ceph/tasks/recovery_bench.py

209 lines
5.5 KiB
Python
Raw Normal View History

"""
Recovery system benchmarking
"""
from cStringIO import StringIO
import contextlib
import gevent
import json
import logging
import random
import time
import ceph_manager
from teuthology import misc as teuthology
log = logging.getLogger(__name__)
@contextlib.contextmanager
def task(ctx, config):
"""
Benchmark the recovery system.
Generates objects with smalliobench, runs it normally to get a
baseline performance measurement, then marks an OSD out and reruns
to measure performance during recovery.
The config should be as follows:
recovery_bench:
duration: <seconds for each measurement run>
num_objects: <number of objects>
io_size: <io size in bytes>
example:
tasks:
- ceph:
- recovery_bench:
duration: 60
num_objects: 500
io_size: 4096
"""
if config is None:
config = {}
assert isinstance(config, dict), \
'recovery_bench task only accepts a dict for configuration'
log.info('Beginning recovery bench...')
first_mon = teuthology.get_first_mon(ctx, config)
(mon,) = ctx.cluster.only(first_mon).remotes.iterkeys()
manager = ceph_manager.CephManager(
mon,
ctx=ctx,
logger=log.getChild('ceph_manager'),
)
num_osds = teuthology.num_instances_of_type(ctx.cluster, 'osd')
while len(manager.get_osd_status()['up']) < num_osds:
time.sleep(10)
bench_proc = RecoveryBencher(
manager,
config,
)
try:
yield
finally:
log.info('joining recovery bencher')
bench_proc.do_join()
class RecoveryBencher:
"""
RecoveryBencher
"""
def __init__(self, manager, config):
self.ceph_manager = manager
self.ceph_manager.wait_for_clean()
osd_status = self.ceph_manager.get_osd_status()
self.osds = osd_status['up']
self.config = config
if self.config is None:
self.config = dict()
else:
def tmp(x):
"""
Local wrapper to print value.
"""
print x
self.log = tmp
log.info("spawning thread")
self.thread = gevent.spawn(self.do_bench)
def do_join(self):
"""
Join the recovery bencher. This is called after the main
task exits.
"""
self.thread.get()
def do_bench(self):
"""
Do the benchmarking.
"""
duration = self.config.get("duration", 60)
num_objects = self.config.get("num_objects", 500)
io_size = self.config.get("io_size", 4096)
osd = str(random.choice(self.osds))
(osd_remote,) = self.ceph_manager.ctx.cluster.only('osd.%s' % osd).remotes.iterkeys()
testdir = teuthology.get_testdir(self.ceph_manager.ctx)
# create the objects
osd_remote.run(
args=[
'adjust-ulimits',
'ceph-coverage',
'{tdir}/archive/coverage'.format(tdir=testdir),
'smalliobench'.format(tdir=testdir),
'--use-prefix', 'recovery_bench',
'--init-only', '1',
'--num-objects', str(num_objects),
'--io-size', str(io_size),
],
wait=True,
)
# baseline bench
log.info('non-recovery (baseline)')
p = osd_remote.run(
args=[
'adjust-ulimits',
'ceph-coverage',
'{tdir}/archive/coverage'.format(tdir=testdir),
'smalliobench',
'--use-prefix', 'recovery_bench',
'--do-not-init', '1',
'--duration', str(duration),
'--io-size', str(io_size),
],
stdout=StringIO(),
stderr=StringIO(),
wait=True,
)
self.process_samples(p.stderr.getvalue())
self.ceph_manager.raw_cluster_cmd('osd', 'out', osd)
time.sleep(5)
# recovery bench
log.info('recovery active')
p = osd_remote.run(
args=[
'adjust-ulimits',
'ceph-coverage',
'{tdir}/archive/coverage'.format(tdir=testdir),
'smalliobench',
'--use-prefix', 'recovery_bench',
'--do-not-init', '1',
'--duration', str(duration),
'--io-size', str(io_size),
],
stdout=StringIO(),
stderr=StringIO(),
wait=True,
)
self.process_samples(p.stderr.getvalue())
self.ceph_manager.raw_cluster_cmd('osd', 'in', osd)
def process_samples(self, input):
"""
Extract samples from the input and process the results
:param input: input lines in JSON format
"""
lat = {}
for line in input.split('\n'):
try:
sample = json.loads(line)
samples = lat.setdefault(sample['type'], [])
samples.append(float(sample['latency']))
except Exception:
pass
for type in lat:
samples = lat[type]
samples.sort()
num = len(samples)
# median
if num & 1 == 1: # odd number of samples
median = samples[num / 2]
else:
median = (samples[num / 2] + samples[num / 2 - 1]) / 2
# 99%
ninety_nine = samples[int(num * 0.99)]
log.info("%s: median %f, 99%% %f" % (type, median, ninety_nine))