Currently we have two plugin managers.
However one of them limits the cloud plugins it loads.
This means that if cloud provider is set to external the plugins will
not be loaded in *that* plugin manager. However they will be loaded in
the other instance of the plugin manager. So it does not actually save
us anything. It does hamper the efforts to actually get stage 1
separation working.
Automatic merge from submit-queue (batch tested with PRs 51574, 51534, 49257, 44680, 48836)
Task 1: Tainted node by condition.
**What this PR does / why we need it**:
Tainted node by condition for MemoryPressure, OutOfDisk and so on.
**Which issue this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close that issue when PR gets merged)*: part of #42001
**Release note**:
```release-note
Tainted nodes by conditions as following:
* 'node.kubernetes.io/network-unavailable=:NoSchedule' if NetworkUnavailable is true
* 'node.kubernetes.io/disk-pressure=:NoSchedule' if DiskPressure is true
* 'node.kubernetes.io/memory-pressure=:NoSchedule' if MemoryPressure is true
* 'node.kubernetes.io/out-of-disk=:NoSchedule' if OutOfDisk is true
```
Automatic merge from submit-queue (batch tested with PRs 51707, 51662, 51723, 50163, 51633)
update GC controller to wait until controllers have been initialized …
fixes#51013
Alternative to https://github.com/kubernetes/kubernetes/pull/51492 which keeps those few controllers (only one) from starting the informers early.
Automatic merge from submit-queue (batch tested with PRs 50775, 51397, 51168, 51465, 51536)
Enable batch/v1beta1.CronJobs by default
This PR moves to CronJobs beta entirely, enabling `batch/v1beta1` by default.
Related issue: #41039
@erictune @janetkuo ptal
```release-note
Promote CronJobs to batch/v1beta1.
```
Automatic merge from submit-queue (batch tested with PRs 49642, 50335, 50390, 49283, 46582)
Improve GC discovery sync performance
Improve GC discovery sync performance by only syncing when discovered
resource diffs are detected. Before, the GC worker pool was shut down
and monitors resynced unconditionally every sync period, leading to
significant processing delays causing test flakes where otherwise
reasonable GC timeouts were being exceeded.
Related to https://github.com/kubernetes/kubernetes/issues/49966.
/cc @kubernetes/sig-api-machinery-bugs
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 50016, 49583, 49930, 46254, 50337)
Break up node controller into packages
This change does NO actual code changes other than moving constituent
parts into packages.
```release-note
NONE
```
Improve GC discovery sync performance by only syncing when discovered
resource diffs are detected. Before, the GC worker pool was shut down
and monitors resynced unconditionally every sync period, leading to
significant processing delays causing test flakes where otherwise
reasonable GC timeouts were being exceeded.
Related to https://github.com/kubernetes/kubernetes/issues/49966.
Automatic merge from submit-queue (batch tested with PRs 50173, 50324, 50288, 50263, 50333)
Honor --use-service-account-credentials and warn when missing private key
Fixes#50275 by logging a warning and failing to start rather than continue to run ignoring the user's specified config
- Move public key functions to client-go/util/cert
- Move pki file helper functions to client-go/util/cert
- Standardize on certutil package alias
- Update dependencies to client-go/util/cert
Automatic merge from submit-queue (batch tested with PRs 49538, 49708, 47665, 49750, 49528)
Enable garbage collection of custom resources
Enhance the garbage collector to periodically refresh the resources it monitors (via discovery) to enable custom resource definition GC (addressing #44507 and reverting #47432).
This is a replacement for #46000.
/cc @lavalamp @deads2k @sttts @caesarxuchao
/ref https://github.com/kubernetes/kubernetes/pull/48065
```release-note
The garbage collector now supports custom APIs added via CustomeResourceDefinition or aggregated apiservers. Note that the garbage collector controller refreshes periodically, so there is a latency between when the API is added and when the garbage collector starts to manage it.
```
Automatic merge from submit-queue (batch tested with PRs 49538, 49708, 47665, 49750, 49528)
Use the core client with version
**What this PR does / why we need it**:
Replace the **deprecated** `clientSet.Core()` with `clientSet.CoreV1()`.
**Which issue this PR fixes**: fixes#49535
**Special notes for your reviewer**:
**Release note**:
```release-note
NONE
```
Enhance the garbage collector to periodically refresh the resources it
monitors (via discovery) to enable custom resource definition GC.
This implementation caches Unstructured structs for any kinds not
covered by a shared informer. The existing meta-only codec only supports
compiled types; an improved codec which supports arbitrary types could
be introduced to optimize caching to store only metadata for all
non-informer types.
Automatic merge from submit-queue (batch tested with PRs 49665, 49689, 49495, 49146, 48934)
make it possible to allow discovery errors for controllers
Update the discovery client to return partial discovery information *and* an error. Since we can aggregate API servers, discovery of some resources can fail independently. Callers of this function who want to tolerate the errors can, existing callers will still get an error and fail in normal blocks.
@kubernetes/sig-api-machinery-misc @sttts
Automatic merge from submit-queue (batch tested with PRs 48976, 49474, 40050, 49426, 49430)
Remove duplicated import and wrong alias name of api package
**What this PR does / why we need it**:
**Which issue this PR fixes**: fixes#48975
**Special notes for your reviewer**:
/assign @caesarxuchao
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 48224, 45431, 45946, 48775, 49396)
add reflector metrics
This adds metrics (optionally prometheus) to reflectors so that you can see when one reflector is behaving poorly and just how poorly its doing.
@eparis
```release-note
Adds metrics for checking reflector health.
```
Automatic merge from submit-queue (batch tested with PRs 49107, 47177, 49234, 49224, 49227)
tighten quota controller interface
While debugging a quota performance problem, I had to chase some references deeper than necessary because the interfaces were overly broad. This tightens them.
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 48377, 48940, 49144, 49062, 49148)
support fc volume attach and detach
**What this PR does / why we need it**:
Support FC volume attach and detach to enforce RWO access
**Which issue this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close that issue when PR gets merged)*: fixes#48953
**Special notes for your reviewer**:
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 47232, 48625, 48613, 48567, 39173)
Include leaderelection in client-go;
Fix#39117
Fix https://github.com/kubernetes/client-go/issues/28
This PR:
* includes the leaderelection to the staging client-go
* to avoid conflict with golang's testing package, renames package /testing to /testutil, and renames cache/testing to cache/testframework
```release-note
client-go now includes the leaderelection package
```
Automatic merge from submit-queue
controller-manager: fix horizontal-pod-autoscaler-use-rest-clients fl…
…ag help info
**What this PR does / why we need it**:
**Which issue this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close that issue when PR gets merged)*: fixes #
**Special notes for your reviewer**:
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue
Use endpoints informer for the endpoint controller
This substantially reduces the number of API calls made by the endpoint
controller. Currently the controller makes an API call per endpoint for
each service that is synced. When the 30s resync is triggered, this
results in an API call for every single endpoint in the cluster. This
quickly exceeds the default qps/burst limit of 20/30 even in small
clusters, leading to delays in endpoint updates.
This change modifies the controller to use the endpoint informer cache
for all endpoint GETs. This means we only make API calls for changes in
endpoints. As a result, qps only depends on the pod activity in the
cluster, rather than the number of services.
**What this PR does / why we need it**:
Address endpoint update delays as described in https://github.com/kubernetes/kubernetes/issues/47597.
**Which issue this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close that issue when PR gets merged)*: fixes #
https://github.com/kubernetes/kubernetes/issues/47597
**Special notes for your reviewer**:
**Release note**:
```release-note
```
This substantially reduces the number of API calls made by the endpoint
controller. Currently the controller makes an API call per endpoint for
each service that is synced. When the 30s resync is triggered, this
results in an API call for every single endpoint in the cluster. This
quickly exceeds the default qps/burst limit of 20/30 even in small
clusters, leading to delays in endpoint updates.
This change modifies the controller to use the endpoint informer cache
for all endpoint GETs. This means we only make API calls for changes in
endpoints. As a result, qps only depends on the pod activity in the
cluster, rather than the number of services.
Internal attach/detach controller timers should be configurable and tests
should use much shorter values.
reconcilerSyncDuration is deliberately left out of TimerConfig because it's
the only one that's not a constant one, it's configurable by user.
Automatic merge from submit-queue
bazel: stamp multiple packages by using x_defs instead of linkstamp in go_binary rules
**What this PR does / why we need it**: Fixes regression introduced sometime in the last few months that prevented bazel-built clusters from identifying version properly.
It does so by updating the bazelbuild/rules_go and kubernetes/repo-infra dependencies to support using stamp values in `go_binary` `x_defs`, and then changing our `go_binary` rules to use `x_defs` instead of `linkstamp`.
This whole charade is necessary because we need to stamp version information in multiple packages.
This pretty much only affects the bazel build, so it should be low risk.
**Which issue this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close that issue when PR gets merged)*: fixes#45298
**Special notes for your reviewer**: depends on https://github.com/kubernetes/repo-infra/pull/18; should not be merged before it.
**Release note**:
```release-note
NONE
```
/assign @spxtr @mikedanese
Automatic merge from submit-queue (batch tested with PRs 44883, 46836, 46765, 46683, 46050)
While deleting a namespace, the TPR instances under this ns should be…
… deleted.
While deleting a namespace, the TPR instances under this ns should be deleted.
Fixed#46736
**Release note**:
```
None
```
Automatic merge from submit-queue (batch tested with PRs 40760, 46706, 46783, 46742, 46751)
complete the controller context for init funcs
This completes the conversion to initFuncs for the controller initialization to make easier and more manageable to add them.
1. Create controllerrevisions (history) and label pods with template
hash for both RollingUpdate and OnDelete update strategy
2. Clean up old, non-live history based on revisionHistoryLimit
3. Remove duplicate controllerrevisions (the ones with the same template)
and relabel their pods
4. Update RBAC to allow DaemonSet controller to manage
controllerrevisions
5. In DaemonSet controller unit tests, create new pods with hash labels
Automatic merge from submit-queue (batch tested with PRs 46076, 43879, 44897, 46556, 46654)
Local storage plugin
**What this PR does / why we need it**:
Volume plugin implementation for local persistent volumes. Scheduler predicate will direct already-bound PVCs to the node that the local PV is at. PVC binding still happens independently.
**Which issue this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close that issue when PR gets merged)*:
Part of #43640
**Release note**:
```
Alpha feature: Local volume plugin allows local directories to be created and consumed as a Persistent Volume. These volumes have node affinity and pods will only be scheduled to the node that the volume is at.
```
Automatic merge from submit-queue (batch tested with PRs 45514, 45635)
refactor certificate controller to break it into two parts
Break pkg/controller/certificates into:
* pkg/controller/certificates/approver: containing the group approver
* pkg/controller/certificates/signer: containing the local signer
* pkg/controller/certificates: containing shared infrastructure
```release-note
Break the 'certificatesigningrequests' controller into a 'csrapprover' controller and 'csrsigner' controller.
```
Allow the list of resources the garbage collector controller should
ignore to be customizable, so downstream integrators can add their own
resources to the list, if necessary.
The garbage collector controller currently needs to list, watch, get,
patch, update, and delete resources. Update the criteria for
deletable resources to reflect this.
Tokens controller previously needed a bit of extra help in order to be
safe for concurrent use. The new MutationCache allows it to keep a local
cache and still use a shared informer. The filtering event handler lets
it only see changes to secrets it cares about.
Automatic merge from submit-queue (batch tested with PRs 45374, 44537, 45739, 44474, 45888)
Allow kcm and scheduler to lock on ConfigMaps.
**What this PR does / why we need it**:
Plumbs through the ability to lock on ConfigMaps through the kcm and scheduler.
**Which issue this PR fixes**
Fixes: #44857
Addresses issues with: #45415
**Special notes for your reviewer**:
**Release note**:
```
Add leader-election-resource-lock support to kcm and scheduler to allow for locking on ConfigMaps as well as Endpoints(default)
```
/cc @kubernetes/sig-cluster-lifecycle-pr-reviews @jamiehannaford @bsalamat @mikedanese
Automatic merge from submit-queue (batch tested with PRs 45382, 45384, 44781, 45333, 45543)
Copy internal types to metrics
Supersedes #45306.
#45306 removed the internal types and suggested whoever needs the internal types should define their own copy, and use the code-gen tools to generated the conversion functions. Per offline discussion with @DirectXMan12, we wanted to go that direction but it's not clear where to put the internal types yet. Hence, as a temporary solution, we decided copy the referred client-go/pkg/api types into metrics api to avoid the dependency.
The commit "remove need of registry from custom_metrics/client.go" is similar to what I did to the fake client in an earlier PR. Let me know if you want to put the commit in another PR.
Automatic merge from submit-queue (batch tested with PRs 45304, 45006, 45527)
increase the QPS for namespace controller
The namespace controller is really chatty. Especially to discovery since that involves two requests for every API version available. This bumps the QPS and burst on the namespace controller to avoid being stuck waiting.
Automatic merge from submit-queue (batch tested with PRs 44862, 42241, 42101, 43181, 44147)
Feature/hpa upscale downscale delay configurable
**What this PR does / why we need it**:
Makes "upscale forbidden window" and "downscale forbidden window" duration configurable in arguments of kube-controller-manager. Those are options of horizontal pod autoscaler.
**Special notes for your reviewer**:
Please have a look @DirectXMan12 , the PR as discussed in Slack.
**Release note**:
```
Make "upscale forbidden window" and "downscale forbidden window" duration configurable in arguments of kube-controller-manager. Those are options of horizontal pod autoscaler. Right now are hardcoded 3 minutes for upscale, and 5 minutes to downscale. But sometimes cluster administrator might want to change this for his own needs.
```
Automatic merge from submit-queue
Add support for IP aliases for pod IPs (GCP alpha feature)
```release-note
Adds support for allocation of pod IPs via IP aliases.
# Adds KUBE_GCE_ENABLE_IP_ALIASES flag to the cluster up scripts (`kube-{up,down}.sh`).
KUBE_GCE_ENABLE_IP_ALIASES=true will enable allocation of PodCIDR ips
using the ip alias mechanism rather than using routes. This feature is currently
only available on GCE.
## Usage
$ CLUSTER_IP_RANGE=10.100.0.0/16 KUBE_GCE_ENABLE_IP_ALIASES=true bash -x cluster/kube-up.sh
# Adds CloudAllocator to the node CIDR allocator (kubernetes-controller manager).
If CIDRAllocatorType is set to `CloudCIDRAllocator`, then allocation
of CIDR allocation instead is done by the external cloud provider and
the node controller is only responsible for reflecting the allocation
into the node spec.
- Splits off the rangeAllocator from the cidr_allocator.go file.
- Adds cloudCIDRAllocator, which is used when the cloud provider allocates
the CIDR ranges externally. (GCE support only)
- Updates RBAC permission for node controller to include PATCH
```
Automatic merge from submit-queue
Remove alphaProvisioner in PVController and AlphaStorageClassAnnotation
remove alpha annotation and alphaProvisioner
**Release note**:
```release-note
NONE
```
If CIDRAllocatorType is set to `CloudCIDRAllocator`, then allocation
of CIDR allocation instead is done by the external cloud provider and
the node controller is only responsible for reflecting the allocation
into the node spec.
- Splits off the rangeAllocator from the cidr_allocator.go file.
- Adds cloudCIDRAllocator, which is used when the cloud provider allocates
the CIDR ranges externally. (GCE support only)
- Updates RBAC permission for node controller to include PATCH
Automatic merge from submit-queue (batch tested with PRs 43144, 42671, 43226, 43314, 43361)
don't start controllers against unhealthy master
Operating against an unhealthy apiserver is unpredictable. Some clients like `kubectl` need to be best effort in this regard so that you can debug broken apiservers. Controllers shouldn't run against unhealthy masters.
Automatic merge from submit-queue (batch tested with PRs 43642, 43170, 41813, 42170, 41581)
Enable storage class support in Azure File volume
**What this PR does / why we need it**:
Support StorageClass in Azure file volume
**Which issue this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close that issue when PR gets merged)*: fixes #
**Special notes for your reviewer**:
**Release note**:
```release-note
Support StorageClass in Azure file volume
```
Automatic merge from submit-queue (batch tested with PRs 42692, 42169, 42173)
DaemonSet: Respect ControllerRef
**What this PR does / why we need it**:
This is part of the completion of the [ControllerRef](https://github.com/kubernetes/community/blob/master/contributors/design-proposals/controller-ref.md) proposal. It brings DaemonSet into full compliance with ControllerRef. See the individual commit messages for details.
**Which issue this PR fixes**:
This ensures that DaemonSet does not fight with other controllers over control of Pods.
**Special notes for your reviewer**:
**Release note**:
```release-note
DaemonSet now respects ControllerRef to avoid fighting over Pods.
```
cc @erictune @kubernetes/sig-apps-pr-reviews
Automatic merge from submit-queue (batch tested with PRs 42692, 42169, 42173)
Add pprof trace support
Add support for `/debug/pprof/trace`
Can wait for master to reopen for 1.7.
cc @smarterclayton @wojtek-t @gmarek @timothysc @jeremyeder @kubernetes/sig-scalability-pr-reviews
This commit switches over the HPA controller to use the custom metrics
API. It also converts the HPA controller to use the generated client
in k8s.io/metrics for the resource metrics API.
In order to enable support, you must enable
`--horizontal-pod-autoscaler-use-rest-clients` on the
controller-manager, which will switch the HPA controller's MetricsClient
implementation over to use the standard rest clients for both custom
metrics and resource metrics. This requires that at the least resource
metrics API is registered with kube-aggregator, and that the controller
manager is pointed at kube-aggregator. For this to work, Heapster
must be serving the new-style API server (`--api-server=true`).
- Add a new type PortworxVolumeSource
- Implement the kubernetes volume plugin for Portworx Volumes under pkg/volume/portworx
- The Portworx Volume Driver uses the libopenstorage/openstorage specifications and apis for volume operations.
Changes for k8s configuration and examples for portworx volumes.
- Add PortworxVolume hooks in kubectl, kube-controller-manager and validation.
- Add a README for PortworxVolume usage as PVs, PVCs and StorageClass.
- Add example spec files
Handle code review comments.
- Modified READMEs to incorporate to suggestions.
- Add a test for ReadWriteMany access mode.
- Use util.UnmountPath in TearDown.
- Add ReadOnly flag to PortworxVolumeSource
- Use hostname:port instead of unix sockets
- Delete the mount dir in TearDown.
- Fix link issue in persistentvolumes README
- In unit test check for mountpath after Setup is done.
- Add PVC Claim Name as a Portworx Volume Label
Generated code and documentation.
- Updated swagger spec
- Updated api-reference docs
- Updated generated code under pkg/api/v1
Godeps update for Portworx Volume Driver
- Adds github.com/libopenstorage/openstorage
- Adds go.pedge.io/pb/go/google/protobuf
- Updates Godep Licenses
Automatic merge from submit-queue (batch tested with PRs 42053, 41282, 42056, 41663, 40927)
Fully remove hand-written listers and informers
Note: the first commit is from #41927. Adding do-not-merge for now as we'll want that to go in first, and then I'll rebase this on top.
Update statefulset controller to use a lister for PVCs instead of a client request. Also replace a unit test's dependency on legacylisters with the generated ones. cc @kargakis @kow3ns @foxish @kubernetes/sig-apps-pr-reviews
Remove all references to pkg/controller/informers and pkg/client/legacylisters, and remove those packages.
@smarterclayton @deads2k this should be it!
cc @gmarek @wojtek-t @derekwaynecarr @kubernetes/sig-scalability-pr-reviews
Automatic merge from submit-queue
Make controller-manager resilient to stale serviceaccount tokens
Now that the controller manager is spinning up controller loops using service accounts, we need to be more proactive in making sure the clients will actually work.
Future additional work:
* make a controller that reaps invalid service account tokens (c.f. https://github.com/kubernetes/kubernetes/issues/20165)
* allow updating the client held by a controller with a new token while the controller is running (c.f. https://github.com/kubernetes/kubernetes/issues/4672)
Automatic merge from submit-queue
Switch service controller to shared informers
Originally part of #40097
cc @deads2k @smarterclayton @gmarek @wojtek-t @timothysc @sttts @liggitt @kubernetes/sig-scalability-pr-reviews
Automatic merge from submit-queue
Remove alpha provisioning
This is the first part of https://github.com/kubernetes/features/issues/36
@kubernetes/sig-storage-misc
**Release note**:
```release-note
Alpha version of dynamic volume provisioning is removed in this release. Annotation
"volume.alpha.kubernetes.io/storage-class" does not have any special meaning. A default storage class
and DefaultStorageClass admission plugin can be used to preserve similar behavior of Kubernetes cluster,
see https://kubernetes.io/docs/user-guide/persistent-volumes/#class-1 for details.
```
Automatic merge from submit-queue
Switch serviceaccounts controller to generated shared informers
Originally part of #40097
cc @deads2k @sttts @liggitt @smarterclayton @gmarek @wojtek-t @timothysc @kubernetes/sig-scalability-pr-reviews
Automatic merge from submit-queue
Change default attach_detach_controller reconciler sync period to 1 minute
When default reconciler sync period is set to 5 second, we often see
rateLimit issue for a large cluster. This PR is changing the period to 1
minute to mitigate this problem.
Make this period longer means that there might be some period of time
that the cached information in master's attach_detach_controller is out
of date. The node might use this information to mount to the wrong
device. For GCE PD, since device path is uniquely associated with volume
id, so mount operation will just fail because of this outdated
information. For AWS, before kubelet might mount to the wrong volume
because device path could be reused immediately once it is available.
But after PR #38818, device path will only be reused after all device
paths have been explored. That means it is very unlikely that kubelet will
mount to a wrong volume that is using the old device path that had been
assigned to the same node.
**Release note**:
```release-note
We change the default attach_detach_controller sync period to 1 minute to reduce the query frequency through cloud provider to check whether volumes are attached or not.
```
When default reconciler sync period is set to 5 second, we often see
rateLimit issue for a large cluster. This PR is change the period to 1
minute to mitigate this problem.
Make this period longer means that there might be some period of time
that the cached information in master's attach_detach_controller is out
of date. The node might use this information to mount to the wrong
device. For GCE PD, since device path is uniquely associated with volume
id, so mount operation will just fail because of this outdated
information. For AWS, before kubelet might mount to the wrong volume
because device path could be reused immediately once it is available.
But after PR #38818, device path will only be reused after all device
paths have been explored. That means it is very unlikely that kubelet will
mount to a wrong volume that is using the old device path that had been
assigned to the same node.
Automatic merge from submit-queue (batch tested with PRs 41137, 41268)
Allow the CertificateController to use any Signer implementation.
**What this PR does / why we need it**:
This will allow developers to create `CertificateController`s with arbitrary `Signer`s, instead of forcing the use of `CFSSLSigner`. It matches the behavior of allowing an arbitrary `AutoApprover` to be passed in the constructor.
**Release note**:
```release-note
NONE
```
CC @mikedanese
Automatic merge from submit-queue (batch tested with PRs 39418, 41175, 40355, 41114, 32325)
TaintController
```release-note
This PR adds a manager to NodeController that is responsible for removing Pods from Nodes tainted with NoExecute Taints. This feature is beta (as the rest of taints) and enabled by default. It's gated by controller-manager enable-taint-manager flag.
```
Automatic merge from submit-queue (batch tested with PRs 39418, 41175, 40355, 41114, 32325)
ResyncPeriod Comment
ResyncPeriod Comment:
// ResyncPeriod returns a function which generates a duration each time it is
// invoked; this is so that multiple controllers don't get into lock-step and all
// hammer the apiserver with list requests simultaneously.
Automatic merge from submit-queue (batch tested with PRs 40796, 40878, 36033, 40838, 41210)
Implement TTL controller and use the ttl annotation attached to node in secret manager
For every secret attached to a pod as volume, Kubelet is trying to refresh it every sync period. Currently Kubelet has a ttl-cache of secrets of its pods and the ttl is set to 1 minute. That means that in large clusters we are targetting (5k nodes, 30pods/node), given that each pod has a secret associated with ServiceAccount from its namespaces, and with large enough number of namespaces (where on each node (almost) every pod is from a different namespace), that resource in ~30 GETs to refresh all secrets every minute from one node, which gives ~2500QPS for GET secrets to apiserver.
Apiserver cannot keep up with it very easily.
Desired solution would be to watch for secret changes, but because of security we don't want a node watching for all secrets, and it is not possible for now to watch only for secrets attached to pods from my node.
So as a temporary solution, we are introducing an annotation that would be a suggestion for kubelet for the TTL of secrets in the cache and a very simple controller that would be setting this annotation based on the cluster size (the large cluster is, the bigger ttl is).
That workaround mean that only very local changes are needed in Kubelet, we are creating a well separated very simple controller, and once watching "my secrets" will be possible it will be easy to remove it and switch to that. And it will allow us to reach scalability goals.
@dchen1107 @thockin @liggitt
Automatic merge from submit-queue (batch tested with PRs 40345, 38183, 40236, 40861, 40900)
refactor approver and signer interfaces to be consisten w.r.t. apiserver interaction
This makes it so that only the controller loop talks to the
API server directly. The signatures for Sign and Approve also
become more consistent, while allowing the Signer to report
conditions (which it wasn't able to do before).
This makes it so that only the controller loop talks to the
API server directly. The signatures for Sign and Approve also
become more consistent, while allowing the Signer to report
conditions (which it wasn't able to do before).
huangjiuyuan