the test to fail.
Another instance of the test failure for Citadel (from July 16th): https://build.osrfoundation.org/job/ignition_rendering-ci-ign-rendering3-bionic-amd64/46
58: [ RUN ] Camera/CameraTest.VisualAt/ogre2
58: [1;32m[Msg] [0m[1;32mLoading plugin [[0m[1;32mignition-rendering-ogre2[0m[1;32m][0m[1;32m[0m
58: /var/lib/jenkins/workspace/ignition_rendering-ci-ign-rendering3-bionic-amd64/ign-rendering/test/integration/camera.cc:251: Failure
58: Expected: (nullptr) != (vis), actual: (nullptr) vs 16-byte object <00-00 00-00 00-00 00-00 00-00 00-00 00-00 00-00>
58: X: 300
58: /var/lib/jenkins/workspace/ignition_rendering-ci-ign-rendering3-bionic-amd64/ign-rendering/test/integration/camera.cc:259: Failure
58: Expected equality of these values:
58: nullptr
58: Which is: (nullptr)
58: vis
58: Which is: 16-byte object <20-0D 64-50 23-56 00-00 00-CC 24-50 23-56 00-00>
58: Found [box] at X [400]
58: /var/lib/jenkins/workspace/ignition_rendering-ci-ign-rendering3-bionic-amd64/ign-rendering/test/integration/camera.cc:265: Failure
58: Expected: (nullptr) != (vis), actual: (nullptr) vs 16-byte object <00-00 00-00 00-00 00-00 00-00 00-00 00-00 00-00>
58: X: 600
58: /var/lib/jenkins/workspace/ignition_rendering-ci-ign-rendering3-bionic-amd64/ign-rendering/test/integration/camera.cc:265: Failure
58: Expected: (nullptr) != (vis), actual: (nullptr) vs 16-byte object <00-00 00-00 00-00 00-00 00-00 00-00 00-00 00-00>
58: X: 700
58: [ FAILED ] Camera/CameraTest.VisualAt/ogre2, where GetParam() = "ogre2" (387 ms)
I took some time to gather metrics about how much this affects the ign-rendering builds across the ignition buildfarms. Taking into account only the builds with a test report available.
ignition_rendering-ci-ign-rendering3-bionic-amd64
has this error 14 out of 25 builds. 56.0% Flaky
ignition_rendering-ci-ign-rendering4-bionic-amd64
has this error 13 out of 23 builds. 56.5% Flaky
ignition_rendering-ci-ign-rendering5-bionic-amd64
has this error 10 out of 21 builds. 47.6% Flaky
ignition_rendering-ci-main-bionic-amd64
has this error 9 out of 34 builds. 26.5% Flaky
I think it would be a good idea to take a closer look into this one, it's a test that fails frequently.
I'm testing this in https://github.com/ignitionrobotics/ign-rendering/pull/450. Setting the device pixel ratio to 1.0f seems to have fixed the issue on the CI machines: https://github.com/ignitionrobotics/ign-rendering/blob/ign-rendering5/ogre2/src/Ogre2Camera.cc#L250
This error is failing consistently on our buildfarm ign-rendering3
job, see:
https://build.osrfoundation.org/job/ignition_rendering-ci-ign-rendering3-focal-amd64/11/
Is it possible to backport this one? or should we disable the test on citadel and document it as a known issue? @chapulina
if we want to backport, these are probably the relevant changes: https://github.com/gazebosim/gz-rendering/pull/446/files#diff-7d09d6004075a16fceed8b6665c62249a2f928e86838438073c983b1c2ad6ad7R107-R110
I took some time to check/backport this one, the link you attached here is showing me this:
src/Utils_TEST.cc
EXPECT_EQ(0u, rayResult.objectId);
VisualPtr root = scene->RootVisual();
Am I correct? To me it looks like the important changes are happening in the ogre2/src/Ogre2SelectionBuffer.cc
, and the changes there were not trivial for me to backport.
oh it should be:
// the scaling factor seems to cause issues with mouse picking.
// see https://github.com/ignitionrobotics/ign-gazebo/issues/147
#if 0
basically forcing that function to return 1.0
issue is back on CI machine for gz-rendering7: https://github.com/gazebosim/gz-rendering/pull/758#issuecomment-1329246890
FYI @Crola1702
I've seen the
Camera/CameraTest.VisualAt/ogre2
failing in some builds for Bionic/amd64. Changes listed in build 20seems to be the starting point of the failures.