Better integration tests#2406
Conversation
dgageot
commented
Jul 4, 2019
- Switch to latest TravisCI env (xenial) because it seems faster and has more disk space for integration tests
- Print more information when an integration test goes wrong
- Re-enable TestDev that wasn't flaky
- Increase the duration of the test suite to increase the chance of running all the tests when one goes wrong.
Signed-off-by: David Gageot <david@gageot.net>
Signed-off-by: David Gageot <david@gageot.net>
Signed-off-by: David Gageot <david@gageot.net>
Signed-off-by: David Gageot <david@gageot.net>
Signed-off-by: David Gageot <david@gageot.net>
d6761d6 to
b056811
Compare
|
Green integration tests: 7 times |
70d995a to
eaf3532
Compare
Signed-off-by: David Gageot <david@gageot.net>
eaf3532 to
57fb190
Compare
|
@corneliusweig would you mind taking a look at this PR? I'm trying to fix all known flaky tests |
|
Sure, I can do that. |
corneliusweig
left a comment
There was a problem hiding this comment.
Looks good. I just have two questions.
| endif | ||
| GCP_ONLY=$(GCP_ONLY) go test -v $(REPOPATH)/integration -timeout 15m $(INTEGRATION_TEST_ARGS) | ||
| kubectl get nodes -oyaml | ||
| GCP_ONLY=$(GCP_ONLY) go test -v $(REPOPATH)/integration -timeout 20m $(INTEGRATION_TEST_ARGS) |
There was a problem hiding this comment.
I don't fully understand how this relates to your comment:
Increase the duration of the test suite to increase the chance of running all the tests when one goes wrong.
Do you mean that this gives enough time for the remaining tests to finish, so that they are not marked as failed?
There was a problem hiding this comment.
Exactly! It's so better said!
| select { | ||
| case <-ctx.Done(): | ||
| k.printDiskFreeSpace() | ||
| k.debug("nodes") |
There was a problem hiding this comment.
Does k.debug("nodes") add anything? After all, the nodes are already unconditionally printed in make integration (https://github.com/GoogleContainerTools/skaffold/pull/2406/files#diff-b67911656ef5d18c4ae36cb6741b7965R112).
There was a problem hiding this comment.
When the TestDebug used to fail, I could see an issue with the k8s nodes.
I added those logs before the test suite and in case of an error, to detect cases where the kind cluster might get broken after running a few tests.
There was a problem hiding this comment.
Ok, that makes sense.