- Move from the old github.com/golang/glog to k8s.io/klog
- klog as explicit InitFlags() so we add them as necessary
- we update the other repositories that we vendor that made a similar
change from glog to klog
* github.com/kubernetes/repo-infra
* k8s.io/gengo/
* k8s.io/kube-openapi/
* github.com/google/cadvisor
- Entirely remove all references to glog
- Fix some tests by explicit InitFlags in their init() methods
Change-Id: I92db545ff36fcec83afe98f550c9e630098b3135
Automatic merge from submit-queue (batch tested with PRs 59683, 59964, 59841, 59936, 59686). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Reevaluate eviction thresholds after reclaim functions
**What this PR does / why we need it**:
When the node comes under `DiskPressure` due to inodes or disk space, the eviction manager runs garbage collection functions to clean up dead containers and unused images.
Currently, we use the strategy of trying to measure the disk space and inodes freed by garbage collection. However, as #46789 and #56573 point out, there are gaps in the implementation that can cause extra evictions even when they are not required. Furthermore, for nodes which frequently cycle through images, it results in a large number of evictions, as running out of inodes always causes an eviction.
This PR changes this strategy to call the garbage collection functions and ignore the results. Then, it triggers another collection of node-level metrics, and sees if the node is still under DiskPressure.
This way, we can simply observe the decrease in disk or inode usage, rather than trying to measure how much is freed.
**Which issue(s) this PR fixes**:
Fixes#46789Fixes#56573
Related PR #56575
**Special notes for your reviewer**:
This will look cleaner after #57802 removes arguments from [makeSignalObservations](https://github.com/kubernetes/kubernetes/pull/57802/files#diff-9e5246d8c78d50ce4ba440f98663f3e9R719).
**Release note**:
```release-note
NONE
```
/sig node
/kind bug
/priority important-soon
cc @kubernetes/sig-node-pr-reviews
Automatic merge from submit-queue (batch tested with PRs 57136, 59920). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Updated PID pressure node condition.
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
part of #54313
**Release note**:
```release-note
Updated PID pressure node condition
```
Primarily, this gives a way to explicitly disable eviction, which is
necessary to use omitempty on EvictionHard.
See: https://github.com/kubernetes/kubernetes/pull/53833#discussion_r166672137
As justification for this approach, neither 0% nor 100% make sense as
eviction thresholds; in the "less-than" case, you can't have less than
0% of a resource and 100% perpetually evicts; in the
"greater-than" case (assuming we ever add a resource with this
semantic), the reasoning is the reverse (not more than 100%, 0%
perpetually evicts).
This moves plugin/pkg/scheduler to pkg/scheduler and
plugin/cmd/kube-scheduler to cmd/kube-scheduler.
Bulk of the work was done with gomvpkg, except for kube-scheduler main
package.
- Changes the following KubeletConfiguration fields from `string` to
`map[string]string`:
- `EvictionHard`
- `EvictionSoft`
- `EvictionSoftGracePeriod`
- `EvictionMinimumReclaim`
- Adds flag parsing shims to maintain Kubelet's public flags API, while
enabling structured input in the file API.
- Also removes `kubeletconfig.ConfigurationMap`, which was an ad-hoc flag
parsing shim living in the kubeletconfig API group, and replaces it
with the `MapStringString` shim introduced in this PR. Flag parsing
shims belong in a common place, not in the kubeletconfig API.
I manually audited these to ensure that this wouldn't cause errors
parsing the command line for syntax that would have previously been
error free (`kubeletconfig.ConfigurationMap` was unique in that it
allowed keys to be provided on the CLI without values. I believe this was
done in `flags.ConfigurationMap` to facilitate the `--node-labels` flag,
which rightfully accepts value-free keys, and that this shim was then
just copied to `kubeletconfig`). Fortunately, the affected fields
(`ExperimentalQOSReserved`, `SystemReserved`, and `KubeReserved`) expect
non-empty strings in the values of the map, and as a result passing the
empty string is already an error. Thus requiring keys shouldn't break
anyone's scripts.
- Updates code and tests accordingly.
Regarding eviction operators, directionality is already implicit in the
signal type (for a given signal, the decision to evict will be made when
crossing the threshold from either above or below, never both). There is
no need to expose an operator, such as `<`, in the API. By changing
`EvictionHard` and `EvictionSoft` to `map[string]string`, this PR
simplifies the experience of working with these fields via the
`KubeletConfiguration` type. Again, flags stay the same.
Other things:
- There is another flag parsing shim, `flags.ConfigurationMap`, from the
shared flag utility. The `NodeLabels` field still uses
`flags.ConfigurationMap`. This PR moves the allocation of the
`map[string]string` for the `NodeLabels` field from
`AddKubeletConfigFlags` to the defaulter for the external
`KubeletConfiguration` type. Flags are layered on top of an internal
object that has undergone conversion from a defaulted external object,
which means that previously the mere registration of flags would have
overwritten any previously-defined defaults for `NodeLabels` (fortunately
there were none).
Automatic merge from submit-queue (batch tested with PRs 44596, 52708, 53163, 53167, 52692). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Fix the bad code comment and make the format unify
**What this PR does / why we need it**:
Fix the bad code comment and make the format unify
**Which issue this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close that issue when PR gets merged)*: fixes #
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 51337, 47080, 52646, 52635, 52666). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>..
Map a resource to multiple signals in eviction manager
It is possible to have multiple signals that point to the same type of
resource, e.g., both SignalNodeFsAvailable and
SignalAllocatableNodeFsAvailable refer to the same resource NodeFs.
Change the map from map[v1.ResourceName]evictionapi.Signal to
map[v1.ResourceName][]evictionapi.Signal
**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#52661
**Special notes for your reviewer**:
**Release note**:
```release-note
```
It is possible to have multiple signals that point to the same type of
resource, e.g., both SignalNodeFsAvailable and
SignalAllocatableNodeFsAvailable refer to the same resource NodeFs.
Change the map from map[v1.ResourceName]evictionapi.Signal to
map[v1.ResourceName][]evictionapi.Signal
This PR fixes the following issues:
1. Use ResourceStorageScratch instead of ResourceStorage API to represent
local storage capacity
2. In eviction manager, use container manager instead of node provider
(kubelet) to retrieve the node capacity and reserved resources. Node
provider (kubelet) has a feature gate so that storagescratch information
may not be exposed if feature gate is not set. On the other hand,
container manager has all the capacity and allocatable resource
information.
This PR adds the support for allocatable local storage (scratch space).
This feature is only for root file system which is shared by kubernetes
componenets, users' containers and/or images. User could use
--kube-reserved flag to reserve the storage for kube system components.
If the allocatable storage for user's pods is used up, some pods will be
evicted to free the storage resource.
Automatic merge from submit-queue (batch tested with PRs 46501, 45944, 46473)
fix func comment in helpers.go
**What this PR does / why we need it**:
fix func comment in helpers.go
**Which issue this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close that issue when PR gets merged)*: fixes #
NONE
**Special notes for your reviewer**:
NONE
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 42369, 42375, 42397, 42435, 42455)
[Bug Fix]: Avoid evicting more pods than necessary by adding Timestamps for fsstats and ignoring stale stats
Continuation of #33121. Credit for most of this goes to @sjenning. I added volume fs timestamps.
**why is this a bug**
This PR attempts to fix part of https://github.com/kubernetes/kubernetes/issues/31362 which results in multiple pods getting evicted unnecessarily whenever the node runs into resource pressure. This PR reduces the chances of such disruptions by avoiding reacting to old/stale metrics.
Without this PR, kubernetes nodes under resource pressure will cause unnecessary disruptions to user workloads.
This PR will also help deflake a node e2e test suite.
The eviction manager currently avoids evicting pods if metrics are old. However, timestamp data is not available for filesystem data, and this causes lots of extra evictions.
See the [inode eviction test flakes](https://k8s-testgrid.appspot.com/google-node#kubelet-flaky-gce-e2e) for examples.
This should probably be treated as a bugfix, as it should help mitigate extra evictions.
cc: @kubernetes/sig-storage-pr-reviews @kubernetes/sig-node-pr-reviews @vishh @derekwaynecarr @sjenning
David Ashpole