This patch fixes an issue where best-effort pods were not admitted
to the node if the single-numa-node policy was set.
This was because the Admit policy in single-numa-node policy does
not admit any pod where the hint is anything but single NUMA node. The 'best hint' in this case is {<set bits for num. Numa Nodes on machine>, true}
So on a machine with 2 NUMA nodes the best hint for a best-effort pod is {11,true} as best-effort pods have no Topology preferences.
The single-numa-node policy fails any pod with a not preferred hint OR a hint where > 1 bits are set, thus the above example resulting in termintaed pods with a Topology Affinity Error.
This is a short term fix for the single-numa-node policy, as there will be code refactoring for the 1.17 release.
This patch fixes an issue in the TopologyManager that wouldn't allow
pods to be admitted if pods were launched with the SingleNUMANode policy
and any of the hint providers had no NUMA preferences.
This is due to 2 factors:
1) Any hint provider that passes back a `nil` as its hints, has its hint
automatically transformed into a single {11 true} hint before merging
2) We added a special casing for the SingleNumaNodePolicy() in the
TopologyManager that essentially turns these hints into a
{11 false} anytime a {11 true} is seen.
The current patch reworks this logic so the that TopologyManager can
tell the difference between a "don't care" hint and a true "{11 true}"
hint returned by the hint provider. Only true "{11 true}" hints will be
converted by the special casing for the SingleNumaNodePolicy(), while
"don't care" hints will not.
This is a short term fix for this issue until we do a larger refactoring
of this code for the 1.17 release.
- As discussed in reviews and other public channels,
this abstraction is used to represent numa nodes, not sockets.
- There is nothing inherently related to sockets in this package anyway.
Added one off fix for single-numa-node policy to correctly
reject pod admission on a resource allocation that spans
NUMA nodes
Co-authored-by: Kevin Klues <kklues@nvidia.com>
Previously it only took a bool, which limited the logic it could perform
to determine if a pod should be admitted or not based on the merged hint
from the policy.
As part of this, update the logic to use the NUMA information instead of
the Socket information when generating and consuming TopologyHints in
the CPUManager.
Unfortunately, the NUMA information is not readily available from
cadvisor, so we have to roll the logic to discover it by hand. In the
future, we should remove this custiom code to use the information
provided by cadvisor once it is made available.
At present, there is no way for a hint provider to return distinct hints
for different resource types via a call to GetTopologyHints(). This
means that hint providers that govern multiple resource types (e.g. the
devicemanager) must do some sort of "pre-merge" on the hints it
generates for each resource type before passing them back to the
TopologyManager.
This patch changes the GetTopologyHints() interface to allow a hint
provider to pass back raw hints for each resource type, and allow the
TopologyManager to merge them using a single unified strategy.
This change also allows the TopologyManager to recognize which
resource type a set of hints originated from, should this information
become useful in the future.
isKernelPid should explicitly check the error returned from os.Readlink and return true
only if the error value is ENOENT. Without this fix, if Readlink
returned say ENAMETOOLONG or EACESS, we would still count the process as
a kernel process (which is not true).
The container manager used in kubelet checks for docker daemon process either via pidfile
or process name. While the pidfile points to the docker daemon process PID, the
dockerProcessName constant stores a docker cli name ( docker ) instead of docker daemon
name ( dockerd ).
This patch fixes a bug in the CPUManager, whereby it doesn't honor the
"effective requests/limits" of a Pod as defined by:
https://kubernetes.io/docs/concepts/workloads/pods/init-containers/#resources
The rule states that a Pod’s "effective request/limit" for a resource
should be the larger of:
* The highest of any particular resource request or limit
defined on all init Containers
* The sum of all app Containers request/limit for a
resource
Moreover, the rule states that:
* The effective QoS tier is the same for init Containers
and app containers alike
This means that the resource requests of init Containers and app
Containers should be able to overlap, such that the larger of the two
becomes the "effective resource request/limit" for the Pod. Likewise,
if a QoS tier of "Guaranteed" is determined for the Pod, then both init
Containers and app Containers should run in this tier.
In its current implementation, the CPU manager honors the effective QoS
tier for both init and app containers, but doesn't honor the "effective
request/limit" correctly.
Instead, it treats the "effective request/limit" as:
* The sum of all init Containers plus the sum of all app
Containers request/limit for a resource
It does this by not proactively removing the CPUs given to previous init
containers when new containers are being created. In the worst case,
this causes the CPUManager to give non-overlapping CPUs to all
containers (whether init or app) in the "Guaranteed" QoS tier before any
of the containers in the Pod actually start.
This effectively blocks these Pods from running if the total number of
CPUs being requested across init and app Containers goes beyond the
limits of the system.
This patch fixes this problem by updating the CPUManager static policy
so that it proactively removes any guaranteed CPUs it has granted to
init Containers before allocating CPUs to app containers. Since all init
container are run sequentially, it also makes sure this proactive
removal happens for previous init containers when allocating CPUs to
later ones.
Previously, the topologymanager would simply fall back to the None() policy
if an invalid policy was specified. This patch updates this to return an
error when an invalid policy is passed, forcing the kubelet to fail
fast when this occurs.
These semantics should be preferable because an invalid policy likely
indicates operator error in setting the policy flag on the kubelet
correctly (e.g. misspelling 'strict' as 'striict'). In this case it is
better to fail fast so the operator can detect this and correct the
mistake, than to mask the error and essentially disable the
topologymanager unexpectedly.
Previously, the cpumanager would simply fall back to the None() policy
if an invalid policy was specified. This patch updates this to return an
error when an invalid policy is passed, forcing the kubelet to fail
fast when this occurs.
These semantics should be preferable because an invalid policy likely
indicates operator error in setting the policy flag on the kubelet
correctly (e.g. misspelling 'static' as 'statiic'). In this case it is
better to fail fast so the operator can detect this and correct the
mistake, than to mask the error and essentially disable the cpumanager
unexpectedly.
Use the exported list from runc that uses "KB" and not "kB".
This issue breaks kubelet on AArch64 (arm 64).
var HugePageSizeUnitList = []string{"B", "KB", "MB", "GB", "TB", "PB"}
The hugetlb cgroup control files (introduced here in 2012:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=abb8206cb0773)
use "KB" and not "kB"
(https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/mm/hugetlb_cgroup.c?h=v5.0#n349).
The behavior in the kernel has not changed since the introduction, and
the current code using "kB" will therefore fail on devices with huge
pages smaller than 1MiB. This is the case for AArch64.
As seen from the code in "mem_fmt" inside hugetlb_cgroup.c, only "KB",
"MB" and "GB" are used, so the others may be removed as well.
Here is a real world example of the files inside the
"/sys/kernel/mm/hugepages/" directory:
- "hugepages-64kB"
- "hugepages-2048kB"
- "hugepages-32768kB"
- "hugepages-1048576kB"
And the corresponding cgroup files:
- "hugetlb.64KB._____"
- "hugetlb.2MB._____"
- "hugetlb.32MB._____"
- "hugetlb.1GB._____"
Signed-off-by: Odin Ugedal <odin@ugedal.com>
Fix an issue in which, when trying to specify the `--kube-reserved-cgroup`
(or `--system-reserved-cgroup`) with `--cgroup-driver=systemd`, we will
not properly convert the `systemd` cgroup name into the internal cgroup
name that k8s expects. Without this change, specifying
`--kube-reserved-cgroup=/test.slice --cgroup-driver=systemd` will fail,
and only `--kube-reserved-cgroup=/test --crgroup-driver=systemd` will succeed,
even if the actual cgroup existing on the host is `/test.slice`.
Additionally, add light unit testing of our process from converting to a
systemd cgroup name to kubernetes internal cgroup name.
These updates are based on discussions had about the preferred semantics
of the TopologyManager and will be reflected in changes to an upcoming
PR that adds the actual TopologyManager implementation.
The cpumanager loops through all init Containers and app Containers when
reconciling its state. However, the current implementation of
findContainerIDByName(), which is call by the reconciler, does not
resolve for init Containers.
This patch updates findContainerIDByName() to account for init
Containers and adds a regression test that fails before the change and
succeeds after.
Minor optimization in the code that attempts to assign whole
sockets/cores in case the number of CPUs requested is higher
than CPUs-per-socket/core: check if the number of requested
CPUs is higher than CPUs-per-socket/core before retrieving
and iterating the free sockets/cores, and break the loops
when that is no longer the case.
Signed-off-by: Arik Hadas <ahadas@redhat.com>
1. Find the minimal thread number within a core using a
single loop rather than by sorting the thread numbers.
2. Inline getUniqueCoreID#err and Discover#numCPUs variables.
3. Narrow the scope of Discover#coreID and Discover#err variables.
Signed-off-by: Arik Hadas <ahadas@redhat.com>
that requests any device plugin resource. If not, re-issue Allocate
grpc calls. This allows us to handle the edge case that a pod got
assigned to a node even before it populates its extended resource
capacity.
Modify kubelet plugin watcher to support older CSI drivers that use an
the old plugins directory for socket registration.
Also modify CSI plugin registration to support multiple versions of CSI
registering with the same name.
- 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 67430, 67550). 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>.
cpumanager: rollback state if updateContainerCPUSet failed
**What this PR does / why we need it**:
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes#63018
If `updateContainerCPUSet` failed, the container will start failed. We should rollback the state to avoid CPU leak.
**Special notes for your reviewer**:
**Release note**:
```release-note
cpumanager: rollback state if updateContainerCPUSet failed
```
Automatic merge from submit-queue. 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>.
Fail container start if its requested device plugin resource is unknown.
With the change, Kubelet device manager now checks whether it has cached option state for the requested device plugin resource to make sure the resource is in ready state when we start the container.
**What this PR does / why we need it**:
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes https://github.com/kubernetes/kubernetes/issues/67107
**Special notes for your reviewer**:
**Release note**:
```release-note
Fail container start if its requested device plugin resource hasn't registered after Kubelet restart.
```
Automatic merge from submit-queue (batch tested with PRs 66512, 66946, 66083). 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>.
kubelet/cm/cpumanager: Fix unused variable "skipIfPermissionsError"
The variable "skipIfPermissionsError" is not needed even when
permission error happened.
Automatic merge from submit-queue. 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>.
clean up useless variables in deviceplugin/types.go
**What this PR does / why we need it**:
some variables is useless for reasons, I think we need a clean up.
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes #
**Special notes for your reviewer**:
**Release note**:
```release-note
```NONE
Automatic merge from submit-queue (batch tested with PRs 66190, 66871, 66617, 66293, 66891). 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>.
Do not set cgroup parent when --cgroups-per-qos is disabled
When --cgroups-per-qos=false (default is true), kubelet sets pod
container management to podContainerManagerNoop implementation and
GetPodContainerName() returns '/' as cgroup parent (default cgroup root).
(1) In case of 'systemd' cgroup driver, '/' is invalid parent as
docker daemon expects '.slice' suffix and throws this error:
'cgroup-parent for systemd cgroup should be a valid slice named as \"xxx.slice\"'
(5fc12449d8/daemon/daemon_unix.go (L618))
'/' corresponds to '-.slice' (root slice) in systemd but I don't think
we want to assign root slice instead of runtime specific default value.
In case of docker runtime, this will be 'system.slice'
(e2593239d9/daemon/oci_linux.go (L698))
(2) In case of 'cgroupfs' cgroup driver, '/' is valid parent but I don't
think we want to assign root instead of runtime specific default value.
In case of docker runtime, this will be '/docker'
(e2593239d9/daemon/oci_linux.go (L695))
Current fix will not set the cgroup parent when --cgroups-per-qos is disabled.
```release-note
Fix pod launch by kubelet when --cgroups-per-qos=false and --cgroup-driver="systemd"
```
Automatic merge from submit-queue (batch tested with PRs 66190, 66871, 66617, 66293, 66891). 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 nil pointer dereference in node_container_manager#enforceExisting
**What this PR does / why we need it**:
fix nil pointer dereference in node_container_manager#enforceExisting
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes#66189
**Special notes for your reviewer**:
NONE
**Release note**:
```release-note
kubelet: fix nil pointer dereference while enforce-node-allocatable flag is not config properly
```
Automatic merge from submit-queue. 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>.
Skip checking when failSwapOn=false
**What this PR does / why we need it**:
Skip checking when failSwapOn=false
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes #
**Special notes for your reviewer**:
NONE
**Release note**:
```
NONE
```
Automatic merge from submit-queue (batch tested with PRs 66623, 66718). 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>.
cpumanager: validate topology in static policy
**What this PR does / why we need it**:
This patch adds a check for the static policy state validation. The check fails if the CPU topology obtained from cadvisor doesn't match with the current topology in the state file.
If the CPU topology has changed in a node, cpumanager static policy might try to assign non-present cores to containers.
For example in my test case, static policy had the default CPU set of `0-1,4-7`. Then kubelet was shut down and CPU 7 was offlined. After restarting the kubelet, CPU manager tries to assign the non-existent CPU 7 to containers which don't have exclusive allocations assigned to them:
Error response from daemon: Requested CPUs are not available - requested 0-1,4-7, available: 0-6)
This breaks the exclusivity, since the CPUs from the shared pool don't get assigned to non-exclusive containers, meaning that they can execute on the exclusive CPUs.
**Release note**:
```release-note
Added CPU Manager state validation in case of changed CPU topology.
```
Test the cases where the number of CPUs available in the system is
smaller or larger than the number of CPUs known in the state, which
should lead to a panic. This covers both CPU onlining and offlining. The
case where the number of CPUs matches is already covered by the
"non-corrupted state" test.
This patch adds a check for the static policy state validation. The
check fails if the CPU topology obtained from cadvisor doesn't match
with the current topology in the state file.
If the CPU topology has changed in a node, cpu manager static policy
might try to assign non-present cores to containers.
For example in my test case, static policy had the default CPU set of
0-1,4-7. Then kubelet was shut down and CPU 7 was offlined. After
restarting the kubelet, CPU manager tries to assign the non-existent CPU
7 to containers which don't have exclusive allocations assigned to them:
Error response from daemon: Requested CPUs are not available - requested 0-1,4-7, available: 0-6)
This breaks the exclusivity, since the CPUs from the shared pool don't
get assigned to non-exclusive containers, meaning that they can execute
on the exclusive CPUs.
the caching layer on endpoint is redundant.
Here are the 3 related objects in picture:
devicemanager <-> endpoint <-> plugin
Plugin is the source of truth for devices
and device health status.
devicemanager maintain healthyDevices,
unhealthyDevices, allocatedDevices based on updates
from plugin.
So there is no point for endpoint caching devices,
this patch is removing this caching layer on endpoint,
Also removing the Manager.Devices() since i didn't
find any caller of this other than test, i am adding a
notification channel to facilitate testing,
If we need to get all devices from manager in future,
it just need to return healthyDevices + unhealthyDevices,
we don't have to call endpoint after all.
This patch makes code more readable, data model been simplified.
Automatic merge from submit-queue (batch tested with PRs 58755, 66414). 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>.
Use probe based plugin watcher mechanism in Device Manager
**What this PR does / why we need it**:
Uses this probe based utility in the device plugin manager.
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes#56944
**Notes For Reviewers**:
Changes are backward compatible and existing device plugins will continue to work. At the same time, any new plugins that has required support for probing model (Identity service implementation), will also work.
**Release note**
```release-note
Add support kubelet plugin watcher in device manager.
```
/sig node
/area hw-accelerators
/cc /cc @jiayingz @RenaudWasTaken @vishh @ScorpioCPH @sjenning @derekwaynecarr @jeremyeder @lichuqiang @tengqm @saad-ali @chakri-nelluri @ConnorDoyle
When --cgroups-per-qos=false (default is true), kubelet sets pod
container management to podContainerManagerNoop implementation and
GetPodContainerName() returns '/' as cgroup parent (default cgroup root).
(1) In case of 'systemd' cgroup driver, '/' is invalid parent as
docker daemon expects '.slice' suffix and throws this error:
'cgroup-parent for systemd cgroup should be a valid slice named as \"xxx.slice\"'
(5fc12449d8/daemon/daemon_unix.go (L618))
'/' corresponds to '-.slice' (root slice) in systemd but I don't think
we want to assign root slice instead of runtime specific default value.
In case of docker runtime, this will be 'system.slice'
(e2593239d9/daemon/oci_linux.go (L698))
(2) In case of 'cgroupfs' cgroup driver, '/' is valid parent but I don't
think we want to assign root instead of runtime specific default value.
In case of docker runtime, this will be '/docker'
(e2593239d9/daemon/oci_linux.go (L695))
Current fix will not set the cgroup parent when --cgroups-per-qos is disabled.
Automatic merge from submit-queue (batch tested with PRs 59214, 65330). 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>.
Migrate cpumanager to use checkpointing manager
**What this PR does / why we need it**:
This PR migrates `cpumanager` to use new kubelet level node checkpointing feature (#56040) to decrease code redundancy and improve consistency.
**Which issue(s) this PR fixes**:
Fixes#58339
**Notes**:
At point of submitting PR the most straightforward approach was used - `state_checkpoint` implementation of `State` interface was added. However, with checkpointing implementation there might be no point to keep `State` interface and just use single implementation with checkpoint backend and in case of different backend than filestore needed just supply `cpumanager` with custom `CheckpointManager` implementation.
/kind feature
/sig node
cc @flyingcougar @ConnorDoyle
Automatic merge from submit-queue (batch tested with PRs 64142, 64426, 62910, 63942, 64548). 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>.
Clean up fake mounters.
**What this PR does / why we need it**:
Fixes https://github.com/kubernetes/kubernetes/issues/61502
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes #
**Special notes for your reviewer**:
list of fake mounters:
- (keep) pkg/util/mount.FakeMounter
- (removed) pkg/kubelet/cm.fakeMountInterface:
- (inherit from mount.FakeMounter) pkg/util/mount.fakeMounter
- (inherit from mount.FakeMounter) pkg/util/removeall.fakeMounter
- (removed) pkg/volume/host_path.fakeFileTypeChecker
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue (batch tested with PRs 65230, 57355, 59174, 63698, 63659). 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>.
TODO has already been implemented
**What this PR does / why we need it**:
TODO has already been implemented, remove the TODO tag.
**Which issue(s) this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close the issue(s) when PR gets merged)*:
Fixes #
**Special notes for your reviewer**:
**Release note**:
```release-note
```NONE
Automatic merge from submit-queue (batch tested with PRs 63348, 63839, 63143, 64447, 64567). 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>.
Containerized subpath
**What this PR does / why we need it**:
Containerized kubelet needs a different implementation of `PrepareSafeSubpath` than kubelet running directly on the host.
On the host we safely open the subpath and then bind-mount `/proc/<pidof kubelet>/fd/<descriptor of opened subpath>`.
With kubelet running in a container, `/proc/xxx/fd/yy` on the host contains path that works only inside the container, i.e. `/rootfs/path/to/subpath` and thus any bind-mount on the host fails.
Solution:
- safely open the subpath and gets its device ID and inode number
- blindly bind-mount the subpath to `/var/lib/kubelet/pods/<uid>/volume-subpaths/<name of container>/<id of mount>`. This is potentially unsafe, because user can change the subpath source to a link to a bad place (say `/run/docker.sock`) just before the bind-mount.
- get device ID and inode number of the destination. Typical users can't modify this file, as it lies on /var/lib/kubelet on the host.
- compare these device IDs and inode numbers.
**Which issue(s) this PR fixes**
Fixes#61456
**Special notes for your reviewer**:
The PR contains some refactoring of `doBindSubPath` to extract the common code. New `doNsEnterBindSubPath` is added for the nsenter related parts.
**Release note**:
```release-note
NONE
```
Automatic merge from submit-queue. 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>.
Correct kill logic for pod processes
Correct the kill logic for processes in the pod's cgroup. os.FindProcess() does not check whether the process exists on POSIX systems.
Automatic merge from submit-queue (batch tested with PRs 60200, 63623, 63406). 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>.
Apply pod name and namespace labels for pod cgroup for cadvisor metrics
**What this PR does / why we need it**:
1. Enable Prometheus users to determine usage by pod name and namespace for pod cgroup sandbox.
1. Label cAdvisor metrics for pod cgroups by pod name and namespace.
1. Aligns with kubelet stats summary endpoint pod cpu and memory stats.
**Special notes for your reviewer**:
This provides parity with the summary API enhancements done here:
https://github.com/kubernetes/kubernetes/pull/55969
**Release note**:
```release-note
Apply pod name and namespace labels to pod cgroup in cAdvisor metrics
```
