Automatic merge from submit-queue
Increase provisioning test timeouts.
We've encountered flakes in our e2e infrastructure when kubelet took more than one minute to detach a volume used by a deleted pod.
Let's increase the wait period from 1 to 3 minutes. This slows down the test by 2 minutes, but it makes the test more stable.
In addition, when kubelet cannot detach a volume for 3 minutes, let the test wait for additional recycle controller retry interval (10 minutes) and hope the volume is deleted by then. This should not increase usual test time, it makes the test stable when kubelet is _extremely_ slow when releasing the volume.
Fixes: #24161
Automatic merge from submit-queue
Fix unintended change of Service.spec.ports[].nodePort during kubectl apply
Please refer #23551 for more detail. @bgrant0607 I think this simple fix should be ok to reuse nodePort. @thockin ptal.
Release note: Fix unintended change of `Service.spec.ports[].nodePort` during `kubectl apply`.
Automatic merge from submit-queue
Add RC and container pors to scheduler benchmark
Fix#23263
Ref #24408
However - scheduler throughput is still ~140 initially, whereas in reality we have 35-40. There are still significant difference we should understand.
@hongchaodeng @xiang90
Automatic merge from submit-queue
Cluster Verification Framework
I've spent the last few days looking at the general patterns of verification we have that we tend to reuse in the e2es. Basically, we need
- label filters
- forEach and WaitFor (where forEach doesn't necessarily waitFor anything).
- timeouts
- multiple phases (reusable definition of state)
- an extensible way to define cluster state that can evolve over time in a data object rather than as a set of parameters that have magic semantics
This PR
- implements the abstract above functionality declaratively, and w/o hidden semantics.
- addresses the sprawling duplicate methods in #23540, so that we can phase out the wrapper methods and replace them with well defined, extensible semantics for cluster state.
- fixes the recently discovered #23730 issue (where kubectl.go is relying on examples.go, which is obviously wacky) by using the new framework to implement forEachPod in just a couple of lines and migrating the wrapper function into framework.go.
There is some cleanup to do here, but this is seemingly working for a couple of use cases that are important (spark,cassandra,...,kubectl) tests. - i played with a few different ideas and this wound up seeming to be the most natural implementation from a usability standpoint...
in any case, just thought id push this up as a first iteration, open to feedback.
@kubernetes/sig-testing @timothysc
Automatic merge from submit-queue
Fix DNS test for larger clusters
On GKE, we scale the number of DNS pods based on the cluster size. For
testing on larger clusters, relax the DNS pod check.
Automatic merge from submit-queue
Incremental improvements to kubelet e2e tests
- Add keep-alive to ssh connection
- Don't try to stop services on image-based runs
- Increase jenkins ci timeout to 90 minutes to accomadate unpredictable go build times
- Remove spammy log statement
Automatic merge from submit-queue
Improve script for running scheduler benchmarks
Without this change, this script didn't work in my environment - it's making more consistent with other scripts.
@hongchaodeng @xiang90
- Add keep-alive to ssh connection
- Don't try to stop services on image-based runs
- Increase jenkins ci timeout to 90 minutes to accomadate unpredictable go build times
- Remove spammy log statement
Automatic merge from submit-queue
shared controller informers
Related to https://github.com/kubernetes/kubernetes/issues/14978
This demonstrates how controllers which use an `Informer`, would be able to share the same watch and store. A similar "setup and run" approach could be done for an `IndexInformer` to share that cache. I found adding listeners here to be easier than intercepting at the watch interface (problems with resourceVersion) or the reflector (same plumbing, but you have to fan out to multiple stores).
We could also use the cache we build here to back several of the admission plugins that currently run their own lookup caches today.
If there's interest, I can finish out the `SharedInformer` and switch the low hanging fruit over.
@kubernetes/rh-cluster-infra @smarterclayton @liggitt @wojtek-t
Automatic merge from submit-queue
Fix PullImage and add corresponding node e2e test
Fixes#24101. This is a bug introduced by #23506, since ref #23563.
The root cause of #24101 is described [here](https://github.com/kubernetes/kubernetes/issues/24101#issuecomment-208547623).
This PR
1) Fixes#24101 by decoding the messages returned during pulling image, and return error if any of the messages contains error.
2) Add the node e2e test to detect this kind of failure.
3) Get present check out of `ConformanceImage.Remove()` and `ConformanceImage.Pull()`. Because sometimes we may expect error to occur in `PullImage()` and `RemoveImage()`, but even that doesn't happen, the `Present()` check will still return error and let the test pass.
@yujuhong @freehan @liangchenye
Also /cc @resouer, because he is doing the image related functions refactoring.
- rebase: ForEach only on Running pods
- add waitFor step in guestbook describe and wrapper
- simplify logs in polling, make panic immediate, give rolluped stats in
the logs.
Improve logging for failure on ForEach
We've encountered flakes in our e2e infrastructure when kubelet took more than
one minute to detach a volume used by a deleted pod.
Let's increase the wait period from 1 to 3 minutes. This slows down the test
by 2 minutes, but it makes the test more stable.
In addition, when kubelet cannot detach a volume for 3 minutes, let the test
wait for additional recycle controller retry interval (10 minutes) and hope the
volume is deleted by then. This should not increase usual test time, it makes
the test stable when kubelet is _extremely_ slow when releasing the volume.
Automatic merge from submit-queue
e2e_node: port privileged pod tests from test/e2e/priviliged.go
The ported test is functionally the same as the original test.
The main difference between the two tests is that the original test relies on
`kubectl` to exec into the container, while the latter directly uses the REST
client of the apiserver. This avoids the need to copy kubectl to the node under
test.
Automatic merge from submit-queue
Move /resetMetrics to DELETE /metrics
Reduces the surface area of the API server slightly and allows
downstream components to have deleteable metrics. After this change
genericapiserver will *not* have metrics unless the caller defines it
(allows different apiserver implementations to make that choice on their
own).
@wojtek-t
Automatic merge from submit-queue
Make etcd cache size configurable
Instead of the prior 50K limit, allow users to specify a more sensible size for their cluster.
I'm not sure what a sensible default is here. I'm still experimenting on my own clusters. 50 gives me a 270MB max footprint. 50K caused my apiserver to run out of memory as it exceeded >2GB. I believe that number is far too large for most people's use cases.
There are some other fundamental issues that I'm not addressing here:
- Old etcd items are cached and potentially never removed (it stores using modifiedIndex, and doesn't remove the old object when it gets updated)
- Cache isn't LRU, so there's no guarantee the cache remains hot. This makes its performance difficult to predict. More of an issue with a smaller cache size.
- 1.2 etcd entries seem to have a larger memory footprint (I never had an issue in 1.1, even though this cache existed there). I suspect that's due to image lists on the node status.
This is provided as a fix for #23323
Automatic merge from submit-queue
Add watch.Until, a conditional watch mechanism
A more powerful tool than wait.Poll, allows a watch interface to drive conditionals to react to changes on a resource or resources. Provide a set of standard conditions that are in common use in the code, and updates e2e to use a few of these.
Extracted from #23567
Automatic merge from submit-queue
phase 2 of cassandra example overhaul
Here's the next iteration in overhauling this example, towards https://github.com/kubernetes/kubernetes/issues/20961. This removes the pod adoption part, but doesn't (yet) otherwise change any of the resources used.
It also includes some README cleanup, and removes some explicit specification of labels in the rc yaml.
This PR doesn't yet add any commentary on how we're using the seed provider (re: https://github.com/kubernetes/kubernetes/issues/20961#issuecomment-190405959 etc.). Maybe we should add that.
Also: LMK if this PR should include any changes to the links out to the docs.
cc @bgrant0607 @johndmulhausen
in e2e/volumes.go: give time to allow pod cleanup and volume unmount happen before volume server exit;
skip cinder volume test if not running with openstack provider
comment on why pause before containerized server is stopped in volume e2e tests, fix#24100
updates NFS server image to 0.6, per #22529
fix persistent_volume e2e test: test cleanup doesn't expect client pod; delete PV after test
Signed-off-by: Huamin Chen <hchen@redhat.com>
Reduces the surface area of the API server slightly and allows
downstream components to have deleteable metrics. After this change
genericapiserver will *not* have metrics unless the caller defines it
(allows different apiserver implementations to make that choice on their
own).
Automatic merge from submit-queue
Do not throw creation errors for containers that fail immediately after being started
Fixes (hopefully) #23607
cc @dchen1107
Some e2e tests use wget to check connectivity, and the default e2e
timeout is 900s. This change allows the timeout to be specified on a
check-by-check basis. This will also make the check useful for negative
checks (like those used by openshift to validate isolation) since a
short timeout is suggested where connectivity is not expected.
