def initialize(self):
super(LibrbdFio, self).initialize()
# Clean and Create the run directory
common.clean_remote_dir(self.run_dir)
common.make_remote_dir(self.run_dir)
logger.info('Pausing for 60s for idle monitoring.')
monitoring.start("%s/idle_monitoring" % self.run_dir)
time.sleep(60)
monitoring.stop()
common.sync_files('%s/*' % self.run_dir, self.out_dir)
self.mkimages()
# populate the fio files
ps = []
logger.info('Attempting to populating fio files...')
if (self.use_existing_volumes == False):
for volnum in xrange(self.volumes_per_client):
rbd_name = 'cbt-librbdfio-`%s`-%d' % (common.get_fqdn_cmd(), volnum)
pre_cmd = 'sudo %s --ioengine=rbd --clientname=admin --pool=%s --rbdname=%s --invalidate=0 --rw=write --numjobs=%s --bs=4M --size %dM %s --output-format=%s > /dev/null' % (self.cmd_path, self.pool_name, rbd_name, self.numjobs, self.vol_size, self.names, self.fio_out_format)
p = common.pdsh(settings.getnodes('clients'), pre_cmd)
ps.append(p)
for p in ps:
p.wait()
return True
def _run(self, mode, run_dir, out_dir):
# We'll always drop caches for rados bench
self.dropcaches()
if self.concurrent_ops:
concurrent_ops_str = '--concurrent-ios %s' % self.concurrent_ops
op_size_str = '-b %s' % self.op_size
common.make_remote_dir(run_dir)
# dump the cluster config
common.dump_config(run_dir)
monitoring.start(run_dir)
# Run rados bench
print 'Running radosbench read test.'
ps = []
for i in xrange(self.concurrent_procs):
out_file = '%s/output.%s' % (run_dir, i)
objecter_log = '%s/objecter.%s.log' % (run_dir, i)
p = common.pdsh(settings.getnodes('clients'), '/usr/bin/rados -p rados-bench-`hostname -s`-%s %s bench %s %s %s --no-cleanup 2> %s > %s' % (i, op_size_str, self.time, mode, concurrent_ops_str, objecter_log, out_file))
ps.append(p)
for p in ps:
p.wait()
monitoring.stop(run_dir)
# Get the historic ops
common.dump_historic_ops(run_dir)
common.sync_files('%s/*' % run_dir, out_dir)
def run(self):
super(CephTestRados, self).run()
# Remake the pool
self.mkpool()
self.dropcaches()
self.cluster.dump_config(self.run_dir)
monitoring.start(self.run_dir)
time.sleep(5)
# Run the backfill testing thread if requested
if 'recovery_test' in self.cluster.config:
recovery_callback = self.recovery_callback
self.cluster.create_recovery_test(self.run_dir, recovery_callback)
logger.info('Running ceph_test_rados.')
ps = []
for i in xrange(1):
p = common.pdsh(settings.getnodes('clients'), self.mkcmd())
ps.append(p)
for p in ps:
p.wait()
# If we were doing recovery, wait until it's done.
if 'recovery_test' in self.cluster.config:
self.cluster.wait_recovery_done()
monitoring.stop(self.run_dir)
# Finally, get the historic ops
self.cluster.dump_historic_ops(self.run_dir)
common.sync_files('%s/*' % self.run_dir, self.out_dir)
def initialize(self):
super(LibrbdFio, self).initialize()
print "Running scrub monitoring."
monitoring.start("%s/scrub_monitoring" % self.run_dir)
self.cluster.check_scrub()
monitoring.stop()
print "Pausing for 60s for idle monitoring."
monitoring.start("%s/idle_monitoring" % self.run_dir)
time.sleep(60)
monitoring.stop()
common.sync_files("%s/*" % self.run_dir, self.out_dir)
self.mkimages()
# Create the run directory
common.make_remote_dir(self.run_dir)
# populate the fio files
ps = []
print "Attempting to populating fio files..."
for i in xrange(self.volumes_per_client):
pre_cmd = (
"sudo %s --ioengine=rbd --clientname=admin --pool=%s --rbdname=cbt-librbdfio-`hostname -s`-%d --invalidate=0 --rw=write --numjobs=%s --bs=4M --size %dM %s > /dev/null"
% (self.cmd_path, self.poolname, i, self.numjobs, self.vol_size, self.names)
)
p = common.pdsh(settings.getnodes("clients"), pre_cmd)
ps.append(p)
for p in ps:
p.wait()
return True
def run(self):
super(LibrbdFio, self).run()
# We'll always drop caches for rados bench
self.dropcaches()
# dump the cluster config
self.cluster.dump_config(self.run_dir)
monitoring.start(self.run_dir)
time.sleep(5)
# Run the backfill testing thread if requested
if "recovery_test" in self.cluster.config:
recovery_callback = self.recovery_callback
self.cluster.create_recovery_test(self.run_dir, recovery_callback)
print "Running rbd fio %s test." % self.mode
ps = []
for i in xrange(self.volumes_per_client):
fio_cmd = self.mkfiocmd(i)
p = common.pdsh(settings.getnodes("clients"), fio_cmd)
ps.append(p)
for p in ps:
p.wait()
# If we were doing recovery, wait until it's done.
if "recovery_test" in self.cluster.config:
self.cluster.wait_recovery_done()
monitoring.stop(self.run_dir)
# Finally, get the historic ops
self.cluster.dump_historic_ops(self.run_dir)
common.sync_files("%s/*" % self.run_dir, self.out_dir)
def initialize(self):
super(RbdFio, self).initialize()
logger.info('Running scrub monitoring.')
monitoring.start("%s/scrub_monitoring" % self.run_dir)
self.cluster.check_scrub()
monitoring.stop()
logger.info('Pausing for 60s for idle monitoring.')
monitoring.start("%s/idle_monitoring" % self.run_dir)
time.sleep(60)
monitoring.stop()
common.sync_files('%s/*' % self.run_dir, self.out_dir)
self.mkimages()
# Create the run directory
common.make_remote_dir(self.run_dir)
# populate the fio files
logger.info('Attempting to populating fio files...')
pre_cmd = 'sudo %s --ioengine=%s --rw=write --numjobs=%s --bs=4M --size %dM %s > /dev/null' % (self.cmd_path, self.ioengine, self.numjobs, self.vol_size*0.9, self.names)
common.pdsh(settings.getnodes('clients'), pre_cmd).communicate()
return True
def _run(self, mode, run_dir, out_dir):
# We'll always drop caches for rados bench
self.dropcaches()
if self.concurrent_ops:
concurrent_ops_str = "--concurrent-ios %s" % self.concurrent_ops
op_size_str = "-b %s" % self.op_size
common.make_remote_dir(run_dir)
monitoring.start(run_dir)
# Run rados bench
print "Running radosbench read test."
ps = []
for i in xrange(self.concurrent_procs):
out_file = "%s/output.%s" % (run_dir, i)
objecter_log = "%s/objecter.%s.log" % (run_dir, i)
p = common.pdsh(
settings.cluster.get("clients"),
"/usr/bin/rados -p rados-bench-%s %s bench %s %s %s --no-cleanup 2> %s > %s"
% (i, op_size_str, self.time, mode, concurrent_ops_str, objecter_log, out_file),
)
ps.append(p)
for p in ps:
p.wait()
monitoring.stop(run_dir)
common.sync_files("%s/*" % run_dir, out_dir)
def run(self):
super(RbdFio, self).run()
# Set client readahead
self.set_client_param('read_ahead_kb', self.client_ra)
# We'll always drop caches for rados bench
self.dropcaches()
common.make_remote_dir(self.run_dir)
monitoring.start(self.run_dir)
# Run rados bench
print 'Running rbd fio %s test.' % self.mode
names = ""
for i in xrange(self.concurrent_procs):
names += "--name=%s/mnt/rbdfio-`hostname -s`-%d/cbt-rbdfio " % (self.tmp_dir, i)
out_file = '%s/output' % self.run_dir
fio_cmd = 'sudo fio --rw=%s -ioengine=%s --runtime=%s --numjobs=1 --direct=1 --bs=%dB --iodepth=%d --size %dM %s > %s' % (self.mode, self.ioengine, self.time, self.op_size, self.iodepth, self.vol_size * 9/10, names, out_file)
common.pdsh(settings.getnodes('clients'), fio_cmd).communicate()
# ps = []
# for i in xrange(self.concurrent_procs):
# out_file = '%s/output.%s' % (self.run_dir, i)
# p = common.pdsh(settings.cluster.get('clients'), 'sudo fio --rw=%s -ioengine=%s --runtime=%s --name=/srv/rbdfio-`hostname -s`-%d/cbt-rbdfio --numjobs=1 --direct=1 --bs=%dB --iodepth=%d --size %dM > %s' % (self.mode, self.ioengine, self.time, i, self.op_size, self.iodepth, self.vol_size * 9/10, out_file))
# ps.append(p)
# for p in ps:
# p.wait()
monitoring.stop(self.run_dir)
common.sync_files('%s/*' % self.run_dir, self.out_dir)
def run(self):
super(KvmRbdFio, self).run()
# We'll always drop caches for rados bench
self.dropcaches()
monitoring.start(self.run_dir)
time.sleep(5)
names = ""
for i in xrange(self.concurrent_procs):
names += "--name=/srv/rbdfio-`hostname -s`-%d/cbt-kvmrbdfio " % i
out_file = '%s/output' % self.run_dir
pre_cmd = 'sudo fio --rw=read -ioengine=sync --numjobs=1 --bs=4M --runtime=1 --size %dM %s > /dev/null' % (self.vol_size * 9/10, names)
fio_cmd = 'sudo fio --rw=%s -ioengine=%s --runtime=%s --numjobs=1 --direct=1 --bs=%dB --iodepth=%d --size %dM %s > %s' % (self.mode, self.ioengine, self.time, self.op_size, self.iodepth, self.vol_size * 9/10, names, out_file)
print 'Attempting to populating fio files...'
common.pdsh(settings.cluster.get('clients'), pre_cmd).communicate()
print 'Running rbd fio %s test.' % self.mode
common.pdsh(settings.cluster.get('clients'), fio_cmd).communicate()
# ps = []
# for i in xrange(self.concurrent_procs):
# out_file = '%s/output.%s' % (self.run_dir, i)
# p = common.pdsh(settings.cluster.get('clients'), 'sudo fio --rw=%s -ioengine=%s --runtime=%s --name=/srv/rbdfio-`hostname -s`-%d/cbt-rbdfio --numjobs=1 --direct=1 --bs=%dB --iodepth=%d --size %dM > %s' % (self.mode, self.ioengine, self.time, i, self.op_size, self.iodepth, self.vol_size * 9/10, out_file))
# ps.append(p)
# for p in ps:
# p.wait()
monitoring.stop(self.run_dir)
common.sync_files('%s/*' % self.run_dir, self.out_dir)
def initialize(self):
super(RbdFio, self).initialize()
self.cleanup()
if not self.use_existing:
self.cluster.initialize()
self.cluster.dump_config(self.run_dir)
# Setup the pools
monitoring.start("%s/pool_monitoring" % self.run_dir)
common.pdsh(settings.getnodes('head'), 'sudo ceph -c %s osd pool create rbdfio %d %d' % (self.tmp_conf, self.pgs, self.pgs)).communicate()
common.pdsh(settings.getnodes('head'), 'sudo ceph -c %s osd pool set rbdfio size 1' % self.tmp_conf).communicate()
print 'Checking Healh after pool creation.'
self.cluster.check_health()
monitoring.stop()
# Mount the filesystem
common.pdsh(settings.getnodes('clients'), 'sudo modprobe rbd').communicate()
for i in xrange(self.concurrent_procs):
common.pdsh(settings.getnodes('clients'), 'sudo rbd -c %s create rbdfio/rbdfio-`hostname -s`-%d --size %d' % (self.tmp_conf, i, self.vol_size)).communicate()
common.pdsh(settings.getnodes('clients'), 'sudo echo "%s %s rbdfio rbdfio-`hostname -s`-%d" | sudo tee /sys/bus/rbd/add && sudo /sbin/udevadm settle' % (self.rbdadd_mons, self.rbdadd_options, i)).communicate()
common.pdsh(settings.getnodes('clients'), 'sudo mkfs.xfs /dev/rbd/rbdfio/rbdfio-`hostname -s`-%d' % i).communicate()
common.pdsh(settings.getnodes('clients'), 'sudo mkdir -p -m0755 -- %s/mnt/rbdfio-`hostname -s`-%d' % (self.tmp_dir, i)).communicate()
common.pdsh(settings.getnodes('clients'), 'sudo mount -t xfs -o noatime,inode64 /dev/rbd/rbdfio/rbdfio-`hostname -s`-%d %s/mnt/rbdfio-`hostname -s`-%d' % (i, self.tmp_dir, i)).communicate()
print 'Running scrub monitoring'
monitoring.start("%s/scrub_monitoring" % self.run_dir)
self.cluster.check_scrub()
monitoring.stop()
# Create the run directory
common.make_remote_dir(self.run_dir)
def run(self):
# First create a credential file for each gateway
self.mkcredfiles()
# We'll always drop caches for rados bench
self.dropcaches()
# dump the cluster config
self.cluster.dump_config(self.run_dir)
# Run the backfill testing thread if requested
if 'recovery_test' in self.cluster.config:
recovery_callback = self.recovery_callback
self.cluster.create_recovery_test(self.run_dir, recovery_callback)
# Run getput
monitoring.start(self.run_dir)
logger.info('Running getput %s test.' % self.test)
ps = []
for i in xrange(0, len(self.auth_urls)):
cmd = self.mkgetputcmd("%s/gw%02d.cred" % (self.run_dir, i), i)
p = common.pdsh(settings.getnodes('clients'), cmd)
ps.append(p)
for p in ps:
p.wait()
monitoring.stop(self.run_dir)
# If we were doing recovery, wait until it's done.
if 'recovery_test' in self.cluster.config:
self.cluster.wait_recovery_done()
# Finally, get the historic ops
self.cluster.dump_historic_ops(self.run_dir)
common.sync_files('%s/*' % self.run_dir, self.out_dir)
def initialize(self):
super(LibrbdFio, self).initialize()
print 'Running scrub monitoring.'
monitoring.start("%s/scrub_monitoring" % self.run_dir)
self.cluster.check_scrub()
monitoring.stop()
print 'Pausing for 60s for idle monitoring.'
monitoring.start("%s/idle_monitoring" % self.run_dir)
time.sleep(60)
monitoring.stop()
common.sync_files('%s/*' % self.run_dir, self.out_dir)
self.mkimages()
# Create the run directory
common.make_remote_dir(self.run_dir)
# populate the fio files
print 'Attempting to populating fio files...'
pre_cmd = 'sudo %s --ioengine=rbd --clientname=admin --pool=%s --rbdname=cbt-librbdfio-`hostname -s` --invalidate=0 --rw=write --numjobs=%s --bs=4M --size %dM %s > /dev/null' % (self.cmd_path, self.poolname, self.numjobs, self.vol_size, self.names)
common.pdsh(settings.getnodes('clients'), pre_cmd).communicate()
return True
def initialize(self):
super(Cosbench, self).initialize()
logger.debug('Running cosbench and radosgw check.')
self.prerun_check()
logger.debug('Running scrub monitoring.')
monitoring.start("%s/scrub_monitoring" % self.run_dir)
self.cluster.check_scrub()
monitoring.stop()
logger.debug('Pausing for 60s for idle monitoring.')
monitoring.start("%s/idle_monitoring" % self.run_dir)
time.sleep(60)
monitoring.stop()
common.sync_files('%s' % self.run_dir, self.out_dir)
# Create the run directory
common.make_remote_dir(self.run_dir)
conf = self.config
if not self.config["template"]:
self.config["template"] = "default"
self.config["workload"] = self.choose_template("default", conf)
self.prepare_xml(self.config["workload"])
return True
def mkimages(self):
monitoring.start("%s/pool_monitoring" % self.run_dir)
self.cluster.rmpool(self.poolname, self.pool_profile)
self.cluster.mkpool(self.poolname, self.pool_profile)
for node in settings.getnodes('clients').split(','):
node = node.rpartition("@")[2]
common.pdsh(settings.getnodes('head'), '/usr/bin/rbd create cbt-librbdfio-%s --size %s --pool %s --order %s' % (node, self.vol_size, self.poolname, self.vol_order)).communicate()
monitoring.stop()
def mkpools(self):
monitoring.start("%s/pool_monitoring" % self.run_dir)
for i in xrange(self.concurrent_procs):
for node in settings.getnodes('clients').split(','):
node = node.rpartition("@")[2]
self.cluster.rmpool('rados-bench-%s-%s' % (node, i), self.pool_profile)
self.cluster.mkpool('rados-bench-%s-%s' % (node, i), self.pool_profile)
monitoring.stop()
def run(self):
super(LibrbdFio, self).run()
# We'll always drop caches for rados bench
self.dropcaches()
# dump the cluster config
self.cluster.dump_config(self.run_dir)
monitoring.start(self.run_dir)
time.sleep(5)
out_file = '%s/output' % self.run_dir
fio_cmd = 'sudo %s --ioengine=rbd --clientname=admin --pool=%s --rbdname=cbt-librbdfio-`hostname -s` --invalidate=0' % (self.cmd_path_full, self.poolname)
fio_cmd += ' --rw=%s' % self.mode
if (self.mode == 'readwrite' or self.mode == 'randrw'):
fio_cmd += ' --rwmixread=%s --rwmixwrite=%s' % (self.rwmixread, self.rwmixwrite)
# fio_cmd += ' --ioengine=%s' % self.ioengine
if self.time is not None:
fio_cmd += ' --runtime=%s' % self.time
if self.ramp is not None:
fio_cmd += ' --ramp_time=%s' % self.ramp
fio_cmd += ' --numjobs=%s' % self.numjobs
fio_cmd += ' --direct=1'
fio_cmd += ' --bs=%dB' % self.op_size
fio_cmd += ' --iodepth=%d' % self.iodepth
fio_cmd += ' --end_fsync=%s' % self.end_fsync
# if self.vol_size:
# fio_cmd += ' -- size=%dM' % self.vol_size
fio_cmd += ' --write_iops_log=%s' % out_file
fio_cmd += ' --write_bw_log=%s' % out_file
fio_cmd += ' --write_lat_log=%s' % out_file
if 'recovery_test' in self.cluster.config:
fio_cmd += ' --time_based'
if self.random_distribution is not None:
fio_cmd += ' --random_distribution=%s' % self.random_distribution
if self.log_avg_msec is not None:
fio_cmd += ' --log_avg_msec=%s' % self.log_avg_msec
fio_cmd += ' %s > %s' % (self.names, out_file)
# Run the backfill testing thread if requested
if 'recovery_test' in self.cluster.config:
recovery_callback = self.recovery_callback
self.cluster.create_recovery_test(self.run_dir, recovery_callback)
print 'Running rbd fio %s test.' % self.mode
common.pdsh(settings.getnodes('clients'), fio_cmd).communicate()
# If we were doing recovery, wait until it's done.
if 'recovery_test' in self.cluster.config:
self.cluster.wait_recovery_done()
monitoring.stop(self.run_dir)
# Finally, get the historic ops
self.cluster.dump_historic_ops(self.run_dir)
common.sync_files('%s/*' % self.run_dir, self.out_dir)
def mkimages(self):
monitoring.start("%s/pool_monitoring" % self.run_dir)
self.cluster.rmpool(self.poolname, self.pool_profile)
self.cluster.mkpool(self.poolname, self.pool_profile)
common.pdsh(settings.getnodes('clients'), '/usr/bin/rbd create cbt-kernelrbdfio-`hostname -s` --size %s --pool %s' % (self.vol_size, self.poolname)).communicate()
common.pdsh(settings.getnodes('clients'), 'sudo rbd map cbt-kernelrbdfio-`hostname -s` --pool %s --id admin' % self.poolname).communicate()
common.pdsh(settings.getnodes('clients'), 'sudo mkfs.xfs /dev/rbd/cbt-kernelrbdfio/cbt-kernelrbdfio-`hostname -s`').communicate()
common.pdsh(settings.getnodes('clients'), 'sudo mkdir -p -m0755 -- %s/cbt-kernelrbdfio-`hostname -s`' % self.cluster.mnt_dir).communicate()
common.pdsh(settings.getnodes('clients'), 'sudo mount -t xfs -o noatime,inode64 /dev/rbd/cbt-kernelrbdfio/cbt-kernelrbdfio-`hostname -s` %s/cbt-kernelrbdfio-`hostname -s`' % self.cluster.mnt_dir).communicate()
monitoring.stop()
def _run(self, mode, run_dir, out_dir):
# We'll always drop caches for rados bench
self.dropcaches()
if self.concurrent_ops:
concurrent_ops_str = '--concurrent-ios %s' % self.concurrent_ops
#determine rados version
rados_version_str, err = common.pdsh(settings.getnodes('head'), '/usr/bin/rados -v').communicate()
m = re.findall("version (\d+)", rados_version_str)
rados_version = int(m[0])
if mode in ['write'] or rados_version < 9:
op_size_str = '-b %s' % self.op_size
else:
op_size_str = ''
common.make_remote_dir(run_dir)
# dump the cluster config
self.cluster.dump_config(run_dir)
# Run the backfill testing thread if requested
if 'recovery_test' in self.cluster.config:
recovery_callback = self.recovery_callback
self.cluster.create_recovery_test(run_dir, recovery_callback)
# Run rados bench
monitoring.start(run_dir)
logger.info('Running radosbench %s test.' % mode)
ps = []
for i in xrange(self.concurrent_procs):
out_file = '%s/output.%s' % (run_dir, i)
objecter_log = '%s/objecter.%s.log' % (run_dir, i)
# default behavior is to use a single storage pool
pool_name = self.pool
run_name = '--run-name %s`hostname -s`-%s'%(self.object_set_id, i)
if self.pool_per_proc: # support previous behavior of 1 storage pool per rados process
pool_name = 'rados-bench-`hostname -s`-%s'%i
run_name = ''
rados_bench_cmd = '%s -c %s -p %s bench %s %s %s %s %s --no-cleanup 2> %s > %s' % \
(self.cmd_path_full, self.tmp_conf, pool_name, op_size_str, self.time, mode, concurrent_ops_str, run_name, objecter_log, out_file)
p = common.pdsh(settings.getnodes('clients'), rados_bench_cmd)
ps.append(p)
for p in ps:
p.wait()
monitoring.stop(run_dir)
# If we were doing recovery, wait until it's done.
if 'recovery_test' in self.cluster.config:
self.cluster.wait_recovery_done()
# Finally, get the historic ops
self.cluster.dump_historic_ops(run_dir)
common.sync_files('%s/*' % run_dir, out_dir)
def run(self):
super(RbdFio, self).run()
# Set client readahead
self.set_client_param('read_ahead_kb', self.client_ra)
# We'll always drop caches for rados bench
self.dropcaches()
monitoring.start(self.run_dir)
# Run the backfill testing thread if requested
if 'recovery_test' in self.cluster.config:
recovery_callback = self.recovery_callback
self.cluster.create_recovery_test(self.run_dir, recovery_callback)
time.sleep(5)
out_file = '%s/output' % self.run_dir
fio_cmd = 'sudo %s' % (self.cmd_path_full)
fio_cmd += ' --rw=%s' % self.mode
if (self.mode == 'readwrite' or self.mode == 'randrw'):
fio_cmd += ' --rwmixread=%s --rwmixwrite=%s' % (self.rwmixread, self.rwmixwrite)
fio_cmd += ' --ioengine=%s' % self.ioengine
if self.time is not None:
fio_cmd += ' --runtime=%s' % self.time
if self.ramp is not None:
fio_cmd += ' --ramp_time=%s' % self.ramp
fio_cmd += ' --numjobs=%s' % self.numjobs
fio_cmd += ' --direct=1'
fio_cmd += ' --bs=%dB' % self.op_size
fio_cmd += ' --iodepth=%d' % self.iodepth
if self.vol_size:
fio_cmd += ' --size=%dM' % (int(self.vol_size) * 0.9)
fio_cmd += ' --write_iops_log=%s' % out_file
fio_cmd += ' --write_bw_log=%s' % out_file
fio_cmd += ' --write_lat_log=%s' % out_file
if 'recovery_test' in self.cluster.config:
fio_cmd += ' --time_based'
if self.random_distribution is not None:
fio_cmd += ' --random_distribution=%s' % self.random_distribution
fio_cmd += ' %s > %s' % (self.names, out_file)
if self.log_avg_msec is not None:
fio_cmd += ' --log_avg_msec=%s' % self.log_avg_msec
logger.info('Running rbd fio %s test.', self.mode)
common.pdsh(settings.getnodes('clients'), fio_cmd).communicate()
# If we were doing recovery, wait until it's done.
if 'recovery_test' in self.cluster.config:
self.cluster.wait_recovery_done()
monitoring.stop(self.run_dir)
# Finally, get the historic ops
self.cluster.dump_historic_ops(self.run_dir)
common.sync_files('%s/*' % self.run_dir, self.out_dir)
def run(self):
super(KvmRbdFio, self).run()
# Set client readahead
self.set_client_param('read_ahead_kb', self.client_ra)
clnts = settings.getnodes('clients')
# We'll always drop caches for rados bench
self.dropcaches()
monitoring.start(self.run_dir)
time.sleep(5)
# Run the backfill testing thread if requested
if 'recovery_test' in self.cluster.config:
recovery_callback = self.recovery_callback
self.cluster.create_recovery_test(self.run_dir, recovery_callback)
logger.info('Starting rbd fio %s test.', self.mode)
fio_process_list = []
for i in range(self.concurrent_procs):
b = self.block_devices[i % len(self.block_devices)]
bnm = os.path.basename(b)
mtpt = '/srv/rbdfio-`hostname -s`-%s' % bnm
fiopath = os.path.join(mtpt, 'fio%d.img' % i)
out_file = '%s/output.%d' % (self.run_dir, i)
fio_cmd = 'sudo %s' % self.fio_cmd
fio_cmd += ' --rw=%s' % self.mode
if (self.mode == 'readwrite' or self.mode == 'randrw'):
fio_cmd += ' --rwmixread=%s --rwmixwrite=%s' % (self.rwmixread, self.rwmixwrite)
fio_cmd += ' --ioengine=%s' % self.ioengine
fio_cmd += ' --runtime=%s' % self.time
fio_cmd += ' --ramp_time=%s' % self.ramp
if self.startdelay:
fio_cmd += ' --startdelay=%s' % self.startdelay
if self.rate_iops:
fio_cmd += ' --rate_iops=%s' % self.rate_iops
fio_cmd += ' --numjobs=%s' % self.numjobs
fio_cmd += ' --direct=1'
fio_cmd += ' --bs=%dB' % self.op_size
fio_cmd += ' --iodepth=%d' % self.iodepth
fio_cmd += ' --size=%dM' % self.vol_size
fio_cmd += ' --write_iops_log=%s' % out_file
fio_cmd += ' --write_bw_log=%s' % out_file
fio_cmd += ' --write_lat_log=%s' % out_file
if 'recovery_test' in self.cluster.config:
fio_cmd += ' --time_based'
fio_cmd += ' --name=%s > %s' % (fiopath, out_file)
fio_process_list.append(common.pdsh(clnts, fio_cmd, continue_if_error=False))
for p in fio_process_list:
p.communicate()
monitoring.stop(self.run_dir)
logger.info('Finished rbd fio test')
common.sync_files('%s/*' % self.run_dir, self.out_dir)
def initialize(self):
super(Radosbench, self).initialize()
logger.info('Pausing for 60s for idle monitoring.')
monitoring.start("%s/idle_monitoring" % self.run_dir)
time.sleep(60)
monitoring.stop()
common.sync_files('%s/*' % self.run_dir, self.out_dir)
return True
def mkimages(self):
monitoring.start("%s/pool_monitoring" % self.run_dir)
if (self.use_existing_volumes == False):
self.cluster.rmpool(self.poolname, self.pool_profile)
self.cluster.mkpool(self.poolname, self.pool_profile)
for node in settings.getnodes('clients').split(','):
for volnum in xrange(0, self.volumes_per_client):
node = node.rpartition("@")[2]
# common.pdsh(settings.getnodes('head'), '/usr/bin/rbd create cbt-librbdfio-%s-%d --size %s --pool %s --order %s' % (node, volnum, self.vol_size, self.poolname, self.vol_order)).communicate()
self.cluster.mkimage('cbt-librbdfio-%s-%d' % (node,volnum), self.vol_size, self.poolname, self.vol_order)
monitoring.stop()
def mkpools(self):
monitoring.start("%s/pool_monitoring" % self.run_dir)
if self.pool_per_proc: # allow use of a separate storage pool per process
for i in xrange(self.concurrent_procs):
for node in settings.getnodes('clients').split(','):
node = node.rpartition("@")[2]
self.cluster.rmpool('rados-bench-%s-%s' % (node, i), self.pool_profile)
self.cluster.mkpool('rados-bench-%s-%s' % (node, i), self.pool_profile)
else: # the default behavior is to use a single Ceph storage pool for all rados bench processes
self.cluster.rmpool('rados-bench-cbt', self.pool_profile)
self.cluster.mkpool('rados-bench-cbt', self.pool_profile)
monitoring.stop()
def mkimages(self):
monitoring.start("%s/pool_monitoring" % self.run_dir)
self.cluster.rmpool(self.poolname, self.pool_profile)
self.cluster.mkpool(self.poolname, self.pool_profile)
for node in settings.getnodes("clients").split(","):
for volnum in xrange(0, self.volumes_per_client):
node = node.rpartition("@")[2]
common.pdsh(
settings.getnodes("head"),
"/usr/bin/rbd create cbt-librbdfio-%s-%d --size %s --pool %s --order %s"
% (node, volnum, self.vol_size, self.poolname, self.vol_order),
).communicate()
monitoring.stop()
def run(self):
super(RawFio, self).run()
# Set client readahead
clnts = settings.getnodes('clients')
# We'll always drop caches for rados bench
self.dropcaches()
monitoring.start(self.run_dir)
time.sleep(5)
logger.info('Starting raw fio %s test.', self.mode)
fio_process_list = []
for i in range(self.concurrent_procs):
b = self.block_devices[i % len(self.block_devices)]
fiopath = b
out_file = '%s/output.%d' % (self.run_dir, i)
fio_cmd = 'sudo %s' % self.fio_cmd
fio_cmd += ' --rw=%s' % self.mode
if (self.mode == 'readwrite' or self.mode == 'randrw'):
fio_cmd += ' --rwmixread=%s --rwmixwrite=%s' % (self.rwmixread, self.rwmixwrite)
fio_cmd += ' --ioengine=%s' % self.ioengine
fio_cmd += ' --runtime=%s' % self.time
fio_cmd += ' --ramp_time=%s' % self.ramp
if self.startdelay:
fio_cmd += ' --startdelay=%s' % self.startdelay
if self.rate_iops:
fio_cmd += ' --rate_iops=%s' % self.rate_iops
fio_cmd += ' --numjobs=%s' % self.numjobs
fio_cmd += ' --direct=%s' % self.direct
fio_cmd += ' --bs=%dB' % self.op_size
fio_cmd += ' --iodepth=%d' % self.iodepth
fio_cmd += ' --size=%dM' % self.vol_size
fio_cmd += ' --write_iops_log=%s' % out_file
fio_cmd += ' --write_bw_log=%s' % out_file
fio_cmd += ' --write_lat_log=%s' % out_file
fio_cmd += ' --output-format=%s' % self.fio_out_format
if 'recovery_test' in self.cluster.config:
fio_cmd += ' --time_based'
fio_cmd += ' --name=%s > %s' % (fiopath, out_file)
logger.debug("FIO CMD: %s" % fio_cmd)
fio_process_list.append(common.pdsh(clnts, fio_cmd, continue_if_error=False))
for p in fio_process_list:
p.communicate()
monitoring.stop(self.run_dir)
logger.info('Finished raw fio test')
common.sync_files('%s/*' % self.run_dir, self.out_dir)
def mkimages(self):
monitoring.start("%s/pool_monitoring" % self.run_dir)
if (self.use_existing_volumes == False):
self.cluster.rmpool(self.pool_name, self.pool_profile)
self.cluster.mkpool(self.pool_name, self.pool_profile, 'rbd')
if self.data_pool_profile:
self.data_pool = self.pool_name + "-data"
self.cluster.rmpool(self.data_pool, self.data_pool_profile)
self.cluster.mkpool(self.data_pool, self.data_pool_profile, 'rbd')
for node in common.get_fqdn_list('clients'):
for volnum in xrange(0, self.volumes_per_client):
node = node.rpartition("@")[2]
self.cluster.mkimage('cbt-librbdfio-%s-%d' % (node,volnum), self.vol_size, self.pool_name, self.data_pool, self.vol_order)
monitoring.stop()
def shutdown(self):
nodes = settings.getnodes('clients', 'osds', 'mons', 'rgws', 'mds')
common.pdsh(nodes, 'sudo killall -9 massif-amd64-li').communicate()
common.pdsh(nodes, 'sudo killall -9 memcheck-amd64-').communicate()
common.pdsh(nodes, 'sudo killall -9 ceph-osd').communicate()
common.pdsh(nodes, 'sudo killall -9 ceph-mon').communicate()
common.pdsh(nodes, 'sudo killall -9 ceph-mds').communicate()
common.pdsh(nodes, 'sudo killall -9 rados').communicate()
common.pdsh(nodes, 'sudo killall -9 rest-bench').communicate()
common.pdsh(nodes, 'sudo killall -9 radosgw').communicate()
common.pdsh(nodes, 'sudo killall -9 radosgw-admin').communicate()
common.pdsh(nodes, 'sudo /etc/init.d/apache2 stop').communicate()
common.pdsh(nodes, 'sudo killall -9 pdsh').communicate()
monitoring.stop()
def shutdown(self):
nodes = settings.getnodes("clients", "osds", "mons", "rgws", "mds")
common.pdsh(nodes, "sudo killall -9 massif-amd64-li").communicate()
common.pdsh(nodes, "sudo killall -9 memcheck-amd64-").communicate()
common.pdsh(nodes, "sudo killall -9 ceph-osd").communicate()
common.pdsh(nodes, "sudo killall -9 ceph-mon").communicate()
common.pdsh(nodes, "sudo killall -9 ceph-mds").communicate()
common.pdsh(nodes, "sudo killall -9 rados").communicate()
common.pdsh(nodes, "sudo killall -9 rest-bench").communicate()
common.pdsh(nodes, "sudo killall -9 radosgw").communicate()
common.pdsh(nodes, "sudo killall -9 radosgw-admin").communicate()
common.pdsh(nodes, "sudo /etc/init.d/apache2 stop").communicate()
common.pdsh(nodes, "sudo killall -9 pdsh").communicate()
monitoring.stop()
def shutdown(self):
nodes = settings.getnodes('clients', 'osds', 'mons', 'rgws', 'mds')
common.pdsh(nodes, 'sudo killall -9 -q massif-amd64-li ', True)
common.pdsh(nodes, 'sudo killall -9 -q memcheck-amd64- ', True)
common.pdsh(nodes, 'sudo killall -9 -q ceph-osd', True)
common.pdsh(nodes, 'sudo killall -9 -q ceph-mon', True)
common.pdsh(nodes, 'sudo killall -9 -q ceph-mds', True)
common.pdsh(nodes, 'sudo killall -9 -q rados', True)
common.pdsh(nodes, 'sudo killall -9 -q rest-bench', True)
common.pdsh(nodes, 'sudo killall -9 -q radosgw', True)
common.pdsh(nodes, 'sudo killall -9 -q radosgw-admin', True)
common.pdsh(nodes, 'sudo /etc/init.d/apache2 stop', True)
common.pdsh(nodes, 'sudo killall -9 -q pdsh', True)
monitoring.stop()
def initialize(self):
super(Radosbench, self).initialize()
print 'Running scrub monitoring.'
monitoring.start("%s/scrub_monitoring" % self.run_dir)
self.cluster.check_scrub()
monitoring.stop()
print 'Pausing for 60s for idle monitoring.'
monitoring.start("%s/idle_monitoring" % self.run_dir)
time.sleep(60)
monitoring.stop()
common.sync_files('%s/*' % self.run_dir, self.out_dir)
return True
def initialize(self):
super(RbdFio, self).initialize()
logger.info('Pausing for 60s for idle monitoring.')
monitoring.start("%s/idle_monitoring" % self.run_dir)
time.sleep(60)
monitoring.stop()
common.sync_files('%s/*' % self.run_dir, self.out_dir)
self.mkimages()
# Create the run directory
common.make_remote_dir(self.run_dir)
# populate the fio files
logger.info('Attempting to populating fio files...')
size = self.vol_size * 0.9 / self.concurrent_procs
pre_cmd = 'sudo %s --ioengine=%s --rw=write --numjobs=%s --bs=4M --size %dM %s > /dev/null' % (
self.cmd_path, self.ioengine, self.numjobs, size, self.names)
common.pdsh(settings.getnodes('clients'), pre_cmd).communicate()
def run(self):
super(Cosbench, self).run()
self.dropcaches()
self.cluster.dump_config(self.run_dir)
monitoring.start(self.run_dir)
# Run cosbench test
try:
self._run()
except KeyboardInterrupt:
logger.warning("accept keyboard interrupt, cancel this run")
conf = self.config
stdout, stderr = common.pdsh(conf["controller"],'sh %s/cli.sh cancel %s' % (conf["cosbench_dir"], self.runid)).communicate()
logger.info("%s", stdout)
self.check_workload_status()
self.check_cosbench_res_dir()
monitoring.stop(self.run_dir)
self.cluster.dump_historic_ops(self.run_dir)
common.sync_files('%s/*' % self.run_dir, self.out_dir)
def initialize(self):
common.cleanup_tests()
if not self.use_existing:
common.setup_cluster()
common.setup_ceph()
# Create the run directory
common.make_remote_dir(self.run_dir)
# Setup the pools
monitoring.start("%s/pool_monitoring" % self.run_dir)
for i in xrange(self.concurrent_procs):
for node in settings.getnodes('clients').split(','):
node = node.rpartition("@")[2]
common.pdsh(settings.getnodes('head'), 'sudo ceph osd pool create rados-bench-%s-%s %d %d' % (node, i, self.pgs_per_pool, self.pgs_per_pool)).communicate()
common.pdsh(settings.getnodes('head'), 'sudo ceph osd pool set rados-bench-%s-%s size 1' % (node, i)).communicate()
# check the health for each pool.
print 'Checking Healh after pool creation.'
common.check_health()
monitoring.stop()
print 'Running scrub monitoring.'
monitoring.start("%s/scrub_monitoring" % self.run_dir)
common.check_scrub()
monitoring.stop()
print 'Pausing for 60s for idle monitoring.'
monitoring.start("%s/idle_monitoring" % self.run_dir)
time.sleep(60)
monitoring.stop()
common.sync_files('%s/*' % self.run_dir, self.out_dir)
return True
def initialize(self):
super(Cosbench, self).initialize()
logger.debug('Running cosbench and radosgw check.')
self.prerun_check()
logger.debug('Pausing for 60s for idle monitoring.')
monitoring.start("%s/idle_monitoring" % self.run_dir)
time.sleep(60)
monitoring.stop()
common.sync_files('%s' % self.run_dir, self.out_dir)
# Create the run directory
common.make_remote_dir(self.run_dir)
conf = self.config
if not self.config["template"]:
self.config["template"] = "default"
self.config["workload"] = self.choose_template("default", conf)
# add a "prepare" stage if mode is read or mix
if not self.container_prepare_check():
workstage_init = {
"name": "init",
"work": {"type":"init", "workers":conf["workers"], "config":"containers=r(1,%s);cprefix=%s-%s-%s" % (conf["containers_max"], conf["obj_size"], conf["mode"], conf["objects_max"])}
}
workstage_prepare = {
"name":"prepare",
"work": {
"type":"prepare",
"workers":conf["workers"],
"config":"containers=r(1,%s);objects=r(1,%s);cprefix=%s-%s-%s;sizes=c(%s)%s" %
(conf["containers_max"], conf["objects_max"], conf["obj_size"], conf["mode"], conf["objects_max"], conf["obj_size_num"], conf["obj_size_unit"])
}
}
self.config["workload"]["workflow"]["workstage"].insert(0, workstage_prepare)
self.config["workload"]["workflow"]["workstage"].insert(0, workstage_init)
self.prepare_xml(self.config["workload"])
def run(self):
super(Cosbench, self).run()
self.dropcaches()
self.cluster.dump_config(self.run_dir)
monitoring.start(self.run_dir)
# Run cosbench test
try:
self._run()
except KeyboardInterrupt:
print "[WARNING] accept keyboard interrupt, cancel this run"
conf = self.config
stdout, stderr = common.pdsh("%s@%s" % (self.user, conf["controller"]),'sh %s/cli.sh cancel %s' % (conf["cosbench_dir"], self.runid)).communicate()
print "[LOG]%s" % stdout
self.check_workload_status()
self.check_cosbench_res_dir()
monitoring.stop(self.run_dir)
self.cluster.dump_historic_ops(self.run_dir)
common.sync_files('%s/*' % self.run_dir, self.out_dir)
common.sync_files('%s/archive/%s*' % (self.config["cosbench_dir"], self.runid), self.out_dir)
def _run(self, mode, run_dir, out_dir):
# We'll always drop caches for rados bench
self.dropcaches()
if self.concurrent_ops:
concurrent_ops_str = '--concurrent-ios %s' % self.concurrent_ops
op_size_str = '-b %s' % self.op_size
common.make_remote_dir(run_dir)
# dump the cluster config
self.cluster.dump_config(run_dir)
# Run the backfill testing thread if requested
if 'recovery_test' in self.cluster.config:
recovery_callback = self.recovery_callback
self.cluster.create_recovery_test(run_dir, recovery_callback)
# Run rados bench
monitoring.start(run_dir)
print 'Running radosbench read test.'
ps = []
for i in xrange(self.concurrent_procs):
out_file = '%s/output.%s' % (run_dir, i)
objecter_log = '%s/objecter.%s.log' % (run_dir, i)
p = common.pdsh(settings.getnodes('clients'), '%s -c %s -p rados-bench-`hostname -s`-%s %s bench %s %s %s --no-cleanup 2> %s > %s' % (self.cmd_path_full, self.tmp_conf, i, op_size_str, self.time, mode, concurrent_ops_str, objecter_log, out_file))
ps.append(p)
for p in ps:
p.wait()
monitoring.stop(run_dir)
# If we were doing recovery, wait until it's done.
if 'recovery_test' in self.cluster.config:
self.cluster.wait_recovery_done()
# Finally, get the historic ops
self.cluster.dump_historic_ops(run_dir)
common.sync_files('%s/*' % run_dir, out_dir)
def initialize(self):
super(Getput, self).initialize()
# create the user and key
self.cluster.add_swift_user(self.user, self.subuser, self.key)
# Clean and Create the run directory
common.clean_remote_dir(self.run_dir)
common.make_remote_dir(self.run_dir)
logger.info('Running scrub monitoring.')
monitoring.start("%s/scrub_monitoring" % self.run_dir)
self.cluster.check_scrub()
monitoring.stop()
logger.info('Pausing for 60s for idle monitoring.')
monitoring.start("%s/idle_monitoring" % self.run_dir)
time.sleep(60)
monitoring.stop()
common.sync_files('%s/*' % self.run_dir, self.out_dir)
return True
def run(self):
super(LibrbdFio, self).run()
# We'll always drop caches for rados bench
self.dropcaches()
# dump the cluster config
self.cluster.dump_config(self.run_dir)
monitoring.start(self.run_dir)
time.sleep(5)
# Run the backfill testing thread if requested
if 'recovery_test' in self.cluster.config:
recovery_callback = self.recovery_callback
self.cluster.create_recovery_test(self.run_dir, recovery_callback)
logger.info('Running rbd fio %s test.', self.mode)
ps = []
for i in range(self.volumes_per_client):
fio_cmd = self.mkfiocmd(i)
p = common.pdsh(settings.getnodes('clients'), fio_cmd)
ps.append(p)
for p in ps:
p.wait()
# If we were doing recovery, wait until it's done.
if 'recovery_test' in self.cluster.config:
self.cluster.wait_recovery_done()
monitoring.stop(self.run_dir)
# Finally, get the historic ops
self.cluster.dump_historic_ops(self.run_dir)
common.sync_files('%s/*' % self.run_dir, self.out_dir)
self.analyze(self.out_dir)
def initialize(self):
super(LibrbdFio, self).initialize()
# Clean and Create the run directory
common.clean_remote_dir(self.run_dir)
common.make_remote_dir(self.run_dir)
logger.info('Running scrub monitoring.')
monitoring.start("%s/scrub_monitoring" % self.run_dir)
self.cluster.check_scrub()
monitoring.stop()
logger.info('Pausing for 60s for idle monitoring.')
monitoring.start("%s/idle_monitoring" % self.run_dir)
time.sleep(60)
monitoring.stop()
common.sync_files('%s/*' % self.run_dir, self.out_dir)
self.mkimages()
# populate the fio files
ps = []
logger.info('Attempting to populating fio files...')
if (self.use_existing_volumes == False):
for volnum in xrange(self.volumes_per_client):
rbd_name = 'cbt-librbdfio-`%s`-%d' % (common.get_fqdn_cmd(),
volnum)
pre_cmd = 'sudo %s --ioengine=rbd --clientname=admin --pool=%s --rbdname=%s --invalidate=0 --rw=write --numjobs=%s --bs=4M --size %dM %s > /dev/null' % (
self.cmd_path, self.pool_name, rbd_name, self.numjobs,
self.vol_size, self.names)
p = common.pdsh(settings.getnodes('clients'), pre_cmd)
ps.append(p)
for p in ps:
p.wait()
return True
def initialize(self):
# safety check to make sure we don't blow away an existing cluster!
if self.use_existing:
raise RuntimeError('initialize was called on an existing cluster! Avoiding touching anything.')
super(Ceph, self).initialize()
# unmount any kernel rbd volumes
self.rbd_unmount()
# shutdown any old processes
self.shutdown()
# Cleanup old junk and create new junk
self.cleanup()
common.mkdir_p(self.tmp_dir)
common.pdsh(settings.getnodes('head', 'clients', 'mons', 'osds', 'rgws', 'mds'), 'mkdir -p -m0755 -- %s' % self.tmp_dir).communicate()
common.pdsh(settings.getnodes('clients', 'mons', 'osds', 'rgws', 'mds'), 'mkdir -p -m0755 -- %s' % self.pid_dir).communicate()
common.pdsh(settings.getnodes('clients', 'mons', 'osds', 'rgws', 'mds'), 'mkdir -p -m0755 -- %s' % self.log_dir).communicate()
common.pdsh(settings.getnodes('clients', 'mons', 'osds', 'rgws', 'mds'), 'mkdir -p -m0755 -- %s' % self.monitoring_dir).communicate()
common.pdsh(settings.getnodes('clients', 'mons', 'osds', 'rgws', 'mds'), 'mkdir -p -m0755 -- %s' % self.core_dir).communicate()
self.distribute_conf()
# Set the core directory
common.pdsh(settings.getnodes('clients', 'mons', 'osds', 'rgws', 'mds'), 'echo "%s/core.%%e.%%p.%%h.%%t" | sudo tee /proc/sys/kernel/core_pattern' % self.tmp_dir).communicate()
# Create the filesystems
self.setup_fs()
# Build the cluster
monitoring.start('%s/creation' % self.monitoring_dir)
self.make_mons()
self.make_osds()
self.start_rgw()
monitoring.stop()
# Check Health
monitoring.start('%s/initial_health_check' % self.monitoring_dir)
self.check_health()
monitoring.stop()
# Disable scrub and wait for any scrubbing to complete
self.disable_scrub()
self.check_scrub()
# Make the crush and erasure profiles
self.make_profiles()
# Peform Idle Monitoring
if self.idle_duration > 0:
monitoring.start("%s/idle_monitoring" % self.monitoring_dir)
time.sleep(self.idle_duration)
monitoring.stop()
return True
def initialize(self):
super(Ceph, self).initialize()
# unmount any kernel rbd volumes
self.rbd_unmount()
# shutdown any old processes
self.shutdown()
# Cleanup old junk and create new junk
self.cleanup()
common.mkdir_p(self.tmp_dir)
common.pdsh(settings.getnodes('head', 'clients', 'mons', 'osds', 'rgws', 'mds'), 'mkdir -p -m0755 -- %s' % self.tmp_dir).communicate()
common.pdsh(settings.getnodes('clients', 'mons', 'osds', 'rgws', 'mds'), 'mkdir -p -m0755 -- %s' % self.pid_dir).communicate()
common.pdsh(settings.getnodes('clients', 'mons', 'osds', 'rgws', 'mds'), 'mkdir -p -m0755 -- %s' % self.log_dir).communicate()
common.pdsh(settings.getnodes('clients', 'mons', 'osds', 'rgws', 'mds'), 'mkdir -p -m0755 -- %s' % self.monitoring_dir).communicate()
common.pdsh(settings.getnodes('clients', 'mons', 'osds', 'rgws', 'mds'), 'mkdir -p -m0755 -- %s' % self.core_dir).communicate()
self.distribute_conf()
# Set the core directory
common.pdsh(settings.getnodes('clients', 'mons', 'osds', 'rgws', 'mds'), 'echo "%s/core.%%e.%%p.%%h.%%t" | sudo tee /proc/sys/kernel/core_pattern' % self.tmp_dir).communicate()
# Create the filesystems
self.setup_fs()
# Build the cluster
monitoring.start('%s/creation' % self.monitoring_dir)
self.make_mons()
self.make_osds()
monitoring.stop()
# Check Health
monitoring.start('%s/initial_health_check' % self.monitoring_dir)
self.check_health()
monitoring.stop()
# Wait for initial scrubbing to complete (This should only matter on pre-dumpling clusters)
self.check_scrub()
# Make the crush and erasure profiles
self.make_profiles()
# Peform Idle Monitoring
if self.idle_duration > 0:
monitoring.start("%s/idle_monitoring" % self.monitoring_dir)
time.sleep(self.idle_duration)
monitoring.stop()
return True
def _run(self, mode, run_dir, out_dir, max_objects, runtime):
# We'll always drop caches for rados bench
self.dropcaches()
if self.concurrent_ops:
concurrent_ops_str = '--concurrent-ios %s' % self.concurrent_ops
rados_version = self.get_rados_version()
# Max Objects
max_objects_str = ''
if max_objects:
if rados_version < 10:
raise ValueError(
'max_objects not supported by rados_version < 10')
max_objects_str = '--max-objects %s' % max_objects
# Operation type
op_type = mode
if mode is 'prefill':
op_type = 'write'
if op_type in ['write'] or rados_version < 9:
op_size_str = '-b %s' % self.op_size
else:
op_size_str = ''
# Write to OMAP
write_omap_str = ''
if self.write_omap and rados_version < 9:
raise ValueError('write_omap not supported by rados_version < 9')
if self.write_omap and rados_version > 9:
write_omap_str = '--write-omap'
run_dir = os.path.join(self.run_dir, run_dir)
common.make_remote_dir(run_dir)
# dump the cluster config
self.cluster.dump_config(run_dir)
# Run the backfill testing thread if requested (but not for prefill)
if mode is not 'prefill' and 'recovery_test' in self.cluster.config:
recovery_callback = self.recovery_callback
self.cluster.create_recovery_test(run_dir, recovery_callback)
# Run rados bench
monitoring.start(run_dir)
logger.info('Running radosbench %s test.' % mode)
ps = []
for i in xrange(self.concurrent_procs):
out_file = '%s/output.%s' % (run_dir, i)
objecter_log = '%s/objecter.%s.log' % (run_dir, i)
# default behavior is to use a single storage pool
pool_name = self.pool
run_name = '--run-name %s`%s`-%s' % (self.object_set_id,
common.get_fqdn_cmd(), i)
if self.pool_per_proc: # support previous behavior of 1 storage pool per rados process
pool_name = 'rados-bench-``-%s' % (common.get_fqdn_cmd(), i)
run_name = ''
rados_bench_cmd = '%s -c %s -p %s bench %s %s %s %s %s %s %s --no-cleanup 2> %s > %s' % \
(self.cmd_path_full, self.tmp_conf, pool_name, op_size_str, runtime, op_type, concurrent_ops_str, max_objects_str, write_omap_str, run_name, objecter_log, out_file)
p = common.pdsh(settings.getnodes('clients'), rados_bench_cmd)
ps.append(p)
for p in ps:
p.wait()
monitoring.stop(run_dir)
# If we were doing recovery, wait until it's done (but not for prefill).
if mode is not 'prefill' and 'recovery_test' in self.cluster.config:
self.cluster.wait_recovery_done()
# Finally, get the historic ops
self.cluster.dump_historic_ops(run_dir)
out_dir = os.path.join(self.out_dir, out_dir)
common.sync_files('%s/*' % run_dir, out_dir)
self.analyze(out_dir)
def _run(self, mode, run_dir, out_dir):
# We'll always drop caches for rados bench
self.dropcaches()
if self.concurrent_ops:
concurrent_ops_str = '--concurrent-ios %s' % self.concurrent_ops
#determine rados version
rados_version_str = self.get_rados_version()
m = re.findall("version (\d+)", rados_version_str)
if not m:
m = re.findall("version v(\d+)", rados_version_str)
rados_version = int(m[0])
if mode in ['write'] or rados_version < 9:
op_size_str = '-b %s' % self.op_size
else:
op_size_str = ''
# Max Objects
if self.max_objects and rados_version < 9:
raise ValueError('max_objects not supported by rados_version < 9')
if self.max_objects and rados_version > 9:
max_objects_str = '--max-objects %s' % self.max_objects
# Write to OMAP
if self.write_omap and rados_version < 9:
raise ValueError('write_omap not supported by rados_version < 9')
if self.write_omap and rados_version > 9:
write_omap_str = '--write-omap'
common.make_remote_dir(run_dir)
# dump the cluster config
self.cluster.dump_config(run_dir)
# Run the backfill testing thread if requested
if 'recovery_test' in self.cluster.config:
recovery_callback = self.recovery_callback
self.cluster.create_recovery_test(run_dir, recovery_callback)
# Run rados bench
monitoring.start(run_dir)
logger.info('Running radosbench %s test.' % mode)
ps = []
for i in xrange(self.concurrent_procs):
out_file = '%s/output.%s' % (run_dir, i)
objecter_log = '%s/objecter.%s.log' % (run_dir, i)
# default behavior is to use a single storage pool
pool_name = self.pool
run_name = '--run-name %s`hostname -s`-%s' % (self.object_set_id,
i)
if self.pool_per_proc: # support previous behavior of 1 storage pool per rados process
pool_name = 'rados-bench-`hostname -s`-%s' % i
run_name = ''
rados_bench_cmd = '%s -c %s -p %s bench %s %s %s %s %s %s %s --no-cleanup 2> %s > %s' % \
(self.cmd_path_full, self.tmp_conf, pool_name, op_size_str, self.time, mode, concurrent_ops_str, max_objects_str, write_omap_str, run_name, objecter_log, out_file)
p = common.pdsh(settings.getnodes('clients'), rados_bench_cmd)
ps.append(p)
for p in ps:
p.wait()
monitoring.stop(run_dir)
# If we were doing recovery, wait until it's done.
if 'recovery_test' in self.cluster.config:
self.cluster.wait_recovery_done()
# Finally, get the historic ops
self.cluster.dump_historic_ops(run_dir)
common.sync_files('%s/*' % run_dir, out_dir)
def mkpool(self):
monitoring.start("%s/pool_monitoring" % self.run_dir)
self.cluster.rmpool('ceph_test_rados', self.pool_profile)
self.cluster.mkpool('ceph_test_rados', self.pool_profile,
'ceph_test_rados')
monitoring.stop()