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
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
|
import math
import os
import os.path
import re
import time
from numpy import mean, median, std
from multiprocessing import cpu_count
from utils.logging import log
from utils.misc import available_ram, run_cmd
class PgBench(object):
'a simple wrapper around pgbench, running TPC-B-like workload by default'
# TODO allow running custom scripts, not just the default
# read-write/read-only tests
# TODO allow running 'prepared' mode
def __init__(self, bin_path, dbname, runs=3, duration=60, csv=False,
results_dir=None):
'''
bin_path - path to PostgreSQL binaries (dropdb, createdb, psql
commands)
dbname - name of the database to use
duration - duration of each execution
runs - number of runs (for each client count)
out_dir - output directory
'''
self._bin = bin_path
self._csv = csv
self._dbname = dbname
self._duration = duration
self._outdir = results_dir
self._runs = runs
self._env = os.environ
self._env['PATH'] = ':'.join([bin_path, self._env['PATH']])
self._results = {}
@staticmethod
def _configure(cpu_count, ram_mbs):
'derive the configurations to benchmark from CPU count and RAM size'
config = []
# TODO allow overriding this from a global config
# scales: 10 (small), 50% of RAM, 200% of RAM
# for s in [10, ram_mbs/15/2, ram_mbs*2/15]:
for s in [10]:
config.append({'scale': int(math.ceil(s / 10) * 10),
'clients': [1, cpu_count, 2 * cpu_count]})
return config
def _init(self, scale):
"""
recreate the database (drop + create) and populate it with given scale
"""
# initialize results for this dataset scale
self._results['results'] = {
'init': None,
'runs': [],
'warmup': None,
}
log("recreating '%s' database" % (self._dbname,))
run_cmd(['dropdb', '--if-exists', self._dbname], env=self._env)
run_cmd(['createdb', self._dbname], env=self._env)
log("initializing pgbench '%s' with scale %s" % (self._dbname, scale))
r = run_cmd(['pgbench', '-i', '-s', str(scale), self._dbname],
env=self._env, cwd=self._outdir)
# remember the init duration
self._results['results']['init'] = r[2]
@staticmethod
def _parse_results(data):
'extract results (including parameters) from the pgbench output'
scale = -1
r = re.search('scaling factor: ([0-9]+)', data)
if r:
scale = r.group(1)
mode = -1
r = re.search('query mode: (.+)', data)
if r:
mode = r.group(1)
clients = -1
r = re.search('number of clients: ([0-9]+)', data)
if r:
clients = r.group(1)
threads = -1
r = re.search('number of threads: ([0-9]+)', data)
if r:
threads = r.group(1)
duration = -1
r = re.search('duration: ([0-9]+) s', data)
if r:
duration = r.group(1)
latency = -1
r = re.search('latency average: ([0-9\.]+) ms', data)
if r:
latency = r.group(1)
tps = -1
r = re.search('tps = ([0-9]+\.[0-9]+) \(excluding connections '
'establishing\)', data)
if r:
tps = r.group(1)
return {'scale': scale,
'mode': mode,
'clients': clients,
'threads': threads,
'duration': duration,
'latency': latency,
'tps': tps}
def check_config(self):
'check pgbench configuration (existence of binaries etc.)'
issues = []
if not os.path.isdir(self._bin):
issues.append("bin_dir='%s' does not exist" % (self._bin,))
elif not os.path.exists('%s/pgbench' % (self._bin,)):
issues.append("pgbench not found in bin_dir='%s'" % (self._bin,))
elif not os.path.exists('%s/createdb' % (self._bin,)):
issues.append("createdb not found in bin_dir='%s'" % (self._bin,))
elif not os.path.exists('%s/dropdb' % (self._bin,)):
issues.append("dropdb not found in bin_dir='%s'" % (self._bin,))
elif not os.path.exists('%s/psql' % (self._bin,)):
issues.append("psql not found in bin_dir='%s'" % (self._bin,))
if type(self._duration) is not int:
issues.append("duration (%s) needs to be an integer" %
self._duration)
elif not self._duration >= 1:
issues.append("duration (%s) needs to be >= 1" % (self._duration,))
if type(self._runs) is not int:
issues.append("runs (%s) needs to be an integer" % self._duration)
elif not self._runs >= 1:
issues.append("runs (%s) needs to be >= 1" % (self._runs,))
return issues
def _run(self, run, scale, duration, nclients=1, njobs=1, read_only=False,
aggregate=True, csv_queue=None):
'run pgbench on the database (either a warmup or actual benchmark run)'
# Create a separate directory for each pgbench run
if read_only:
rtag = "ro"
else:
rtag = "rw"
rdir = "%s/pgbench-%s-%d-%d-%s" % (self._outdir, rtag, scale, nclients,
str(run))
os.mkdir(rdir)
args = ['pgbench', '-c', str(nclients), '-j', str(njobs), '-T',
str(duration)]
# aggregate on per second resolution
if aggregate:
args.extend(['-l', '--aggregate-interval', '1'])
if read_only:
args.extend(['-S'])
args.extend([self._dbname])
# do an explicit checkpoint before each run
run_cmd(['psql', self._dbname, '-c', 'checkpoint'], env=self._env)
log("pgbench: clients=%d, jobs=%d, aggregate=%s, read-only=%s, "
"duration=%d" % (nclients, njobs, aggregate, read_only, duration))
start = time.time()
r = run_cmd(args, env=self._env, cwd=rdir)
end = time.time()
r = PgBench._parse_results(r[1])
r.update({'read-only': read_only})
r.update({'start': start, 'end': end})
if csv_queue is not None:
csv_queue.put([start, end, r['scale'], nclients, njobs, mode,
duration, latency, tps])
return r
def run_tests(self, csv_queue):
"""
execute the whole benchmark, including initialization, warmup and
benchmark runs
"""
# derive configuration for the CPU count / RAM size
configs = PgBench._configure(cpu_count(), available_ram())
results = {'ro': {}, 'rw': {}}
j = 0
for config in configs:
scale = config['scale']
if scale not in results['ro']:
results['ro'][scale] = {}
if scale not in results['rw']:
results['rw'][scale] = {}
# init for the dataset scale and warmup
self._init(scale)
warmup = self._run('w%d' % j, scale, self._duration, cpu_count(),
cpu_count())
j += 1
# read-only & read-write
for ro in [True, False]:
if ro:
tag = 'ro'
else:
tag = 'rw'
for i in range(self._runs):
log("pgbench : %s run=%d" % (tag, i))
for clients in config['clients']:
if clients not in results[tag][scale]:
results[tag][scale][clients] = {}
results[tag][scale][clients]['results'] = []
r = self._run(i, scale, self._duration, clients,
clients, ro, True, csv_queue)
r.update({'run': i})
results[tag][scale][clients]['results'].append(r)
tps = []
for result in results[tag][scale][clients]['results']:
tps.append(float(result['tps']))
results[tag][scale][clients]['metric'] = mean(tps)
results[tag][scale][clients]['median'] = median(tps)
results[tag][scale][clients]['std'] = std(tps)
self._results['pgbench'] = results
return self._results
|
