xref: /aosp_15_r20/external/autotest/server/control_segments/provision (revision 9c5db1993ded3edbeafc8092d69fe5de2ee02df7)
1# Copyright (c) 2013 The Chromium OS Authors. All rights reserved.
2# Use of this source code is governed by a BSD-style license that can be
3# found in the LICENSE file.
4
5import datetime
6import re
7
8from autotest_lib.client.bin import sysinfo
9from autotest_lib.client.cros import constants
10from autotest_lib.server import utils
11from autotest_lib.server.cros import provision
12
13try:
14    from autotest_lib.utils.frozen_chromite.lib import metrics
15except ImportError:
16    metrics = utils.metrics_mock
17
18
19LABEL_REGEX = r',.*:'
20_LABEL_UPDATE_DURATION_METRIC = metrics.SecondsDistribution(
21        'chromeos/autotest/provision/label_update_durations')
22
23# job_labels should be a string like "name:setting,name:setting"
24# However setting might also contain ',' therefore we need more advanced logic
25# than split.
26# non-provisionable labels are currently skipped, so they're safe to pass in.
27job_labels = locals().get('job_labels') or ','.join(args)
28labels_list = []
29while job_labels:
30    # Split based off of a comma followed by colon regex.
31    split = re.split(LABEL_REGEX, job_labels)
32    # First value found is a proper key value pair.
33    labels_list.append(split[0].strip())
34    # Remove this key value pair.
35    job_labels = job_labels[len(split[0]):]
36    # If a comma remains at the start of the remaining labels, remove it.
37    # This should happen on every loop except the last one.
38    if job_labels.startswith(','):
39        job_labels = job_labels.lstrip(',')
40
41
42def provision_machine(machine):
43    """
44    Run the appropriate provisioning tests to make the machine's labels match
45    those given in job_labels.
46    """
47    job.record('START', None, 'provision')
48    # Determine if we should initialize servo based on request type as
49    # we don't need servo in OS only provision.
50    need_servo = False
51    for label in labels_list:
52        if (label.startswith(provision.FW_RW_VERSION_PREFIX) or
53            label.startswith(provision.FW_RO_VERSION_PREFIX)):
54            need_servo = True
55    host = hosts.create_target_machine(machine, try_lab_servo=need_servo)
56    try:
57        job.sysinfo.add_logdir(
58                sysinfo.logdir(constants.AUTOUPDATE_PRESERVE_LOG))
59        provision.Provision.run_task_actions(job, host, labels_list)
60        host.verify()
61
62        # Let's update the labels on the host and track how long it takes.
63        # Don't fail while updating the labels, provision is flaky enough by
64        # itself.
65        label_update_success = True
66        start_time = datetime.datetime.now()
67        try:
68            host.labels.update_labels(host, keep_pool=True)
69        except Exception:
70            logging.exception('Exception while updating labels.')
71            label_update_success = False
72
73        end_time = datetime.datetime.now()
74        duration = (end_time - start_time).total_seconds()
75
76        fields = {'success': label_update_success,
77                  'board': host.get_board()}
78        _LABEL_UPDATE_DURATION_METRIC.add(duration, fields=fields)
79    except Exception:
80        logging.exception('Provision failed due to Exception.')
81        job.record('END FAIL', None, 'provision')
82        # Raising a blank exception is done here because any message we can
83        # give here would be less useful than whatever the failing test left as
84        # its own exception message.
85        #
86        # The gory details of how raising a blank exception accomplishes this
87        # is as follows:
88        #
89        # The scheduler only looks at the return code of autoserv to see if
90        # the special task failed.  Therefore we need python to exit because
91        # of an unhandled exception or because someone called sys.exit(1).
92        #
93        # We can't call sys.exit, since there's post-job-running logic (like
94        # cleanup) that we'd be skipping out on.  So therefore, we need to
95        # raise an exception.  However, if we raise an exception, this
96        # exception ends up triggering server_job to write an INFO line with
97        # job_abort_reason equal to str(e), which the tko parser then picks
98        # up as the reason field for the job when the status.log we generate is
99        # parsed as the job's results.
100        #
101        # So therefore, we raise a blank exception, which then generates an
102        # empty job_abort_reason which the tko parser ignores just inserts as
103        # a SERVER_JOB failure with no reason, which we then ignore at suite
104        # results reporting time.
105        raise Exception('')
106    else:
107        # If we finish successfully, nothing in autotest ever looks at the
108        # status.log, so it's purely for human consumption and tracability.
109        hostname = utils.get_hostname_from_machine(machine)
110        job.record('END GOOD', None, 'provision',
111                   '%s provisioned successfully' % hostname)
112
113
114job.parallel_simple(provision_machine, machines)
115
116# vim: set syntax=python :
117