Ceph

h1. 2F - Testing buildrelease & Teuthology

h3. Live Pad

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

h3. Summit Snapshot

Building and testing Ceph automatically

Build infrastructure (gitbuilders): 

p(. https://github.com/ceph/gitbuilder + https://github.com/ceph/autobuild-ceph  on-demand

    Ceph gitbuilder status: http://ceph.com/gitbuilder.cgi

Release builds

1. target platforms?

   ubuntu (precise, quantal, soon raring), squeeze, Centos, SuSE

2. process

Teuthology http://github.com/ceph/teuthology

p(. automated test cluster setup/run/collect output

   uses physical or virtual machines

   cooperative locking of machine resources to avoid trampling other users (optional)

   write YAML configuration files to select machines/roles, Ceph versions to install, tests to run

   a few hundred physical machines internal to Inktank to run nightly tests

gtest for unit tests: (make check)

p(. require tests for new code

   refactor existing code to be testable

Test suite  http://github.com/ceph/ceph-qa-suite

* combination/permutation of teuthology tests, cluster configurations

   many different functional/regression tests, with and without failure injection

upgrade testing

p(. run tests in mixed verison environment with slow rolling upgrades

integration tests:

p(.  openstack

    cloudstack

    chef.py

    libvirt (pools and volumes)

    qemu

Allocate/create VMs using EC2 or Openstack APIs?

Improve documentation on how to get it set up and working

- how to setup test machines (http://github.com/ceph/ceph-qa-chef )

Work items:

p(. build out a large cluster test suite

    parallel.py and sequential.py task

    rados.py radosmodel test should infer the list of clients and run them in parallel

    task to slurp up/archive perf counters

    identifying key metrics to monitor

p((.   osd: small/large write performance

       mds: metadata ops/sec

       ...

p(. qemu gitbuilder

    build large long-term clusters on burnupi?

    samba (and others?) don't register as running daemons and thus can't be restarted by the upgrade task

Performance:

p(. need to be able to identify performance regressions

    memory, cpu usage, network usage(please) data

p((.   perf task?

       collectl?

       store aggregated data in summary.yaml for each runl

p(. time (have raw timer task; need to log results)

    identify data warehouse and make something to import into it

    build chart.io graphs :)

    aggregate/slurp the osd perf counters at end of run?

scribe (facebook)

flume (cloudera)