so I found a bit of time to look again to the testbed of openshift
in the cloud (https://fedoraproject.org/wiki/Cloud/OpenShift_Origin_Testbed
), and now
is 2 atomic VM running, I realize that we didn't define exactly the details.
Since one of the goal is to use fedora atomic, the vm run fedora atomic. So far, so good.
On what is needed to deploy, we have:
- the master (so openshift itself)
- a docker registry (may requires some storage)
- a router (so haproxy managed by openshift)
- and various nodes to run conatiners (so kubernetes and/or etcd and docker)
Atomic come with kubernetes, etcd and docker pre installed. AFAIK, there isn't yet
So the first question is about the layout.
When the first idea were discussed, people did propose
to have 2 nodes for the master, 2 node for infra and all others as worker nodes, and
add one storage node. That's likely a fair production setup, but we are not yet doing
And discussing around in the office, my co-workers suggest that we merge the master and
the router on
one single node. I would also suggest to skip HA for now, and keep for later. So the
proposed layout would be:
- 1 node, with a public IP, running the master, the router and the registry
- a few nodes behind running container
Then once that's working, start to make regitry be replicated (not too hard to do on
since it can use the block storage), make the router replicated and the master/openshift
Then come the question that currently kinda block me, about where does the binaries
need to come from.
As people might know, all stuff on atomic is either preinstalled,
or come from a docker container.
Openshift can bundle kubernetes and etcd, or can likely use external ones.
So we have to decide:
- does etcd come from atomic (ie preinstalled), or do we use the one bundled in openshift
- do we use a external registry (using the rpm), or a external registry using a container,
or using the
one deployed by openshift ?
- do we want to use openshift coming from F24 in a container, or directly consume the one
openshift upstream on docker hub ?
I am not sure what is the answer for all of this, because in the end, the policy of fedora
infra are maybe
not suitable to decide for that, and I am not sure if the tooling is ready. For example,
if we take openshift from
the rpm in a container, someone need to build that container.
Any toughts ?
(I will likely decide by fiat anyway, but without the illusion of having got feedback if
people do not give it)