Ceph

h1. 2E - Chef Cookbook Consolidation & ceph-deploy Improvements

h3. Live Pad

The live pad can be found here: "[pad]":http://pad.ceph.com/p/chef-ceph-deploy

h3. Summit Snapshot

Welcome to the Ceph Etherpad!

Chef: ( http://github.com/ceph/ceph-cookbooks )

Past few months: refactored recipes, included Newdream input

Soliciting input for recipe features or anything else!

Current features:

p(. Deploy mon/osd

    Supports latest Bobtail and Cuttlefish only

Possible extensions:

p(. Does anyone need other release support?

    Limiting power of deployment keys (monitor key opens everything)

    Error management/recovery

Other issues:

p(. Consolidate best practices from many Chef cookbook authors

    Compatibility with ceph-deploy approach would be nice, at least from ceph-deploy -> Chef

* -> Documentation on migration path.

Look at https://github.com/wonko/ceph-cookbook to see if we can take some features.

Questions:

p(. Q: How will the Chef scripts be synced with the Barclamp Chef recipes?

    A: Dell/SuSE are working on it most; ideally they would simply include upstream recipes

* Q: LWRP for creating pools? clients? manage keys? 

* A: the intent is to deploy, but not administer, so focused on setup for now

* Lightweigth resource Resources and Providers - kind of abstraction for objects/resources in chef - http://docs.opscode.com/lwrp.html

* Presentation on LWRP from chefconf (thanks kbader):

* https://docs.google.com/presentation/d/1gQj_QNbzIecZQtbb4oVNMZKlPqiO1q6d08g5l8BqWOI/edit#slide=id.p

p(. Q: when someone want to put effort in the cookbooks, what sections would you want to be tackled first? (cleanup, more distributions, hardening the runs, ...)

   A: first priority is better testing on Centos/Redhat/SuSE/Fedora; error handling also needs help

p(. Q: are there any tests in the cookbooks? 

   A: no, and we'd like to have some

   What kind?  Ideally we'd use something like test-kitchen with a handful of minitest handlers

   Not sure what the convention is for test frameworks for Chef, but "as nonsurprising as possible" always seems right to me

p(. Q: previously there was some goal of running a setup adding additional disks (osds) through some labels on the disks - and then auto-add them to the cluster - is this still the way to go?

  A: yes, ceph-create-keys/ceph-disk (in the ceph repo itself, http://github.com/ceph/ceph ) and the upstart/sysvinit scripts support auto-configuring disks today, and ceph-deploy will mark disks for that sort of delayed activation

  A2: We have some newer chef cookbooks that dreamhost worked with inktank on and they use the ceph-disk-prepare, ceph-disk-activate tooling, we'll open that up and share it

ceph-deploy:

* Syncing efforts between ceph-deploy and Chef is a requirement; publicizing changes/conventions for other environments (Juju, Salt, Puppet) will probably be done in the mailing list

* can it have it own mailing list please