Fix random test failures in github #939
Conversation
We see many failures when starting podman based environment on github. Add verbose logging to get more info about the failures. Signed-off-by: Nir Soffer <nsoffer@redhat.com>
We used 64 bit random number, which should be enough, but we see random
failures in github with this error:
time="2023-06-22T15:04:59Z" level=warning msg="Error validating CNI
config file /etc/cni/net.d/test-77836bdde14be1d8-cluster.conflist:
[plugin bridge does not support config version \"1.0.0\" plugin
portmap does not support config version \"1.0.0\" plugin firewall
does not support config version \"1.0.0\" plugin tuning does not
support config version \"1.0.0\"]" Error: volume with name
test-77836bdde14be1d8-cluster already exists: volume already exists
This is probably not the reason for "volume already exists", but lets
make sure it cannot be the problem by using a 128 bit random prefix.
Signed-off-by: Nir Soffer <nsoffer@redhat.com>
To report a bug in minikube we need to use the --alsologtostderr option. Enable this option when using debug mode. With this option minikube logs huge amount of debug log to stderr, but we will see the logs only if the command fails in the commands.Error traceback. Signed-off-by: Nir Soffer <nsoffer@redhat.com>
Log the minikube commands in debug level to make it eaiser to understand
failures on remote machines and verify changes in minikube commands.
Example of the new logs:
$ drenv start envs/test.yaml -v
2023-06-22 19:25:07,827 INFO [test] Starting environment
2023-06-22 19:25:07,920 INFO [cluster] Starting minikube cluster
2023-06-22 19:25:07,920 DEBUG [cluster] Running ['minikube', 'start', '--profile', 'cluster',
'--driver', 'podman', '--container-runtime', 'cri-o', '--disk-size', '20g', '--nodes', '1',
'--cni', 'auto', '--cpus', '2', '--memory', '2g', '--alsologtostderr']
...
$ drenv delete envs/test.yaml -v
2023-06-22 19:26:05,932 INFO [test] Deleting environment
2023-06-22 19:26:05,933 INFO [cluster] Deleting cluster
2023-06-22 19:26:05,933 DEBUG [cluster] Running ['minikube', 'delete', '--profile', 'cluster']
...
Signed-off-by: Nir Soffer <nsoffer@redhat.com>
nirs
force-pushed
the
tmpenv
branch
3 times, most recently
from
97a3c38
to
d7ff7cb
Compare
nirs
changed the title
| @@ -37,21 +37,23 @@ environment. | |||
| [Install clusteradm CLI tool](https://open-cluster-management.io/getting-started/installation/start-the-control-plane/#install-clusteradm-cli-tool) | |||
| for the details. Version 0.5.0 or later is required. | |||
|
|
|||
| 1. Install `podman` | |||
| 1. Install `docker` | |||
There was a problem hiding this comment.
I think this is a pretty big change and is claiming incompatibility with podman. I suggest that you update the commit message with example issues that you have seen with podman. If possible, also link to one or more runs on github actions where the failures have occurred.
There was a problem hiding this comment.
poadmn is considered experimental in minikube:
https://minikube.sigs.k8s.io/docs/drivers/podman/
We used it because it was a way to get rid of docker, and it seems to work, but
since we added it was removed from all environments. We kept it only in the test
env used for running the tests locally and in github. Now that it does not work
in github there is no reason to use it.
We can consider using podman again when it is officially supported and works
for us in github.
If you look in the latest github actions in this week you will find that all of
them failed in drenv tests in the same way:
- https://github.com/RamenDR/ramen/actions/runs/5348065862
- https://github.com/RamenDR/ramen/actions/runs/5347722247
- https://github.com/RamenDR/ramen/actions/runs/5347705377
- https://github.com/RamenDR/ramen/actions/runs/5347594713
- https://github.com/RamenDR/ramen/actions/runs/5347589593
- More: https://github.com/RamenDR/ramen/actions?query=is%3Afailure
|
@nirs If you are confident about the changes, then we don't need any change in the code; but please provide references to the failures such that podman can be enabled for drenv tests later by verifying that the issues are resolved. |
The podman driver is flakeky locally and consistently failing in github actions now. Switching to docker fix the issues in github. Developers need to install docker to run drenv test locally. I tried to use the podman-docker pacakge but it does not emulate docker good enough for minikube. Signed-off-by: Nir Soffer <nsoffer@redhat.com>
All the changes are useful regardless of replacing pomand with docker for the drenv
Provided in the other comment. |
No chance that the volume was created by the test, the only place calling drenv Even if we assume that pytest is bugy and this is called more than once, starting I think this is an issue with the podman driver, maybe only on Ubuntu since locally The most important thing now is to unbreak our CI - once the CI is green again we have I filed a minikube bug for this issue, hopefully they will have some ideas how to |
Recently we see random failures in drenv tests when starting the vm. Try to debug
and avoid the failures.