def test_1stLevel_agg_die(self, logger, cfg, obj):
kill_ldmsd(hosts=[obj['agg_1stLevel_host']],
xprt=cfg.AGG_XPRT,
port=cfg.AGG_PORT)
is_running = is_ldmsd_running(hosts=[obj['agg_1stLevel_host']],
xprt=cfg.AGG_XPRT,
port=cfg.AGG_PORT)
if is_running[obj['agg_1stLevel_host']]:
kill_9_ldmsd(hosts=[obj['agg_1stLevel_host']],
xprt=cfg.AGG_XPRT,
port=cfg.AGG_PORT)
remove_file(hosts=[obj['agg_1stLevel_host']],
filepath=cfg.AGG_SOCK)
is_running = is_ldmsd_running(hosts=[obj['agg_1stLevel_host']],
xprt=cfg.AGG_XPRT,
port=cfg.AGG_PORT)
assert (not is_running[obj['agg_1stLevel_host']])
sleep(3)
is_samplerd_running = is_ldmsd_running(hosts=[obj['samplerd_host']],
xprt=cfg.SAMPLERD_XPRT,
port=cfg.SAMPLERD_PORT)
assert (is_samplerd_running[obj['samplerd_host']])
is_2ndLevel_agg_running = is_ldmsd_running(
hosts=[obj['agg_2ndLevel_host']],
xprt=cfg.AGG2_XPRT,
port=cfg.AGG2_PORT)
assert (is_2ndLevel_agg_running[obj['agg_2ndLevel_host']])
def test_samplerd_die(self, logger, cfg, obj):
kill_ldmsd(hosts=obj['samplerd_host'],
xprt=cfg.SAMPLERD_XPRT,
port=cfg.SAMPLERD_PORT)
is_running = is_ldmsd_running(hosts=obj['samplerd_host'],
xprt=cfg.SAMPLERD_XPRT,
port=cfg.SAMPLERD_PORT)
if is_running[obj['samplerd_host']]:
kill_9_ldmsd(hosts=obj['samplerd_host'],
xprt=cfg.SAMPLERD_XPRT,
port=cfg.SAMPLERD_PORT)
remove_file(hosts=[obj['samplerd_host']],
filepath=cfg.SAMPLERD_SOCK)
# Sleep to make sure that the aggregators do not crash.
# If the test doesn't sleep, the aggregators could die after
# the check point.
sleep(3)
is_agg_running = is_ldmsd_running(hosts=[obj['agg_1stLevel_host']],
xprt=cfg.AGG_XPRT,
port=cfg.AGG_PORT)
assert (is_agg_running[obj['agg_1stLevel_host']])
is_agg_running = is_ldmsd_running(hosts=[obj['agg_2ndLevel_host']],
xprt=cfg.AGG2_XPRT,
port=cfg.AGG2_PORT)
assert (is_agg_running[obj['agg_2ndLevel_host']])
def stop_test_ldmsds(hosts, xprt, port, sock, log):
kill_ldmsd(hosts=hosts, xprt=xprt, port=port)
is_running = is_ldmsd_running(hosts, xprt, port)
not_died = filter(lambda host: is_running[host], is_running)
if len(not_died) > 0:
log.debug("---- kill 9 samplerd: {0}".format(",".join(not_died)))
kill_9_ldmsd(hosts=not_died, xprt=xprt, port=port)
remove_file(hosts=hosts, filepath=sock)
def test_1stLevel_agg_comeback(self, logger, cfg, obj):
start_ldmsd(hosts = [obj['agg_1stLevel_host']], xprt = cfg.AGG_XPRT,
port = cfg.AGG_PORT, log = cfg.AGG_LOG,
sock = cfg.AGG_SOCK,
inet_ctrl_port = cfg.AGG_INET_CTRL_PORT)
is_1st_agg_started = is_ldmsd_running(hosts = obj['agg_1stLevel_host'],
xprt = cfg.AGG_XPRT,
port = cfg.AGG_PORT)
assert(is_1st_agg_started[obj['agg_1stLevel_host']])
def test_samplerd_die(self, logger, cfg, obj):
kill_ldmsd(hosts=obj['samplerd_host'],
xprt=cfg.SAMPLERD_XPRT,
port=cfg.SAMPLERD_PORT)
is_running = is_ldmsd_running(hosts=obj['samplerd_host'],
xprt=cfg.SAMPLERD_XPRT,
port=cfg.SAMPLERD_PORT)
if is_running[obj['samplerd_host']]:
kill_9_ldmsd(hosts=obj['samplerd_host'],
xprt=cfg.SAMPLERD_XPRT,
port=cfg.SAMPLERD_PORT)
remove_file(hosts=obj['samplerd_host'], filepath=cfg.SAMPLERD_SOCK)
sleep(3)
is_agg_running = is_ldmsd_running(hosts=obj['agg_host'],
xprt=cfg.AGG_XPRT,
port=cfg.AGG_PORT)
assert (is_agg_running[obj['agg_host']])
def test_agg_running_after_store(self, cfg, obj):
nap = 5
sleep(nap)
t = nap
while (t <= 60):
is_running = is_ldmsd_running(hosts=obj['agg_host'],
xprt=cfg.AGG_XPRT,
port=cfg.AGG_PORT)
assert (is_running[obj['agg_host']])
sleep(nap)
t += nap
def test_samplerd_comeback(self, logger, cfg, obj):
start_ldmsd(hosts=obj['samplerd_host'],
xprt=cfg.SAMPLERD_XPRT,
port=cfg.SAMPLERD_PORT,
log=cfg.SAMPLERD_LOG,
sock=cfg.SAMPLERD_SOCK,
test_set_name=cfg.TEST_INSTANCE_PREFIX_NAME,
test_set_count=cfg.NUM_TEST_INSTANCES_PER_HOST,
test_metric_count=cfg.TEST_INSTANCE_NUM_METRICS,
inet_ctrl_port=cfg.SAMPLERD_INET_CTRL_PORT)
is_samplerd_started = is_ldmsd_running(hosts=obj['samplerd_host'],
xprt=cfg.SAMPLERD_XPRT,
port=cfg.SAMPLERD_PORT)
assert (is_samplerd_started[obj['samplerd_host']])
sleep(3)
is_running = is_ldmsd_running(hosts=obj['agg_1stLevel_host'],
xprt=cfg.AGG_XPRT,
port=cfg.AGG_PORT)
assert (is_running[obj['agg_1stLevel_host']])
def test_2ndLevel_running_after_store(self, cfg, obj, logger):
nap = 5
sleep(nap)
t = nap
while (t <= 60):
is_running = is_ldmsd_running(hosts=obj['agg_2ndLevel_host'],
xprt=cfg.AGG2_XPRT,
port=cfg.AGG2_PORT)
assert (is_running[obj['agg_2ndLevel_host']])
logger.info("sleeping for {0}".format(nap))
sleep(nap)
t += nap
def start_test_ldmsds(**kwargs):
hosts = kwargs['hosts']
xprt = kwargs['xprt']
port = kwargs['port']
start_ldmsd(**kwargs)
sleep(1)
is_running = is_ldmsd_running(hosts=hosts, xprt=xprt, port=port)
not_started = filter(lambda host: not is_running[host], is_running)
if len(not_started) > 0:
raise Exception("Failed to start ldmsd_samplerd on " +
",".join(not_started))
def main(argv = None):
if argv is None:
argv = sys.argv
else:
sys.argv.extend(argv)
try:
parser = ArgumentParser()
parser.add_argument('--config-file', help = "Path to the config file", required = True)
parser.add_argument('--start-samplerd', help = "start samplerd ldmsd",
action = "store_true")
parser.add_argument('--kill-samplerd', action = "store_true",
help = "kill samplerd ldmsd")
parser.add_argument('--start-agg', action = "store_true",
help = "start agg ldmsd")
parser.add_argument('--kill-agg', action = "store_true",
help = "kill agg ldmsd")
parser.add_argument('--kill-9-samplerd', action = "store_true",
help = "kill 9 samplerd ldmsd")
parser.add_argument('--kill-9-agg', action = "store_true",
help = "kill 9 agg ldmsd")
parser.add_argument('--remove-samplerd-files', action = "store_true",
help = "Remove the samplerd's log and sock files")
parser.add_argument('--remove-agg-files', action = "store_true",
help = "Remove the aggregataor(s)'s log and sock files")
parser.add_argument('--check-samplerd', action = "store_true",
help = "Check samplerd ldmsd running")
parser.add_argument('--check-agg', action = "store_true",
help = "Check agg ldmsd running")
parser.add_argument('--samplerd-pid', help = "Get the samplerd PIDs",
action = "store_true")
parser.add_argument('--agg-pid', help = "Get the agg PIDs",
action = "store_true")
args = parser.parse_args()
cfg = get_var_from_file(module_name = "cfg", filepath = args.config_file)
if args.start_samplerd:
print "start samplerd.."
start_ldmsd(hosts = cfg.SAMPLERD_HOSTS,
xprt = cfg.SAMPLERD_XPRT, port = cfg.SAMPLERD_PORT,
log = cfg.SAMPLERD_LOG, sockname = cfg.SAMPLERD_SOCK)
if args.start_agg:
print "start agg.."
start_ldmsd(hosts = cfg.AGG_HOSTS, xprt = cfg.AGG_XPRT, port = cfg.AGG_PORT,
log = cfg.AGG_LOG, sockname = cfg.AGG_SOCK)
if args.kill_samplerd:
print "kill samplerd.."
kill_ldmsd(hosts = cfg.SAMPLERD_HOSTS, xprt = cfg.SAMPLERD_XPRT,
port = cfg.SAMPLERD_PORT)
if args.kill_agg:
print "kill agg.."
kill_ldmsd(hosts = cfg.AGG_HOSTS, xprt = cfg.AGG_XPRT, port = cfg.AGG_PORT)
if args.kill_9_samplerd:
print "kill 9 samplerd.."
kill_9_ldmsd(hosts = cfg.SAMPLERD_HOSTS, xprt = cfg.SAMPLERD_XPRT,
port = cfg.SAMPLERD_PORT)
if args.kill_9_agg:
print "kill 9 agg.."
kill_9_ldmsd(hosts = cfg.AGG_HOSTS, xprt = cfg.AGG_XPRT, port = cfg.AGG_PORT)
if args.remove_samplerd_files:
print "Removing the files of samplerd"
remove_file(cfg.SAMPLERD_HOSTS, cfg.SAMPLERD_LOG)
remove_file(cfg.SAMPLERD_HOSTS, cfg.SAMPLERD_SOCK)
if args.remove_agg_files:
print "Removing the files of aggregators"
remove_file(cfg.AGG_HOSTS, cfg.AGG_LOG)
remove_file(cfg.AGG_HOSTS, cfg.AGG_SOCK)
if args.check_samplerd:
print "Check samplerd ldmsd running? ...."
print is_ldmsd_running(hosts = cfg.SAMPLERD_HOSTS,
xprt = cfg.SAMPLERD_XPRT,
port = cfg.SAMPLERD_PORT)
if args.check_agg:
print "Check agg ldmsd running? ...."
print is_ldmsd_running(hosts = cfg.AGG_HOSTS, xprt = cfg.AGG_XPRT,
port = cfg.AGG_PORT)
if args.samplerd_pid:
print "Getting samplerd pid"
print get_ldmsd_pid(hosts = cfg.SAMPLERD_HOSTS, xprt = cfg.SAMPLERD_XPRT,
port = cfg.SAMPLERD_PORT)
if args.agg_pid:
print "Getting agg pid"
print get_ldmsd_pid(hosts = cfg.AGG_HOSTS, xprt = cfg.AGG_XPRT,
port = cfg.AGG_PORT)
except KeyboardInterrupt:
return 0
except Exception:
traceback.print_exc()
return 2
