#333 removing gluster/ceph rpms from base atomic host
Closed: Fixed 3 years ago Opened 3 years ago by dustymabe.

These were removed as part of the remove of kube. They were originally added here as a result of this BZ

We need to determine if tools like openshift-ansible depend on them being in the host for gluster storage to work or if it all works inside containers for them.

A longer term approach would be to get package layering support in an ansible module so that we could worry about stuff like this less.

Metadata Update from @dustymabe:
- Issue tagged with: F27, host

3 years ago

The mount.glusterfs command (provided in glusterfs-fuse RPM) is required to be able to make use of GlusterFS volumes in kube/openshift. Without it the nodes won't be able to provide those volumes to their containers.

@jarrpa - i'll ask the naive question before someone else does in this thread; is there is no way to do this by running mount.glusterfs command from inside a container?

The answer can be "yes, but with pain". We just want to gauge the level of pain :)

@dustymabe Honestly, I wouldn't know. :) Never tried! I don't know enough about Atomic to truly understand how that would work, unfortunately. I'm also a bit unfamiliar with the glusterfs-fuse packaging, so i don't know if the RPM has other packaging or configuration dependencies.

@jarrpa - could we bring in someone who knows better? maybe we can chat on IRC tomorrow and help figure some things out.

@dustymabe I'm out in CA for a conference through tomorrow. I'm still in CA on Friday but would be more open to an IRC or audio/video meeting.

@dustymabe I'm out in CA for a conference through tomorrow. I'm still in CA on Friday but would be more open to an IRC or audio/video meeting.

@jarrpa ok - let's get something together and chat with some people who know more about glusterfs-fuse and openshift-ansible.

I just looked at this on f27 w/ layering. Here's what's added:


note: to get this information I package layered dnf then ostree admin unlock --hotfix then dnf install blah.

here's some more detailed information:

to add glusterfs-fuse:

[root@vanilla-f27atomic ~]# dnf install glusterfs-fuse | tee
Fedora 27 - x86_64                              9.9 MB/s |  66 MB     00:06    
Last metadata expiration check: 0:00:06 ago on Mon 18 Sep 2017 06:24:59 PM UTC.
Dependencies resolved.
 Package                    Arch     Version            Repository         Size
 glusterfs-fuse             x86_64   3.12.1-1.fc27      updates-testing   149 k
Installing dependencies:
 glusterfs                  x86_64   3.12.1-1.fc27      updates-testing   589 k
 glusterfs-client-xlators   x86_64   3.12.1-1.fc27      updates-testing   875 k
 glusterfs-libs             x86_64   3.12.1-1.fc27      updates-testing   406 k

Transaction Summary
Install  4 Packages

Total download size: 2.0 M
Installed size: 7.5 M

for ceph it's a little more:

[root@vanilla-f27atomic ~]# dnf install ceph-common | tee                                                                                                                                    
'Last metadata expiration check: 0:00:42 ago on Mon 18 Sep 2017 06:24:59 PM UTC.
Dependencies resolved.
 Package               Arch      Version               Repository          Size
 ceph-common           x86_64    1:12.2.0-1.fc27       updates-testing     15 M
Installing dependencies:
 gperftools-libs       x86_64    2.6.1-3.fc27          fedora             287 k
 leveldb               x86_64    1.18-1.fc26           fedora             148 k
 libbabeltrace         x86_64    1.5.3-1.fc27          fedora             195 k
 libcephfs2            x86_64    1:12.2.0-1.fc27       updates-testing    451 k
 librados2             x86_64    1:12.2.0-1.fc27       updates-testing    2.9 M
 libradosstriper1      x86_64    1:12.2.0-1.fc27       updates-testing    346 k
 librbd1               x86_64    1:12.2.0-1.fc27       updates-testing    1.1 M
 librgw2               x86_64    1:12.2.0-1.fc27       updates-testing    1.8 M
 lttng-ust             x86_64    2.10.0-2.fc27         fedora             260 k
 python-cephfs         x86_64    1:12.2.0-1.fc27       updates-testing    125 k
 python-prettytable    noarch    0.7.2-11.fc27         fedora              41 k
 python-rados          x86_64    1:12.2.0-1.fc27       updates-testing    309 k
 python-rbd            x86_64    1:12.2.0-1.fc27       updates-testing    174 k
 python-rgw            x86_64    1:12.2.0-1.fc27       updates-testing    119 k
 snappy                x86_64    1.1.4-5.fc27          fedora              61 k
 userspace-rcu         x86_64    0.10.0-3.fc27         fedora              98 k

Transaction Summary
Install  17 Packages

Total download size: 24 M
Installed size: 77 M
Is this ok [y/N]: N

so gluster is 7M installed and ceph is 77M.

talked in the atomic working group meeting today. the current proposal is that we keep gluster in since it is rather small, and leave ceph out.

Please voice objections/concerns here, otherwise we'll take action soonish.

It looks like while support may have come in docker, kubernetes still doesn't support the sidecar style container that would allow us to pull these from AH. There is a proposal from the OpenShift team to add that support to k8s here:


The card is old, and the various links are some what historical, but it looks like the team is still working on it.

My thoughts are things still need to be shipped in the host until that's fixed.

in that case we should add them both back. Any objections?

+1 on adding them both back until they can be side-loaded.

What about package layering?

I'm fine w/ leaving them in, though.

What about package layering?

I'd prefer to have ansible support for package layering and livefs support to not be experimental before relying on layering :(

I'm fine w/ leaving them in, though.


The gluser/ceph rpms are now back in f27 and rawhide. Closing this issue.

Metadata Update from @dustymabe:
- Issue close_status updated to: Fixed
- Issue status updated to: Closed (was: Open)

3 years ago

Login to comment on this ticket.