Typecheck is still hitting memory limits semi-regularly on periodic CI
jobs. This bumps the default parallelism down to 3 from 4 to make it
slightly less memory intensive.
Signed-off-by: hasheddan <georgedanielmangum@gmail.com>
Some of the tests are negative test cases which are supposed to ensure that those
invalid usecases are handled properly.
However, some of the tests are false positives, they can pass due to various reasons.
One such example is: "should fail substituting values in a volume subpath with absolute path".
This test can pass if:
- the Pod cannot start due to various reasons (e.g.: the container image cannot be pulled or does
not exist).
- the Pod ran to completion, even though the container was not supposed to start in the first place.
- Revert "Fix integration test flake on TestFilter and TestPostFilter"
This reverts commit 94fc18c2dc.
- Relax checking logic on expected Filter/PostFilter counters.
- Move "ForgetPod" after "RunReservePluginsUnreserve", so that the cache would hold the pod to
avoid it's being retried simutaneously until Unreserve is completed.
- Move "assume" ahead of "RunReservePluginsReserve". This is based on the fact that "ForgetPod" is
the last step of failure path, so "assume" should be reversly treated as the first step. The
current failure path is like this:
assume -> reserve -> unreserve -> forgetPod -> recordingFailure
- Make subtests of TestReservePluginUnreserve stateless
When these tests failed it was unclear that the reason for the failure
could have been more EndpointSlices than expected. It was also unclear
what EndpointSlices were actually found when that occurred. This fixes
both of those issues.
There's currently no way to know whether an error is for SCTP or
UDP, for example:
Jul 24 09:55:54.469: INFO: netserver-0[e2e-nettest-3476].container[webserver].log
2020/07/24 09:53:52 Started UDP server
2020/07/24 09:53:52 Error occurred. error:protocol not supported
In this case the "Error occurred. error:protocol not supported" is
actually for the SCTP socket. Make that more apparent.
This adjusts tests that were waiting for Pods to be ready to instead
just wait for them to have IPs assigned to them. This relies on the
associated publishNotReadyAddresses field on Services. Additionally this
increases the the length of time we'll wait for EndpointSlices to be garbage
collected from 12s to 30s. Finally, this adds additional logging to
ExpectNoError calls so it's easier to understand where and why a test
failed.
