Problem:
Sometimes get etcd unhealth alert, but may cause by network or disk problem, actually etcd is health

Solution:
Improve group wait seconds to 10 mins as issue mentioned

Issue:
https://github.com/rancher/rancher/issues/19474

Add logic to stream

Update websockets

Adding a worker node to an RKE cluster does not lead to rke up.
But the provisioner controller still calls the update, and sets the
taints capability after it. This commit checks if the update actually let
to rke up and sets taints capability based on that

k8s 1.16 has moved deployments to apps/v1 apigroup
DeploymentRollback and RollbackTo are deprecated in apps/v1
This commit handles deployment rollback separately for apps/v1,
by updating podSpec of a deployment with the podSpec of the replicaSet
This is the recommended approach from k8s client-go issues
https://github.com/kubernetes/client-go/issues/398#issuecomment-382546959

Problem:
After monitoring charts upgraded, this expression value change from no data point to 0

Solution:
Update the thredshold for this expression

Issue:
https://github.com/rancher/rancher/issues/22680

Bump telemetry to version 0.5.7

Remove monitoring and logging catalog id setting

Cleanup legacy globalrolebindings finalizers to deleted clusters

Problem:
Alert depends on the monitoring catalog setting when deployment, it need update setting when release, this isn't convenient and easy cause problem

Solution:
Remove monitoring catalog from setting and remove legacy code

Issue:
https://github.com/rancher/rancher/issues/22615

Problem: service-sidekick was being every plan execution, docker containers were started while already running and because of that, log-linker was run for every container that was started while it was already running
Root cause: service-sidekick was not excluded from plan execution (should be similar to share-mnt which is excluded), and there is no check for containers already running
Solution: exclude service-sidekick from being checked besides being present, check container is running before starting + running log linker, exclude service-sidekick from running log-linker

So as to make resolved notifications recognizable.

Problem:
When multiple recipients are configured in pipeline notification, one of
them get notified multiple times.

Solution:
Avoid using loop iterator variable in go routines.

**Problem:**
Setting node pool taints and node template taints doesn't work.

**Solution:**
Add kubelet args `--register-with-taints` parameters to kubelet worker
process. So those taints will be added after kubelet is up and running.
If `--register-with-taints` args is set by kubelet service extra_args,
taints from node template, node pool and kubelet will be merged.

The global ns rbac controller calls DeepEqual on the entire Role and
RoleBinding objects, this can cause continuous updates since some fields
such as CreationTimestamp have to be different. This commit calls DeepEqual
only on Subjects of RoleBinding, and on Rules in case of Roles.

Previously, this only removed finalizers from deleted grbs with a 1 hour delay.
This meant that when a user had their admin access removed it would take one hour for
the change to propogate if they had a finalizer to a non-existent cluster in their list of finalizers.

Now, we premptively remove "bad" finalziers from grbs that point to non-existent clusters so that on
change they will not need to be cleaned. This only occurs once and all grbs created after this fix will
have the "field.cattle.io/grbUpgrade" annotation to avoid running through this sync handler.

The global ns rbac controller calls DeepEqual on the entire Role and
RoleBinding objects, this can cause continuous updates since some fields
such as CreationTimestamp have to be different. This commit calls DeepEqual
only on Subjects of RoleBinding, and on Rules in case of Roles.

- sync only if interval!=0 and hash changes
- git clone instead of using raw github content
- updated log msgs

**Problem:**
HPA with cpu rules for newly created workload will get
`transitioning=error` because the HPA controller can't get the cpu
metrics for newly created workload.

**Solutions:**
Ignore the transitioning error for the first 60s after HPA has been
created.