ceph/doc/dev/sepia.rst
Sage Weil b61468deb7 doc/dev/sepia: notes about sepia email list and irc channels
Signed-off-by: Sage Weil <sage@redhat.com>
2014-10-14 14:43:04 -07:00

88 lines
2.3 KiB
ReStructuredText

Notes on the Sepia community test lab
=====================================
The Ceph community maintains a test lab that is open to active
contributors to the Ceph project.
The lab is currently located in DreamHost's Irvine, CA data center. There are
about 8 racks of gear.
Hardware loans or donations are gladly accepted and will be put to
good use running regression and other automated testing.
E-mail and IRC
--------------
Acitivity within the lab is coordinated in two places:
* `sepia@ceph.com`_ email discussion list.
* #sepia on irc.oftc.net
.. _sepia@ceph.com: http://lists.ceph.com/listinfo.cgi/ceph-qa-ceph.com/
Hardware overview
-----------------
* **96 plana**: 1u Dell R410. 8 core Intel E5620 2.4GHz. 4x 500GB SATA. 1gig and 10gig network.
* **64 burnupi**: 2u Dell R515. 6 core. 8 disk. 1gig and 10gig network. Crummy SAS expanders.
* **120 mira**: 8 core. 4-8 1TB disks. 1gig network.
* 8 vercoi: 24 core, RAID; VM hosts
* 4 senta: 24 core: VM hosts
* 36 saya: Calxeda armv7l nodes (for testing)
* 24 tala: Calxeda armv7l nodes (gitbuilder)
* ~200 vps: VMs running on mira hardware
* 4 rex
* 4 rhoda
* force 10 switches
Access
------
We use openvpn to grant access to the lab network. Public ssh keys are used to
grant access to individual machines.
Locking machines
----------------
* All tests pull their builds from gitbuilder.ceph.com.
* Anybody can lock machines with ``teuthology-lock --lock-many NUM
--machine-type TYPE``.
* Machines are locked as ``whoami''@``hostname -s``. --owner to
choose otherwise.
* Automated tests current run on the ``plana``; please avoid locking
these for personal use.
* To unlock, please use ``teuthology-nuke -t list.yaml -r -u``, which
will reboot and clean up any leftover test state before unlocking
(or fail to unlock). It looks for a ``targets::`` section in the
yaml, so the regular job yaml will work. You can get a list of all
locked machines with ``teuthology-lock --list-targets``.
* ``teuthology-lock -a --brief`` or ``teuthology-lock --summary`` to
see what is locked and by whom.
* Be conscientious about scheduling entire qa runs. Coordinate
utilization on IRC. Make sure you are running the latest version
ceph-qa-suite.git and teuthology.git.
* Results for scheduled runs appear in /a/$jobname on the teuthology
machine.