1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
|
"""
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))
|
