The script for adding user to dcoker group would fail, since there
is no group with name 'docker' in current version of atomic image.
Use 'dockerroot' instead.
The default kubenetes kubelet config file specified a wrong option.
At the file /etc/kubernetes/kubelet, it specified:
"--api_server=127.0.0.1:8080"
First, the flag should be "--api_servers". Second, it should use
the IP address of the master node instead of "127.0.0.1".
See: https://bugzilla.redhat.com/show_bug.cgi?id=1200924
Currently, the list of minions is statically specified in master node,
and the list of IP addresses of minion nodes is passed to master node
through user_data. This approach doesn't work well on scaling. For example,
adding a minion node through stack-update will cause an update on user_data,
which cause the master node to be replaced.
This commit addresses this issue by dynamically register minion node.
There are two cases:
* Scale out: the new minion nodes will register themselves to the cluster
when they are booting.
* Scale in: a subset of minion nodes are deleted by Heat, and the
master node will discover it and automatically deregister the
deleted nodes.
the parameteters scetion specifies `type: string`, but YAML converts
bare numbers into integer types.
Change-Id: I3a430839eee7e4ad4e0dd3aa66720ed6ca0c49ec
The default replacement_policy is REPLACE_ALWAYS, which means a stack update
will always replace the port regardless of any property changes. As a result,
all the VMs of the kubecluster will be replaced, which is undesired.
it turns out that flannel's vxlan backend will crash centos 7, but works
fine under fedora 21. added a new 'flannel_use_vxlan' parameter to make
the backend selection conditional.
we now put selinux in permissive mode to avoid problems with fedora 21.
create a default user in the templates rather than relying on whatever
is used in the underlying images, which may be "centos" or "fedora"
or "ec2-user" or something else.
this splits the giant user-data script into smaller modules. this
makes things easier to manage and also permits us to use cloud-config
scripts in addition to shell scripts.
- fixed an order-of-operations issue that could have resulted in
conflicts between docker-io and legacy "docker" package
- customize a copy of downloaded image rather than the original (to make
it easier to start from scratch)
- make sure to clean up yum cache