Jessie image can be selected by name using AWS_IMAGE_NAME. This is a
little easier than specifying the AMI ID.

Given the problems with docker's installation, we should be baking it
into the image.

This means we don't have to code the AMI ids for each region.

It isn't available (and hopefully not needed) on Debian

Don't rely on events in deployment rollback e2e test

Remove potential corrupted docker network checkpoint file before star…

again.

Remove racy check in waitForDeploymentStatus; wait for underlying resources to reach the desired state instead

Auto commit by PR queue bot

Auto commit by PR queue bot

Auto commit by PR queue bot

Plumb node labels through salt

Expose node-labels flag on kubelet for gce via a startup script
env var.

Teach the kubelet about promiscuous mode, take 2

Add image version to glbc name.

Remove racy check in waitForDeploymentStatus; wait for underlying resources to reach the desired state instead

Add milestone tag clarifications

Auto commit by PR queue bot

Make sure docker is properly shutdown before startup.

Auto commit by PR queue bot

Auto commit by PR queue bot

Auto commit by PR queue bot

Updated label selector for daemon set

Auto commit by PR queue bot

Mark flaky Deployment e2e tests until the problems are resolved.

Based on the official debian image, with the following changes:

* Switched extlinux -> grub, because we need to change kernel options
to enable the memory cgroup controller, and extlinux is harder and has
reboot problems
* Added packages that would otherwise be installed as part of the boot
(just an optimization)
* Also add the cloud-initramfs-growroot package; with it the root
volume will resize.
* We add panic=10 & oops=panic to kernel options
* We install the packages as per the base image, except we install
  awscli from pip, because the repo version is really old.

Auto commit by PR queue bot

Ref #21463, #21467, #21491, #19299

Auto commit by PR queue bot

Once we've built the master, we can build kubeconfig.  By doing so, if
we time out waiting for the nodes, the system is still configured
correctly.

In particular, spot instances can be slow to launch.

Related to issue #21200

Auto commit by PR queue bot

Auto commit by PR queue bot

Helps debug problem when running tests on OS X.

Auto commit by PR queue bot

Revert "kubelet: fix duplicated status updates at pod cleanup"