[Gluster-users] managing lifecycle of a gluster volume in openshift

[Joseph Lorenzini]

Hi all,

I am happy to report that I finally got a container in an openshift pod to
mount a gluster volume successfully. This has nothing to do with gluster,
which works fine, and everything to do with openshift interfaces being less
than ideal. Note to self: turn off the settings in openshift that prevent
containers from running as root (what a silly restriction).

So now I need to tackle a much more complicated problem. How to handle the
lifecycle of a gluster volume in openshift....

Here are the things I am considering and I'd be interested to see how
others have addressed this problem. Lets assume for the purposes of this
conversation we have a single gluster cluster that will use a replicated
volumes with a replica count of 3 (nothing is distributed) and the cluster
consists of three nodes. So each volume minimally has three bricks, where
each server only has one of the bricks. Total available disk throughput
gluster is 1 terabyte.


   - do you use a single gluster volume for multiple pods or one gluster
   volume for each pod Until gluster supports mounting a subdirectory in a
   volume natively (can't wait for that feature!!), it seems like you'd want
   to go the route of volume per pod for reasons of multi-tenancy and security.
   -  if you do a gluster volume per pod, how do you handle the physical
   storage that backs the gluster cluster? For example, lets say each gluster
   server has three devices (/dev/sdb,/dev/sdc,/dev/sdd) that can be used by
   bricks. Would it be a good idea to create a volume for each openshift pod,
   where there are multiple brick processes writing to the same device on the
   same disk for different volumes? Or would that have unacceptable
   performance implications? The reason I ask is that the gluster docs seems
   to recommend having physical devices dedicated to a single volume only.


I did take a look at heketi but have a variety of concerns/questions on
that, which are probably more appropriate for whatever email list discusses
heketi.

https://github.com/screeley44/openshift-docs/blob/
ce684e3c4c581db3b4aa27ecc1dba2ea65f51eda/install_config/
storage_examples/external_gluster_dynamic_example.adoc

Thanks,
Joe
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.gluster.org/pipermail/gluster-users/attachments/20170226/13cecdb0/attachment.html>