-
Notifications
You must be signed in to change notification settings - Fork 0
/
run_spark_benchmark.py
171 lines (149 loc) · 5.76 KB
/
run_spark_benchmark.py
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
import argparse
import benchmark
import logging
import subprocess
from typing import List
logging.getLogger(__name__).setLevel(logging.DEBUG)
class SparkBenchmark(benchmark.BaseBenchmark):
def __init__(self,
model_path: str,
data_path: str,
jar_path: str,
sparknlp_jar: str = None,
classname: str = None,
batch_sizes: List[int] = None,
input_lengths: List[int] = None,
seq_lengths: List[int] = None,
n_iter: int = 1,
input_cols: List[str] = None,
memcpu: bool = False,
name: str = "spark_benchmark"):
super().__init__(batch_sizes=batch_sizes,
input_lengths=input_lengths,
seq_lengths=seq_lengths,
n_iter=n_iter,
input_cols=input_cols,
model_path=model_path,
memcpu=memcpu,
name=name)
self.data_path = data_path
self.jar_path = jar_path
self.sparknlp_jar = sparknlp_jar
self.classname = classname
def run_iter(self, batch_size: int, input_length: int, output_length: int):
cmd = [
"spark-submit", "--executor-memory", "15G", "--driver-memory",
"12G", "--jars", self.sparknlp_jar, "--class", self.classname,
self.jar_path, self.model_path, self.data_path,
str(batch_size),
str(output_length), ",".join(self.input_cols)
]
result = subprocess.run(cmd)
if result.returncode != 0:
raise ValueError("Process exited wih non-zero return code...")
def measure_resource_usage(self, batch_size, input_length, output_length):
cmd = [
"spark-submit", "--jars", self.sparknlp_jar, "--class",
self.classname, self.jar_path, self.model_path, self.data_path,
str(batch_size),
str(output_length), ",".join(self.input_cols)
]
proc = subprocess.Popen(cmd)
res_monitor = benchmark.ProcMonitor(proc.pid)
res_monitor.start()
proc.wait()
res_monitor.stop()
res_monitor.join()
result = dict()
result['cpu_percent'] = res_monitor.cpu_percent
result['mem_usage'] = res_monitor.mem_kb
result['peak_memory'] = max(result['mem_usage']) / 1024
return result
def parse_config(args) -> dict:
assert args.jar_path is not None, "Missing benchmark app jar..."
config = dict()
config['jar_path'] = args.jar_path
config['n_iter'] = args.n_iter
config['input_cols'] = [s.strip() for s in args.input_cols.split(",")]
config['memcpu'] = args.resource_usage
config['batch_sizes'] = [
int(s.strip()) for s in args.batch_sizes.split(",")
] if args.batch_sizes else [1]
config['input_lengths'] = [
int(s.strip()) for s in args.input_lengths.split(",")
] if args.input_lengths else [16]
config['seq_lengths'] = [
int(s.strip()) for s in args.output_lengths.split(",")
] if args.output_lengths else [16]
return config
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('jar_path',
type=str,
help='Path to the compiled jar file.')
parser.add_argument('classname',
type=str,
help='Benchmark script main class name.')
parser.add_argument('sparknlp_jar',
type=str,
help='Path to the spark nlp jar file.')
parser.add_argument(
'--batch_sizes',
type=str,
help=
'Batch sizes to benchmark (pass multiple values as a comma-separated list). Default [4].'
)
parser.add_argument(
'--input_lengths',
type=str,
help=
'Input lengths to benchmark (pass multiple values as a comma-separated list). Default [16].'
)
parser.add_argument(
'--output_lengths',
type=str,
help=
'Output sequence lengths to benchmark (pass multiple values as a comma-separated list). Default [16].'
)
parser.add_argument(
'--model_path',
type=str,
help=
'Path to the model to import for benchmarking custom pre-trained models.'
)
parser.add_argument('--conll',
type=str,
help='Path to the CONLL formatted data file.')
parser.add_argument('--input_cols',
type=str,
help='Input columns to use for benchmarking.',
default='document')
parser.add_argument('--resource_usage',
type=bool,
help='Measure memory and cpu usage.',
default=False)
parser.add_argument("--n_iter",
type=int,
help="Number of iterations of each case.",
default=1)
args = parser.parse_args()
assert args.sparknlp_jar is not None, "Missing Spark NLP jar..."
benchmark_conf = parse_config(args)
if "model_path" in args:
model_path = args.model_path
else:
raise ValueError("Missing model path...")
if "conll" in args:
benchmark_conf["data_path"] = args.conll
else:
raise ValueError("Missing data path...")
if "classname" in args:
benchmark_conf["classname"] = args.classname
else:
raise ValueError("Missing classname...")
bm = SparkBenchmark(model_path=model_path,
sparknlp_jar=args.sparknlp_jar,
**benchmark_conf)
bm.run()
bm.print_results()
bm.save_results()