Automatic merge from submit-queue (batch tested with PRs 39260, 40216, 40213, 40325, 40333)
Fixed propagation of kube master certs during master replication.
Fixed propagation of kube-master-certs during master replication.
Automatic merge from submit-queue
Enable lazy initialization of ext3/ext4 filesystems
**What this PR does / why we need it**: It enables lazy inode table and journal initialization in ext3 and ext4.
**Which issue this PR fixes** *(optional, in `fixes #<issue number>(, fixes #<issue_number>, ...)` format, will close that issue when PR gets merged)*: fixes#30752, fixes#30240
**Release note**:
```release-note
Enable lazy inode table and journal initialization for ext3 and ext4
```
**Special notes for your reviewer**:
This PR removes the extended options to mkfs.ext3/mkfs.ext4, so that the defaults (enabled) for lazy initialization are used.
These extended options come from a script that was historically located at */usr/share/google/safe_format_and_mount* and later ported to GO so this dependency to the script could be removed. After some search, I found the original script here: https://github.com/GoogleCloudPlatform/compute-image-packages/blob/legacy/google-startup-scripts/usr/share/google/safe_format_and_mount
Checking the history of this script, I found the commit [Disable lazy init of inode table and journal.](4d7346f7f5). This one introduces the extended flags with this description:
```
Now that discard with guaranteed zeroing is supported by PD,
initializing them is really fast and prevents perf from being affected
when the filesystem is first mounted.
```
The problem is, that this is not true for all cloud providers and all disk types, e.g. Azure and AWS. I only tested with magnetic disks on Azure and AWS, so maybe it's different for SSDs on these cloud providers. The result is that this performance optimization dramatically increases the time needed to format a disk in such cases.
When mkfs.ext4 is told to not lazily initialize the inode tables and the check for guaranteed zeroing on discard fails, it falls back to a very naive implementation that simply loops and writes zeroed buffers to the disk. Performance on this highly depends on free memory and also uses up all this free memory for write caching, reducing performance of everything else in the system.
As of https://github.com/kubernetes/kubernetes/issues/30752, there is also something inside kubelet that somehow degrades performance of all this. It's however not exactly known what it is but I'd assume it has something to do with cgroups throttling IO or memory.
I checked the kernel code for lazy inode table initialization. The nice thing is, that the kernel also does the guaranteed zeroing on discard check. If it is guaranteed, the kernel uses discard for the lazy initialization, which should finish in a just few seconds. If it is not guaranteed, it falls back to using *bio*s, which does not require the use of the write cache. The result is, that free memory is not required and not touched, thus performance is maxed and the system does not suffer.
As the original reason for disabling lazy init was a performance optimization and the kernel already does this optimization by default (and in a much better way), I'd suggest to completely remove these flags and rely on the kernel to do it in the best way.
Automatic merge from submit-queue (batch tested with PRs 39694, 39383, 39651, 39691, 39497)
Bump container-linux and gci timeout for docker health check
The command `docker ps` can take longer time to respond under heavy load or
when encountering some known issues. In these cases, the containers are running
fine, so aggressive health check could cause serious disruption. Bump the
timeout to 60s to be consistent with the debian-based containerVM.
This addresses #38588
The command `docker ps` can take longer time to respond under heavy load or
when encountering some known issues. In these cases, the containers are running
fine, so aggressive health check could cause serious disruption. Bump the
timeout to 60s to be consistent with the debian-based containerVM.
Automatic merge from submit-queue
Remove all MAINTAINER statements in the codebase as they are deprecated
**What this PR does / why we need it**:
ref: https://github.com/docker/docker/pull/25466
**Release note**:
```release-note
Remove all MAINTAINER statements in Dockerfiles in the codebase as they are deprecated by docker
```
@ixdy @thockin (who else should be notified?)
Automatic merge from submit-queue (batch tested with PRs 38294, 37009, 36778, 38130, 37835)
Only configure basic auth on gci if KUBE_USER and KUBE_PASSWORD are specified.
This should not change the existing flow when KUBE_USER/KUBE_PASSWORD are specified.
It makes not specifying those a valid option that means "don't turn on basic auth".
I only did it for cluster/gce/gci for now, but others should be somewhat similar.
Automatic merge from submit-queue
add a configuration for kubelet to register as a node with taints
and deprecate --register-schedulable
ref #28687#29178
cc @dchen1107 @davidopp @roberthbailey
Automatic merge from submit-queue (batch tested with PRs 37692, 37785, 37647, 37941, 37856)
Use unified gcp fluentd image for gci and cvm
Follow-up of https://github.com/kubernetes/kubernetes/pull/37681
Actually unify the pod specs for CVM and GCI, to simplify the configuration
CC @piosz
Automatic merge from submit-queue
Deploy a default StorageClass instance on AWS and GCE
This needs a newer kubectl in kube-addons-manager container. It's quite tricky to test as I cannot push new container image to gcr.io and I must copy the newer container manually.
cc @kubernetes/sig-storage
**Release note**:
```release-note
Kubernetes now installs a default StorageClass object when deployed on AWS, GCE and
OpenStack with kube-up.sh scripts. This StorageClass will automatically provision
a PeristentVolume in corresponding cloud for a PersistentVolumeClaim that cannot be
satisfied by any existing matching PersistentVolume in Kubernetes.
To override this default provisioning, administrators must manually delete this default StorageClass.
```
Automatic merge from submit-queue
Modify GCI mounter to enable NFSv3
In order to make NFSv3 work, mounter needs to start rpcbind daemon. This
change modify mounter's Dockerfile and mounter script to start the
rpcbind daemon if it is not running on the host.
After this change, need to make push the image and update the sha number in Changelog.
In order to make NFSv3 work, mounter needs to start rpcbind daemon. This
change modify mounter's Dockerfile and mounter script to start the
rpcbind daemon if it is not running on the host.
After this change, need to make push the image and update the sha number in Changelog.
Automatic merge from submit-queue
Change master to advertise external IP in kubernetes service.
Change master to advertise external IP in kubernetes service.
In effect, in HA mode in case of multiple masters, IP of external load
balancer will be advertise in kubernetes service.
Automatic merge from submit-queue
Migrates addons from RCs to Deployments
Fixes#33698.
Below addons are being migrated:
- kube-dns
- GLBC default backend
- Dashboard UI
- Kibana
For the new deployments, the version suffixes are removed from their names. Version related labels are also removed because they are confusing and not needed any more with regard to how Deployment and the new Addon Manager works.
The `replica` field in `kube-dns` Deployment manifest is removed for the incoming DNS horizontal autoscaling feature #33239.
The `replica` field in `Dashboard` Deployment manifest is also removed because the rescheduler e2e test is manually scaling it.
Some resource limit related fields in `heapster-controller.yaml` are removed, as they will be set up by the `addon resizer` containers. Detailed reasons in #34513.
Three e2e tests are modified:
- `rescheduler.go`: Changed to resize Dashboard UI Deployment instead of ReplicationController.
- `addon_update.go`: Some namespace related changes in order to make it compatible with the new Addon Manager.
- `dns_autoscaling.go`: Changed to examine kube-dns Deployment instead of ReplicationController.
Both of above two tests passed on my own cluster. The upgrade process --- from old Addons with RCs to new Addons with Deployments --- was also tested and worked as expected.
The last commit upgrades Addon Manager to v6.0. It is still a work in process and currently waiting for #35220 to be finished. (The Addon Manager image in used comes from a non-official registry but it mostly works except some corner cases.)
@piosz @gmarek could you please review the heapster part and the rescheduler test?
@mikedanese @thockin
cc @kubernetes/sig-cluster-lifecycle
---
Notes:
- Kube-dns manifest still uses *-rc.yaml for the new Deployment. The stale file names are preserved here for receiving faster review. May send out PR to re-organize kube-dns's file names after this.
- Heapster Deployment's name remains in the old fashion(with `-v1.2.0` suffix) for avoiding describe this upgrade transition explicitly. In this way we don't need to attach fake apply labels to the old Deployments.
Kubernetes Submit Queue