This metrics is measured in seconds so it makes no sense starting from
1000 as init value. This breaks also the scheduler e2e metric thus make
users unable to compute, for example, their SLO for the scheduler.
Even if this metric is deprecated, it should behave correctly until it is
completely removed to avoid user confusion.
For example, for each volume created, the minimum value exposed
as a metric is 16.6min (1000sec/60) which is obviously wrong as logic.
In this commit, we migrate bucket creation to start from reasonable
numbers, copying the incrementation from the conventions that the
scheduler follows itself.
Signed-off-by: dntosas <ntosas@gmail.com>
This updates the StaleSlices() method in EndpointSliceTracker to also
ensure that the tracker does not have more slices than have been
provided.
Co-Authored-By: Swetha Repakula <srepakula@google.com>
Add support to the endpoint slice mirroring controller to mirror
annotations, in addition to labels, but don´t mirror endpoint
triggertime annotation.
Also, fix a bug in the endpointslice mirroring controller, that
wasn't updating the mirrored slice with the new labels, in case
that only the endpoint labels were modified.
This updates the EndpointSlice controller to make use of the
EndpointSlice tracker to identify when expected changes are not present
in the cache yet. If this is detected, the controller will wait to sync
until all expected updates have been received. This should help avoid
race conditions that would result in duplicate EndpointSlices or failed
attempts to update stale EndpointSlices. To simplify this logic, this
also moves the EndpointSlice tracker from relying on resource versions
to generations.
Pod with PVC will not be scheduled if the PVC is being deleted.
This can happen when the PVC has finalizers of storage plugins.
Such a pod becomes pending. Unfortunately, after the finalizer
finishes and PVC is deleted, the pod remains pending forever.
The StatefulSet controller does nothing for this pending pod.
This commit prevents the StatefulSet controller from creating
such pods when PVC is to be deleted.
The range allocator in pkg/controller/nodeipam/ipam/range_allocator.go
may call Occupy() on the same range twice:
1. Just before subscribing to the NodeInformer
2. From a callback given to the NodeInformer soon after registration
When a pod is deleted, it is given a deletion timestamp. However the
pod might still run for some time during graceful shutdown. During
this time it might still produce CPU utilization metrics and be in a
Running phase.
Currently the HPA replica calculator attempts to ignore deleted pods
by skipping over them. However by not adding them to the ignoredPods
set, their metrics are not removed from the average utilization
calculation. This allows pods in the process of shutting down to drag
down the recommmended number of replicas by producing near 0%
utilization metrics.
In fact the ignoredPods set is misnomer. Those pods are not fully
ignored. When the replica calculator recommends to scale up, 0%
utilization metrics are filled in for those pods to limit the scale
up. This prevents overscaling when pods take some time to startup. In
fact, there should be 4 sets considered (readyPods, unreadyPods,
missingPods, ignoredPods) not just 3.
This change renames ignoredPods as unreadyPods and leaves the scaleup
limiting semantics. Another set (actually) ignoredPods is added to
which delete pods are added instead of being skipped during
grouping. Both ignoredPods and unreadyPods have their metrics removed
from consideration. But only unreadyPods have 0% utilization metrics
filled in upon scaleup.
This fixes a bug that occurred when a Service was rapidly recreated.
This relied on an unfortunate series of events:
1. When the Service is deleted, the EndpointSlice controller removes it
from the EndpointSliceTracker along with any associated EndpointSlices.
2. When the Service is recreated, the EndpointSlice controller sees that
there are still appropriate EndpointSlices for the Service and does
nothing. (They have not yet been garbage collected).
3. When the EndpointSlice is deleted, the EndpointSlice controller
checks with the EndpointSliceTracker to see if it thinks we should have
this EndpointSlice. This check was intended to ensure we wouldn't
requeue a Service every time we delete an EndpointSlice for it.
This adds a check in reconciler to ensure that EndpointSlices it is
working with are owned by a Service with a matching UID. If not, it will
mark those EndpointSlices for deletion (assuming they're about to be
garbage collected anyway) and create new EndpointSlices.
Previously the controllers would proceed with additional creates,
updates, or deletes if 1 failed. That could potentially result in
scenarios where an EndpointSlice create or update failing while a delete
worked. This updates the logic so that removals will not happen if
additions fail.
The KEP specifies that the controller will "mirror all labels from the
Endpoints resource and all endpoints and ports from the corresponding subset".
I'd missed that in my initial implementation, this should fix that.
