2014-04-09 00:26:37 +00:00
|
|
|
- to get the reporting stuff working, you'll need to have gnuplot
|
|
|
|
|
installed on your local machine, at at least version 4.6.3. If you
|
|
|
|
|
need to install from source, you'll need at least these two deps:
|
2013-12-30 18:01:25 +00:00
|
|
|
|
2014-04-09 00:26:37 +00:00
|
|
|
- sudo apt-get install libgd2-xpm-dev libreadline6-dev
|
2013-12-30 18:01:25 +00:00
|
|
|
|
2014-04-09 00:26:37 +00:00
|
|
|
- symlink the stuff in perf/reporting to ~/bin/, assuming that is on
|
|
|
|
|
your path.
|
2013-12-30 18:01:25 +00:00
|
|
|
|
2014-04-09 00:26:37 +00:00
|
|
|
- drop_caches.sh needs to be visible on the remote machine. the
|
|
|
|
|
initial version of these scripts assumes that everyone is on the
|
|
|
|
|
same filer. This isn't going to be true everywhere. drop is meant
|
|
|
|
|
to contain your username, and be in /etc/sudoers.d/.
|
|
|
|
|
|
|
|
|
|
- ssh needs to be setup so that the user doesn't need to provide
|
|
|
|
|
credentials. The current way is that I just generated a bos-local
|
|
|
|
|
key and use that. This isn't a terrible idea, but should at least
|
|
|
|
|
be automated on setup.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
- ~/.riak_test.config should have a harness entry like this:
|
|
|
|
|
|
|
|
|
|
{rtperf,
|
|
|
|
|
[
|
|
|
|
|
{rt_deps, ["/usr/local/basho/evan/riak2.0/deps"]},
|
|
|
|
|
%% should be really long to allow full bitcasks to
|
|
|
|
|
%% come up
|
|
|
|
|
{rt_max_wait_time, 600000000000000},
|
|
|
|
|
{basho_bench, "/usr/local/basho/evan/basho_bench/"},
|
|
|
|
|
{basho_bench_statedir, "/tmp/bb_seqstate/"},
|
|
|
|
|
{rt_retry_delay, 500},
|
|
|
|
|
{rt_harness, rtperf},
|
|
|
|
|
{load_intercepts, false},
|
|
|
|
|
{perf_builds, "/usr/local/basho/evan/"},
|
|
|
|
|
{perf_loadgens, ["localhost", "r2s24"]},
|
|
|
|
|
{rtdev_path, []}
|
|
|
|
|
|
|
|
|
|
]}.
|
|
|
|
|
|
|
|
|
|
at least right now. this will problably change as the configuration
|
|
|
|
|
stuff gets sorted out. some of these may not be necessary.
|
|
|
|
|
rt_max_wait_time is (could maybe be set to infinity? maybe by the
|
|
|
|
|
harness?), perf_* and basho_bench* are also critical. rt_deps
|
|
|
|
|
should maybe be dynamic?
|
