By default check the KCM and scheduler on 127.0.0.1:<port> as that is the
defaall --bind-address kubeamd uses for these components.
For kube-apiserver take the value from APIEndpoint.AdvertiseAddress which is
dynamically detected from the host. Unless the user has passed explicitly --advertise-address
as an extra arg.
Read the <port> values for all components from the --secure-port flag
value if needed. Otherwise use defaults.
Use /livez for apiserver and scheduler. Add TODO for KCM to
switch to /livez as well.
[InPlacePodVerticalScaling] fix restore checkpoint bug: failed to verify pod status checkpoint checksum because of different behaviors of func Quantity.Marshal and Quantity.Unmarshal
For one thing, this release decouples device management from
libcontainer/cgroups. You can see the result of this in a dropped
cilium/ebpf dependency (which is only needed for device management).
NOTE that due to an issue with go mod / go list,
github.com/opencontainers/runc had to be added to
hack/unwanted-dependencies.json under x/exp. This is bogus because
opencontainers/runc does not use x/exp directly, only via cilium/ebpf
dependency (which is not vendored here).
Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>
chore(leader-election): modify leader-election comment to OnStoppedLeading callback is always called when the LeaderElector exits, even if it did not start leading.
If a client is configured to encode request bodies to CBOR, but the server does not support CBOR,
the server will respond with HTTP 415 (Unsupported Media Type). By feeding this response back to the
RESTClient, subsequent requests can fall back to JSON, which is assumed to be acceptable.
"Allocated devices" are the ones which can be observed from the informer. "All
allocated devices" also includes those which are in flight and haven't been
written back to the apiserver.
The logic for skipping "admin access" was repeated in three different places. A
single foreachAllocatedDevices with a callback puts it into one function.
The devices which are allocated before starting the allocation always remain
allocated. They can be stored once in a set, then each Filter call for the
different nodes can reuse that set instead of allocating it anew for each node.
goos: linux
goarch: amd64
pkg: k8s.io/kubernetes/test/integration/scheduler_perf
cpu: Intel(R) Core(TM) i9-7980XE CPU @ 2.60GHz
│ before │ after │
│ SchedulingThroughput/Average │ SchedulingThroughput/Average vs base │
PerfScheduling/SchedulingWithResourceClaimTemplateStructured/5000pods_500nodes-36 36.89 ± 2% 54.70 ± 6% +48.26% (p=0.002 n=6)
PerfScheduling/SteadyStateClusterResourceClaimTemplateStructured/empty_100nodes-36 105.7 ± 5% 106.4 ± 4% ~ (p=0.970 n=6)
PerfScheduling/SteadyStateClusterResourceClaimTemplateStructured/empty_500nodes-36 117.8 ± 3% 120.0 ± 4% ~ (p=0.134 n=6)
PerfScheduling/SteadyStateClusterResourceClaimTemplateStructured/half_100nodes-36 119.5 ± 4% 112.5 ± 4% -5.86% (p=0.009 n=6)
PerfScheduling/SteadyStateClusterResourceClaimTemplateStructured/half_500nodes-36 63.22 ± 2% 87.13 ± 4% +37.82% (p=0.002 n=6)
PerfScheduling/SteadyStateClusterResourceClaimTemplateStructured/full_100nodes-36 109.5 ± 2% 113.4 ± 2% +3.65% (p=0.006 n=6)
PerfScheduling/SteadyStateClusterResourceClaimTemplateStructured/full_500nodes-36 27.56 ± 5% 65.55 ± 3% +137.84% (p=0.002 n=6)
geomean 72.44 90.81 +25.37%
