experiments.py 6.98 KB
Newer Older
1
2
3
4
5
6
import os
import asyncio
import modes.exectools as exectools
import shlex
import time
import json
7
import traceback
8
9
10
11
12
13
14
15
16
17
18

class Experiment(object):
    name = None
    timeout = None
    checkpoint = False

    def __init__(self, name):
        self.name = name
        self.hosts = []
        self.nics = []
        self.networks = []
19
        self.metadata = {}
20
21

    def add_host(self, sim):
22
23
24
        for h in self.hosts:
            if h.name == sim.name:
                raise Exception('Duplicate host name')
25
26
27
        self.hosts.append(sim)

    def add_nic(self, sim):
28
29
30
        for n in self.nics:
            if n.name == sim.name:
                raise Exception('Duplicate nic name')
31
32
33
        self.nics.append(sim)

    def add_network(self, sim):
34
35
36
        for n in self.networks:
            if n.name == sim.name:
                raise Exception('Duplicate net name')
37
38
        self.networks.append(sim)

39
    async def prepare(self, env, verbose=False):
40
41
42
        # generate config tars
        for host in self.hosts:
            path = env.cfgtar_path(host)
43
44
            if verbose:
                print('preparing config tar:', path)
45
46
47
48
49
50
            host.node_config.make_tar(path)

        # prepare all simulators in parallel
        sims = []
        for sim in self.hosts + self.nics + self.networks:
            prep_cmds = [pc for pc in sim.prep_cmds(env)]
51
52
            sims.append(exectools.run_cmdlist('prepare_' + self.name, prep_cmds,
                verbose=verbose))
53
54
        await asyncio.wait(sims)

55
    async def run(self, env, verbose=False):
56
57
58
59
60
61
        running = []
        sockets = []
        out = ExpOutput(self)
        try:
            out.set_start()

62
63
            if verbose:
                print('%s: starting NICS' % self.name)
64
            for nic in self.nics:
65
66
                if verbose:
                    print('start NIC:', nic.run_cmd(env))
67
                sc = exectools.SimpleComponent(nic.full_name(),
Hejing Li's avatar
Hejing Li committed
68
                        shlex.split(nic.run_cmd(env)), verbose=verbose, canfail=True)
69
70
71
72
73
74
75
                await sc.start()
                running.append((nic, sc))

                sockets.append(env.nic_pci_path(nic))
                sockets.append(env.nic_eth_path(nic))
                sockets.append(env.nic_shm_path(nic))

76
77
78
            if verbose:
                print('%s: waiting for sockets' % self.name)

79
            for s in sockets:
80
                await exectools.await_file(s, verbose=verbose)
81
82
            await asyncio.sleep(0.5)

83
84
85

            # start networks
            for net in self.networks:
86
87
88
                if verbose:
                    print('start Net:', net.run_cmd(env))

89
                sc = exectools.SimpleComponent(net.full_name(),
Hejing Li's avatar
Hejing Li committed
90
                        shlex.split(net.run_cmd(env)), verbose=verbose, canfail=True)
91
92
93
94
95
96
                await sc.start()
                running.append((net, sc))

            # start hosts
            wait_hosts = []
            for host in self.hosts:
97
98
99
                if verbose:
                    print('start Host:', host.run_cmd(env))

100
                sc = exectools.SimpleComponent(host.full_name(),
Hejing Li's avatar
Hejing Li committed
101
                        shlex.split(host.run_cmd(env)), verbose=verbose, canfail=True)
102
103
104
105
106
107
                await sc.start()
                running.append((host,sc))

                if host.wait:
                    wait_hosts.append(sc)

108
109
110
                if host.sleep > 0:
                    await asyncio.sleep(host.sleep)

111
112
            if verbose:
                print('%s: waiting for hosts to terminate' % self.name)
113
114
115
116
117
            for sc in wait_hosts:
                await sc.wait()
            # wait for necessary hosts to terminate
        except:
            out.set_failed()
118
            traceback.print_exc()
119
120
121
122
123

        finally:
            out.set_end()

            # shut things back down
124
125
            if verbose:
                print('%s: cleaning up' % self.name)
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
            scs = []
            for _,sc in running:
                scs.append(sc.int_term_kill())
            await asyncio.wait(scs)

            for _,sc in running:
                await sc.wait()

            for sock in sockets:
                os.remove(sock)

            for sim,sc in running:
                out.add_sim(sim, sc)
        return out



    def resreq_mem(self):
        mem = 0
        for h in self.hosts:
            mem += h.resreq_mem()
        for n in self.nics:
            mem += n.resreq_mem()
        for n in self.networks:
            mem += n.resreq_mem()
        return mem

    def resreq_cores(self):
        cores = 0
        for h in self.hosts:
            cores += h.resreq_cores()
        for n in self.nics:
            cores += n.resreq_cores()
        for n in self.networks:
            cores += n.resreq_cores()
        return cores

class ExpEnv(object):
164
    def __init__(self, repo_path, workdir, cpdir):
165
166
        self.repodir = os.path.abspath(repo_path)
        self.workdir = os.path.abspath(workdir)
167
        self.cpdir = os.path.abspath(cpdir)
168
169
        self.qemu_img_path = self.repodir + '/qemu/build/qemu-img'
        self.qemu_path = self.repodir + '/qemu/build/x86_64-softmmu/qemu-system-x86_64'
170
        self.qemu_kernel_path = self.repodir + '/images/bzImage'
171
172
173
        self.gem5_path = self.repodir + '/gem5/build/X86/gem5.opt'
        self.gem5_py_path = self.repodir + '/gem5/configs/cosim/cosim.py'
        self.gem5_kernel_path = self.repodir + '/images/vmlinux'
174
175

    def hdcopy_path(self, sim):
176
        return '%s/hdcopy.%s' % (self.workdir, sim.name)
177
178
179
180

    def hd_path(self, hd_name):
        return '%s/images/output-%s/%s' % (self.repodir, hd_name, hd_name)

181
182
183
    def hd_raw_path(self, hd_name):
        return '%s/images/output-%s/%s.raw' % (self.repodir, hd_name, hd_name)

184
    def cfgtar_path(self, sim):
185
        return '%s/cfg.%s.tar' % (self.workdir, sim.name)
186
187

    def nic_pci_path(self, sim):
188
        return '%s/nic.pci.%s' % (self.workdir, sim.name)
189
190

    def nic_eth_path(self, sim):
191
        return '%s/nic.eth.%s' % (self.workdir, sim.name)
192
193

    def nic_shm_path(self, sim):
194
        return '%s/nic.shm.%s' % (self.workdir, sim.name)
195

196
197
198
199
200
201
    def gem5_outdir(self, sim):
        return '%s/gem5-out.%s' % (self.workdir, sim.name)

    def gem5_cpdir(self, sim):
        return '%s/gem5-cp.%s' % (self.cpdir, sim.name)

202
203
204
class ExpOutput(object):
    def __init__(self, exp):
        self.exp_name = exp.name
205
        self.metadata = exp.metadata
206
207
208
209
210
211
212
213
214
        self.start_time = None
        self.end_time = None
        self.sims = {}
        self.success = True

    def set_start(self):
        self.start_time = time.time()

    def set_end(self):
215
        self.end_time = time.time()
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233

    def set_failed(self):
        self.success = False

    def add_sim(self, sim, comp):
        obj = {
            'class': sim.__class__.__name__,
            'cmd': comp.cmd_parts,
            'stdout': comp.stdout,
            'stderr': comp.stderr,
        }
        self.sims[sim.full_name()] = obj

    def dumps(self):
        return json.dumps(self.__dict__)



234
235
236
def run_exp_local(exp, env, verbose=False):
    asyncio.run(exp.prepare(env, verbose=verbose))
    return asyncio.run(exp.run(env, verbose=verbose))