Problem:

Monitoring get redeployed due to etcd params updated. However, it is not expected as the etcd address doesn't change at all.

Solution:

Sort it before assign.

Issue:

https://github.com/rancher/rancher/issues/19945

Sort etcd endpoints and controlplane endpoints.

If we can't obtain a token for launching the telemetry service,
do not retry forever, give up after 3 tries.

http.Transports will leak connections if they are garbage collected.  This
situation is made really bad when we are using a cluster dialer.

**Problem:**
Although delete monitoring action from schema resources, UI still can
seen `enableMonitoring` action link

**Solution:**
Remove actions addition from `Formatter`

**Issue:**
https://github.com/rancher/rancher/issues/19528

**Problem:**
Cannot see the workload/container/pod metrics as `ProjectMetricGraph` has been deleted.

**Solution:**
Cannot delete `ProjectMetricGraph` directly.

**Issue:**
https://github.com/rancher/rancher/issues/19516

Added content handler to gzip chain and
implemented Hijacker in content handler.
Prior, monitoring would break on certain
browsers because header was not being.

Properly chain gzip handler with v3 handler.

To handle an upgrade situation from v2.1 to v2.2, we added logic to set
the kubernetesVersion field on eks cluster config. We were overly cautious
in deciding when we would set it such that when EKS added suppport for k8s 1.12,
users could create a cluster with it in Rancher 2.1, but when they upgraded to
2.2, we would not set the kubernetsVersion appropriately.

Backport performance improvements

Added tests for logic in project setter store.

Now, identifies and supplies namespaces to namespace filter for requests
associated with given project. Prior, the backend listed items belonging
to all namespaces in all projects no matter what, and then filtered. This
led to longer request times for projects that contained a fraction of
total resources. Now, request times are improved and largely dependent on
resource amount for the given project.

Backport of d040b643 from master

Norman now tracks and assigns project namespaces
which are necessary to filter project requests.

Manual backport of commit f2b601cc
from master.

Norman now has a recognized option for opting out of pagination.
Prior, passing a limit of -1 would set the pagination limit to
the default, 1000. The backend would serve a paginated response
that the UI automatically aggregates. This led multiple api calls,
each of which taking as long to return as a single non-paginated
result. Now, one non-paginated response will be returned when
limit is -1.

Test script now runs go tests. Disabled existing test setup in e2e
package, as the the necessary dependencies for it appear to be
missing.

This reverts commit 3bb7a2778d186daaf61fc1ec8f28959ad8616dad.

Increased size of parent cache to 100k from 1k. Prior, when retrieving
a large number of resources (over 1k) items would shift out of the cache
before being called again. This caused misses on most requests. Now,
size of cache has increased 10x and misses have been largely reduced. Using
non-expiration lru cache, as the data is static and does not need to be
discarded after a set time.

**Problem:**
After enabling monitoring, the `ClusterConditionMonitoringEnabled` is no longer perceived.

**Solution:**
- Remove Cluster Monitoring detection
  + Correct judgement condition on Monitoring
  + Doesn't detect Cluster Monitoring component installation status
  + Remove useless component detections
- Detect cluster level Prometheus Endpoint is ready or not

**Issue:**
- https://github.com/rancher/rancher/issues/17733
- https://github.com/rancher/rancher/issues/19197

Problem:
single pod deploy by yaml don't have ownerRef, need add check before use
pod attributes

Solution:
check ownerRef array len

Issue:
https://github.com/rancher/rancher/issues/19358

Problem:
When users configure a custom git ca cert, it is not available for the publish catalog steps.

Solution:
Inject the custom ca cert to publish catalog containers.