run.py 7.77 KB
Newer Older
Antoine Kaufmann's avatar
Antoine Kaufmann committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
# Copyright 2021 Max Planck Institute for Software Systems, and
# National University of Singapore
#
# Permission is hereby granted, free of charge, to any person obtaining
# a copy of this software and associated documentation files (the
# "Software"), to deal in the Software without restriction, including
# without limitation the rights to use, copy, modify, merge, publish,
# distribute, sublicense, and/or sell copies of the Software, and to
# permit persons to whom the Software is furnished to do so, subject to
# the following conditions:
#
# The above copyright notice and this permission notice shall be
# included in all copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
# EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
# MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
# IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
# CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
# TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
# SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

23
24
25
26
import argparse
import sys
import os
import importlib
Hejing Li's avatar
Hejing Li committed
27
import importlib.util
28
import json
29
import pickle
30
import fnmatch
31
import simbricks.exectools as exectools
32
33
import simbricks.experiments as exp
import simbricks.runtime as runtime
34
35
36
37
38

def mkdir_if_not_exists(path):
    if not os.path.exists(path):
        os.mkdir(path)

39

40
41
42
parser = argparse.ArgumentParser()
parser.add_argument('experiments', metavar='EXP', type=str, nargs='+',
        help='An experiment file to run')
43
44
parser.add_argument('--filter', metavar='PATTERN', type=str, nargs='+',
        help='Pattern to match experiment names against')
45
46
47
parser.add_argument('--pickled', action='store_const', const=True,
        default=False,
        help='Read exp files as pickled runs instead of exp.py files')
48
parser.add_argument('--runs', metavar='N', type=int, default=1,
49
        help='Number of repetition for each experiment')
50
51
parser.add_argument('--firstrun', metavar='N', type=int, default=1,
        help='ID for first run')
52
53
parser.add_argument('--force', action='store_const', const=True, default=False,
        help='Run experiments even if output already exists')
54
55
56
parser.add_argument('--verbose', action='store_const', const=True,
        default=False,
        help='Verbose output')
57
58
parser.add_argument('--pcap', action='store_const', const=True, default=False,
        help='Dump pcap file (if supported by simulator)')
59
60
61

g_env = parser.add_argument_group('Environment')
g_env.add_argument('--repo', metavar='DIR', type=str,
62
        default='..', help='Repo directory')
63
g_env.add_argument('--workdir', metavar='DIR', type=str,
64
        default='./out/', help='Work directory base')
65
g_env.add_argument('--outdir', metavar='DIR',  type=str,
66
        default='./out/', help='Output directory base')
67
68
g_env.add_argument('--cpdir', metavar='DIR',  type=str,
        default='./out/', help='Checkpoint directory base')
69
70
g_env.add_argument('--hosts', metavar='JSON_FILE', type=str,
        default=None, help='List of hosts to use (json)')
71

72
73
74
75
76
77
78
79
80
81
g_par = parser.add_argument_group('Parallel Runtime')
g_par.add_argument('--parallel', dest='runtime', action='store_const',
        const='parallel', default='sequential',
        help='Use parallel instead of sequential runtime')
g_par.add_argument('--cores', metavar='N', type=int,
        default=len(os.sched_getaffinity(0)),
        help='Number of cores to use for parallel runs')
g_par.add_argument('--mem', metavar='N', type=int, default=None,
        help='Memory limit for parallel runs (in MB)')

82
83
84
85
86
87
88
g_slurm = parser.add_argument_group('Slurm Runtime')
g_slurm.add_argument('--slurm', dest='runtime', action='store_const',
        const='slurm', default='sequential',
        help='Use slurm instead of sequential runtime')
g_slurm.add_argument('--slurmdir', metavar='DIR',  type=str,
        default='./slurm/', help='Slurm communication directory')

89
90
91
92
g_dist = parser.add_argument_group('Distributed Runtime')
g_par.add_argument('--dist', dest='runtime', action='store_const',
        const='dist', default='sequential',
        help='Use sequential distributed runtime instead of local')
93
94
95
g_par.add_argument('--auto-dist', action='store_const', const=True,
        default=False,
        help='Automatically distribute non-distributed experiments')
96
args = parser.parse_args()
97

98
99
100
101
102
103
104
105
106

def load_executors(path):
    """ Load hosts list from json file and return list of executors. """
    with open(path, 'r') as f:
        hosts = json.load(f)

        exs = []
        for h in hosts:
            if h['type'] == 'local':
107
                ex = exectools.LocalExecutor()
108
            elif h['type'] == 'remote':
109
                ex = exectools.RemoteExecutor(h['host'], h['workdir'])
110
111
            else:
                raise RuntimeError('invalid host type "' + h['type'] + '"')
112
113
            ex.ip = h['ip']
            exs.append(ex)
114
115
116
117
118
119
120
121
122
123
124
125
    return exs

if args.hosts is None:
    executors = [exectools.LocalExecutor()]
else:
    executors = load_executors(args.hosts)

def warn_multi_exec():
    if len(executors) > 1:
        print('Warning: multiple hosts specified, only using first one for now',
                file=sys.stderr)

126
# initialize runtime
127
if args.runtime == 'parallel':
128
    warn_multi_exec()
129
    rt = runtime.LocalParallelRuntime(cores=args.cores, mem=args.mem,
130
            verbose=args.verbose, exec=executors[0])
131
132
elif args.runtime == 'slurm':
    rt = runtime.SlurmRuntime(args.slurmdir, args, verbose=args.verbose)
133
134
elif args.runtime == 'dist':
    rt = runtime.DistributedSimpleRuntime(executors, verbose=args.verbose)
135
else:
136
137
    warn_multi_exec()
    rt = runtime.LocalSimpleRuntime(verbose=args.verbose, exec=executors[0])
138

139
def add_exp(e, run, prereq, create_cp, restore_cp, no_simbricks):
140
    outpath = '%s/%s-%d.json' % (args.outdir, e.name, run)
141
    if os.path.exists(outpath) and not args.force:
142
143
144
145
        print('skip %s run %d' % (e.name, run))
        return None

    workdir = '%s/%s/%d' % (args.workdir, e.name, run)
146
    cpdir = '%s/%s/%d' % (args.cpdir, e.name, 0)
147

148
    env = exp.ExpEnv(args.repo, workdir, cpdir)
149
150
    env.create_cp = create_cp
    env.restore_cp = restore_cp
151
    env.no_simbricks=no_simbricks
152
153
154
    env.pcap_file = ''
    if args.pcap:
        env.pcap_file = workdir+'/pcap'
155
156
157
158
159

    run = runtime.Run(e, run, env, outpath, prereq)
    rt.add_run(run)
    return run

160
161
162
163
164
165
166
167
168
169
170
# load experiments
if not args.pickled:
    # default: load python modules with experiments
    experiments = []
    for path in args.experiments:
        modname, _ = os.path.splitext(os.path.basename(path))

        spec = importlib.util.spec_from_file_location(modname, path)
        mod = importlib.util.module_from_spec(spec)
        spec.loader.exec_module(mod)
        experiments += mod.experiments
171

172
    for e in experiments:
173
174
        if args.auto_dist and not isinstance(e, exp.DistributedExperiment):
            e = runtime.auto_dist(e, executors)
175
        # apply filter if any specified
176
        if (args.filter) and (len(args.filter) > 0):
177
178
179
180
181
182
183
184
            match = False
            for f in args.filter:
                if fnmatch.fnmatch(e.name, f):
                    match = True
                    break
            if not match:
                continue

185
        # if this is an experiment with a checkpoint we might have to create it
186
187
188
189
        if e.no_simbricks:
                no_simbricks = True
        else:
                no_simbricks = False
190
        if e.checkpoint:
191
            prereq = add_exp(e, 0, None, True, False, no_simbricks)
192
193
        else:
            prereq = None
194

195
        for run in range(args.firstrun, args.firstrun + args.runs):
196
            add_exp(e, run, prereq, False, e.checkpoint, no_simbricks)
197
198
199
200
201
else:
    # otherwise load pickled run object
    for path in args.experiments:
        with open(path, 'rb') as f:
            rt.add_run(pickle.load(f))
202

203
rt.start()