autoserv revision cb6f1e2209ae6ea09c78317b114db6e51f6e255b 
1#!/usr/bin/python -u
2# Copyright 2007-2008 Martin J. Bligh <mbligh@google.com>, Google Inc.
3# Released under the GPL v2
4
5"""
6Run a control file through the server side engine
7"""
8
9import sys, os, re, traceback, signal, time, logging, getpass
10
11import common
12
13from autotest_lib.client.common_lib.global_config import global_config
14require_atfork = global_config.get_config_value(
15        'AUTOSERV', 'require_atfork_module', type=bool, default=True)
16
17try:
18    import atfork
19    atfork.monkeypatch_os_fork_functions()
20    import atfork.stdlib_fixer
21    # Fix the Python standard library for threading+fork safety with its
22    # internal locks.  http://code.google.com/p/python-atfork/
23    import warnings
24    warnings.filterwarnings('ignore', 'logging module already imported')
25    atfork.stdlib_fixer.fix_logging_module()
26except ImportError, e:
27    from autotest_lib.client.common_lib import global_config
28    if global_config.global_config.get_config_value(
29            'AUTOSERV', 'require_atfork_module', type=bool, default=False):
30        print >>sys.stderr, 'Please run utils/build_externals.py'
31        print e
32        sys.exit(1)
33
34from autotest_lib.server import server_logging_config
35from autotest_lib.server import server_job, utils, autoserv_parser, autotest
36from autotest_lib.client.common_lib import pidfile, logging_manager
37
38def log_alarm(signum, frame):
39    logging.error("Received SIGALARM. Ignoring and continuing on.")
40    sys.exit(1)
41
42def run_autoserv(pid_file_manager, results, parser):
43    # send stdin to /dev/null
44    dev_null = os.open(os.devnull, os.O_RDONLY)
45    os.dup2(dev_null, sys.stdin.fileno())
46    os.close(dev_null)
47
48    # Create separate process group
49    os.setpgrp()
50
51    # Implement SIGTERM handler
52    def handle_sigterm(signum, frame):
53        if pid_file_manager:
54            pid_file_manager.close_file(1, signal.SIGTERM)
55        os.killpg(os.getpgrp(), signal.SIGKILL)
56
57    # Set signal handler
58    signal.signal(signal.SIGTERM, handle_sigterm)
59
60    # Ignore SIGTTOU's generated by output from forked children.
61    signal.signal(signal.SIGTTOU, signal.SIG_IGN)
62
63    # If we received a SIGALARM, let's be loud about it.
64    signal.signal(signal.SIGALRM, log_alarm)
65
66    # Server side tests that call shell scripts often depend on $USER being set
67    # but depending on how you launch your autotest scheduler it may not be set.
68    os.environ['USER'] = getpass.getuser()
69
70    if parser.options.machines:
71        machines = parser.options.machines.replace(',', ' ').strip().split()
72    else:
73        machines = []
74    machines_file = parser.options.machines_file
75    label = parser.options.label
76    group_name = parser.options.group_name
77    user = parser.options.user
78    client = parser.options.client
79    server = parser.options.server
80    install_before = parser.options.install_before
81    install_after = parser.options.install_after
82    verify = parser.options.verify
83    repair = parser.options.repair
84    cleanup = parser.options.cleanup
85    provision = parser.options.provision
86    no_tee = parser.options.no_tee
87    parse_job = parser.options.parse_job
88    execution_tag = parser.options.execution_tag
89    if not execution_tag:
90        execution_tag = parse_job
91    host_protection = parser.options.host_protection
92    ssh_user = parser.options.ssh_user
93    ssh_port = parser.options.ssh_port
94    ssh_pass = parser.options.ssh_pass
95    collect_crashinfo = parser.options.collect_crashinfo
96    control_filename = parser.options.control_filename
97    test_retry = parser.options.test_retry
98    verify_job_repo_url = parser.options.verify_job_repo_url
99
100    # can't be both a client and a server side test
101    if client and server:
102        parser.parser.error("Can not specify a test as both server and client!")
103
104    if provision and client:
105        parser.parser.error("Cannot specify provisioning and client!")
106
107    is_special_task = (verify or repair or cleanup or collect_crashinfo or
108                       provision)
109    if len(parser.args) < 1 and not is_special_task:
110        parser.parser.error("Missing argument: control file")
111
112    # We have a control file unless it's just a verify/repair/cleanup job
113    if len(parser.args) > 0:
114        control = parser.args[0]
115    else:
116        control = None
117
118    if machines_file:
119        machines = []
120        for m in open(machines_file, 'r').readlines():
121            # remove comments, spaces
122            m = re.sub('#.*', '', m).strip()
123            if m:
124                machines.append(m)
125        print "Read list of machines from file: %s" % machines_file
126        print ','.join(machines)
127
128    if machines:
129        for machine in machines:
130            if not machine or re.search('\s', machine):
131                parser.parser.error("Invalid machine: %s" % str(machine))
132        machines = list(set(machines))
133        machines.sort()
134
135    if group_name and len(machines) < 2:
136        parser.parser.error("-G %r may only be supplied with more than one machine."
137               % group_name)
138
139    kwargs = {'group_name': group_name, 'tag': execution_tag}
140    if control_filename:
141        kwargs['control_filename'] = control_filename
142    job = server_job.server_job(control, parser.args[1:], results, label,
143                                user, machines, client, parse_job,
144                                ssh_user, ssh_port, ssh_pass, test_retry,
145                                **kwargs)
146    job.logging.start_logging()
147    job.init_parser()
148
149    # perform checks
150    job.precheck()
151
152    # run the job
153    exit_code = 0
154    try:
155        try:
156            if repair:
157                job.repair(host_protection)
158            elif verify:
159                job.verify()
160            elif provision:
161                job.provision(provision)
162            else:
163                job.run(cleanup, install_before, install_after,
164                        verify_job_repo_url=verify_job_repo_url,
165                        only_collect_crashinfo=collect_crashinfo)
166        finally:
167            while job.hosts:
168                host = job.hosts.pop()
169                host.close()
170    except:
171        exit_code = 1
172        traceback.print_exc()
173
174    if pid_file_manager:
175        pid_file_manager.num_tests_failed = job.num_tests_failed
176        pid_file_manager.close_file(exit_code)
177    job.cleanup_parser()
178
179    sys.exit(exit_code)
180
181
182def main():
183    # grab the parser
184    parser = autoserv_parser.autoserv_parser
185    parser.parse_args()
186
187    if len(sys.argv) == 1:
188        parser.parser.print_help()
189        sys.exit(1)
190
191    if parser.options.no_logging:
192        results = None
193    else:
194        results = parser.options.results
195        if not results:
196            results = 'results.' + time.strftime('%Y-%m-%d-%H.%M.%S')
197        results  = os.path.abspath(results)
198        resultdir_exists = False
199        for filename in ('control.srv', 'status.log', '.autoserv_execute'):
200            if os.path.exists(os.path.join(results, filename)):
201                resultdir_exists = True
202        if not parser.options.use_existing_results and resultdir_exists:
203            error = "Error: results directory already exists: %s\n" % results
204            sys.stderr.write(error)
205            sys.exit(1)
206
207        # Now that we certified that there's no leftover results dir from
208        # previous jobs, lets create the result dir since the logging system
209        # needs to create the log file in there.
210        if not os.path.isdir(results):
211            os.makedirs(results)
212
213    logging_manager.configure_logging(
214            server_logging_config.ServerLoggingConfig(), results_dir=results,
215            use_console=not parser.options.no_tee,
216            verbose=parser.options.verbose,
217            no_console_prefix=parser.options.no_console_prefix)
218    if results:
219        logging.info("Results placed in %s" % results)
220
221        # wait until now to perform this check, so it get properly logged
222        if parser.options.use_existing_results and not resultdir_exists:
223            logging.error("No existing results directory found: %s", results)
224            sys.exit(1)
225
226
227    if parser.options.write_pidfile:
228        pid_file_manager = pidfile.PidFileManager(parser.options.pidfile_label,
229                                                  results)
230        pid_file_manager.open_file()
231    else:
232        pid_file_manager = None
233
234    autotest.BaseAutotest.set_install_in_tmpdir(
235        parser.options.install_in_tmpdir)
236
237    exit_code = 0
238    try:
239        try:
240            run_autoserv(pid_file_manager, results, parser)
241        except SystemExit, e:
242            exit_code = e.code
243        except:
244            traceback.print_exc()
245            # If we don't know what happened, we'll classify it as
246            # an 'abort' and return 1.
247            exit_code = 1
248    finally:
249        if pid_file_manager:
250            pid_file_manager.close_file(exit_code)
251    sys.exit(exit_code)
252
253
254if __name__ == '__main__':
255    main()
256