You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

615 lines
24 KiB

#
# Copyright 2008 Google Inc. All Rights Reserved.
"""
The job module contains the objects and methods used to
manage jobs in Autotest.
The valid actions are:
list: lists job(s)
create: create a job
abort: abort job(s)
stat: detailed listing of job(s)
The common options are:
See topic_common.py for a High Level Design and Algorithm.
"""
# pylint: disable=missing-docstring
from __future__ import print_function
import getpass, re
from autotest_lib.cli import topic_common, action_common
from autotest_lib.client.common_lib import control_data
from autotest_lib.client.common_lib import priorities
class job(topic_common.atest):
"""Job class
atest job [create|clone|list|stat|abort] <options>"""
usage_action = '[create|clone|list|stat|abort]'
topic = msg_topic = 'job'
msg_items = '<job_ids>'
def _convert_status(self, results):
for result in results:
total = sum(result['status_counts'].values())
status = ['%s=%s(%.1f%%)' % (key, val, 100.0*float(val)/total)
for key, val in result['status_counts'].iteritems()]
status.sort()
result['status_counts'] = ', '.join(status)
def backward_compatibility(self, action, argv):
""" 'job create --clone' became 'job clone --id' """
if action == 'create':
for option in ['-l', '--clone']:
if option in argv:
argv[argv.index(option)] = '--id'
action = 'clone'
return action
class job_help(job):
"""Just here to get the atest logic working.
Usage is set by its parent"""
pass
class job_list_stat(action_common.atest_list, job):
def __init__(self):
super(job_list_stat, self).__init__()
self.topic_parse_info = topic_common.item_parse_info(
attribute_name='jobs',
use_leftover=True)
def __split_jobs_between_ids_names(self):
job_ids = []
job_names = []
# Sort between job IDs and names
for job_id in self.jobs:
if job_id.isdigit():
job_ids.append(job_id)
else:
job_names.append(job_id)
return (job_ids, job_names)
def execute_on_ids_and_names(self, op, filters={},
check_results={'id__in': 'id',
'name__in': 'id'},
tag_id='id__in', tag_name='name__in'):
if not self.jobs:
# Want everything
return super(job_list_stat, self).execute(op=op, filters=filters)
all_jobs = []
(job_ids, job_names) = self.__split_jobs_between_ids_names()
for items, tag in [(job_ids, tag_id),
(job_names, tag_name)]:
if items:
new_filters = filters.copy()
new_filters[tag] = items
jobs = super(job_list_stat,
self).execute(op=op,
filters=new_filters,
check_results=check_results)
all_jobs.extend(jobs)
return all_jobs
class job_list(job_list_stat):
"""atest job list [<jobs>] [--all] [--running] [--user <username>]"""
def __init__(self):
super(job_list, self).__init__()
self.parser.add_option('-a', '--all', help='List jobs for all '
'users.', action='store_true', default=False)
self.parser.add_option('-r', '--running', help='List only running '
'jobs', action='store_true')
self.parser.add_option('-u', '--user', help='List jobs for given '
'user', type='string')
def parse(self):
options, leftover = super(job_list, self).parse()
self.all = options.all
self.data['running'] = options.running
if options.user:
if options.all:
self.invalid_syntax('Only specify --all or --user, not both.')
else:
self.data['owner'] = options.user
elif not options.all and not self.jobs:
self.data['owner'] = getpass.getuser()
return options, leftover
def execute(self):
return self.execute_on_ids_and_names(op='get_jobs_summary',
filters=self.data)
def output(self, results):
keys = ['id', 'owner', 'name', 'status_counts']
if self.verbose:
keys.extend(['priority', 'control_type', 'created_on'])
self._convert_status(results)
super(job_list, self).output(results, keys)
class job_stat(job_list_stat):
"""atest job stat <job>"""
usage_action = 'stat'
def __init__(self):
super(job_stat, self).__init__()
self.parser.add_option('-f', '--control-file',
help='Display the control file',
action='store_true', default=False)
self.parser.add_option('-N', '--list-hosts',
help='Display only a list of hosts',
action='store_true')
self.parser.add_option('-s', '--list-hosts-status',
help='Display only the hosts in these statuses '
'for a job.', action='store')
def parse(self):
status_list = topic_common.item_parse_info(
attribute_name='status_list',
inline_option='list_hosts_status')
options, leftover = super(job_stat, self).parse([status_list],
req_items='jobs')
if not self.jobs:
self.invalid_syntax('Must specify at least one job.')
self.show_control_file = options.control_file
self.list_hosts = options.list_hosts
if self.list_hosts and self.status_list:
self.invalid_syntax('--list-hosts is implicit when using '
'--list-hosts-status.')
if len(self.jobs) > 1 and (self.list_hosts or self.status_list):
self.invalid_syntax('--list-hosts and --list-hosts-status should '
'only be used on a single job.')
return options, leftover
def _merge_results(self, summary, qes):
hosts_status = {}
for qe in qes:
if qe['host']:
job_id = qe['job']['id']
hostname = qe['host']['hostname']
hosts_status.setdefault(job_id,
{}).setdefault(qe['status'],
[]).append(hostname)
for job in summary:
job_id = job['id']
if hosts_status.has_key(job_id):
this_job = hosts_status[job_id]
job['hosts'] = ' '.join(' '.join(host) for host in
this_job.itervalues())
host_per_status = ['%s="%s"' %(status, ' '.join(host))
for status, host in this_job.iteritems()]
job['hosts_status'] = ', '.join(host_per_status)
if self.status_list:
statuses = set(s.lower() for s in self.status_list)
all_hosts = [s for s in host_per_status if s.split('=',
1)[0].lower() in statuses]
job['hosts_selected_status'] = '\n'.join(all_hosts)
else:
job['hosts_status'] = ''
if not job.get('hosts'):
self.generic_error('Job has unassigned meta-hosts, '
'try again shortly.')
return summary
def execute(self):
summary = self.execute_on_ids_and_names(op='get_jobs_summary')
# Get the real hostnames
qes = self.execute_on_ids_and_names(op='get_host_queue_entries',
check_results={},
tag_id='job__in',
tag_name='job__name__in')
self._convert_status(summary)
return self._merge_results(summary, qes)
def output(self, results):
if self.list_hosts:
keys = ['hosts']
elif self.status_list:
keys = ['hosts_selected_status']
elif not self.verbose:
keys = ['id', 'name', 'priority', 'status_counts', 'hosts_status']
else:
keys = ['id', 'name', 'priority', 'status_counts', 'hosts_status',
'owner', 'control_type', 'synch_count', 'created_on',
'run_verify', 'reboot_before', 'reboot_after',
'parse_failed_repair']
if self.show_control_file:
keys.append('control_file')
super(job_stat, self).output(results, keys)
class job_create_or_clone(action_common.atest_create, job):
"""Class containing the code common to the job create and clone actions"""
msg_items = 'job_name'
def __init__(self):
super(job_create_or_clone, self).__init__()
self.hosts = []
self.data_item_key = 'name'
self.parser.add_option('-p', '--priority',
help='Job priority (int)', type='int',
default=priorities.Priority.DEFAULT)
self.parser.add_option('-b', '--labels',
help='Comma separated list of labels '
'to get machine list from.', default='')
self.parser.add_option('-m', '--machine', help='List of machines to '
'run on')
self.parser.add_option('-M', '--mlist',
help='File listing machines to use',
type='string', metavar='MACHINE_FLIST')
self.parser.add_option('--one-time-hosts',
help='List of one time hosts')
self.parser.add_option('-e', '--email',
help='A comma seperated list of '
'email addresses to notify of job completion',
default='')
def _parse_hosts(self, args):
""" Parses the arguments to generate a list of hosts and meta_hosts
A host is a regular name, a meta_host is n*label or *label.
These can be mixed on the CLI, and separated by either commas or
spaces, e.g.: 5*Machine_Label host0 5*Machine_Label2,host2 """
hosts = []
meta_hosts = []
for arg in args:
for host in arg.split(','):
if re.match('^[0-9]+[*]', host):
num, host = host.split('*', 1)
meta_hosts += int(num) * [host]
elif re.match('^[*](\w*)', host):
meta_hosts += [re.match('^[*](\w*)', host).group(1)]
elif host != '' and host not in hosts:
# Real hostname and not a duplicate
hosts.append(host)
return (hosts, meta_hosts)
def parse(self, parse_info=[]):
host_info = topic_common.item_parse_info(attribute_name='hosts',
inline_option='machine',
filename_option='mlist')
job_info = topic_common.item_parse_info(attribute_name='jobname',
use_leftover=True)
oth_info = topic_common.item_parse_info(attribute_name='one_time_hosts',
inline_option='one_time_hosts')
label_info = topic_common.item_parse_info(attribute_name='labels',
inline_option='labels')
options, leftover = super(job_create_or_clone, self).parse(
[host_info, job_info, oth_info, label_info] + parse_info,
req_items='jobname')
self.data = {
'priority': options.priority,
}
jobname = getattr(self, 'jobname')
if len(jobname) > 1:
self.invalid_syntax('Too many arguments specified, only expected '
'to receive job name: %s' % jobname)
self.jobname = jobname[0]
if self.one_time_hosts:
self.data['one_time_hosts'] = self.one_time_hosts
if self.labels:
label_hosts = self.execute_rpc(op='get_hosts',
multiple_labels=self.labels)
for host in label_hosts:
self.hosts.append(host['hostname'])
self.data['name'] = self.jobname
(self.data['hosts'],
self.data['meta_hosts']) = self._parse_hosts(self.hosts)
self.data['email_list'] = options.email
return options, leftover
def create_job(self):
job_id = self.execute_rpc(op='create_job', **self.data)
return ['%s (id %s)' % (self.jobname, job_id)]
def get_items(self):
return [self.jobname]
class job_create(job_create_or_clone):
"""atest job create [--priority <int>]
[--synch_count] [--control-file </path/to/cfile>]
[--on-server] [--test <test1,test2>]
[--mlist </path/to/machinelist>] [--machine <host1 host2 host3>]
[--labels <list of labels of machines to run on>]
[--reboot_before <option>] [--reboot_after <option>]
[--noverify] [--timeout <timeout>] [--max_runtime <max runtime>]
[--one-time-hosts <hosts>] [--email <email>]
[--dependencies <labels this job is dependent on>]
[--parse-failed-repair <option>]
[--image <http://path/to/image>] [--require-ssp]
job_name
Creating a job is rather different from the other create operations,
so it only uses the __init__() and output() from its superclass.
"""
op_action = 'create'
def __init__(self):
super(job_create, self).__init__()
self.ctrl_file_data = {}
self.parser.add_option('-y', '--synch_count', type=int,
help='Number of machines to use per autoserv '
'execution')
self.parser.add_option('-f', '--control-file',
help='use this control file', metavar='FILE')
self.parser.add_option('-s', '--server',
help='This is server-side job',
action='store_true', default=False)
self.parser.add_option('-t', '--test',
help='List of tests to run')
self.parser.add_option('-d', '--dependencies', help='Comma separated '
'list of labels this job is dependent on.',
default='')
self.parser.add_option('-B', '--reboot_before',
help='Whether or not to reboot the machine '
'before the job (never/if dirty/always)',
type='choice',
choices=('never', 'if dirty', 'always'))
self.parser.add_option('-a', '--reboot_after',
help='Whether or not to reboot the machine '
'after the job (never/if all tests passed/'
'always)',
type='choice',
choices=('never', 'if all tests passed',
'always'))
self.parser.add_option('--parse-failed-repair',
help='Whether or not to parse failed repair '
'results as part of the job',
type='choice',
choices=('true', 'false'))
self.parser.add_option('-n', '--noverify',
help='Do not run verify for job',
default=False, action='store_true')
self.parser.add_option('-o', '--timeout_mins',
help='Job timeout in minutes.',
metavar='TIMEOUT')
self.parser.add_option('--max_runtime',
help='Job maximum runtime in minutes')
self.parser.add_option('-i', '--image',
help='OS image to install before running the '
'test.')
self.parser.add_option('--require-ssp',
help='Require server-side packaging',
default=False, action='store_true')
def parse(self):
deps_info = topic_common.item_parse_info(attribute_name='dependencies',
inline_option='dependencies')
options, leftover = super(job_create, self).parse(
parse_info=[deps_info])
if (len(self.hosts) == 0 and not self.one_time_hosts
and not options.labels):
self.invalid_syntax('Must specify at least one machine.'
'(-m, -M, -b or --one-time-hosts).')
if not options.control_file and not options.test:
self.invalid_syntax('Must specify either --test or --control-file'
' to create a job.')
if options.control_file and options.test:
self.invalid_syntax('Can only specify one of --control-file or '
'--test, not both.')
if options.control_file:
try:
control_file_f = open(options.control_file)
try:
control_file_data = control_file_f.read()
finally:
control_file_f.close()
except IOError:
self.generic_error('Unable to read from specified '
'control-file: %s' % options.control_file)
self.data['control_file'] = control_file_data
if options.test:
if options.server:
self.invalid_syntax('If you specify tests, then the '
'client/server setting is implicit and '
'cannot be overriden.')
tests = [t.strip() for t in options.test.split(',') if t.strip()]
self.ctrl_file_data['tests'] = tests
if options.image:
self.data['image'] = options.image
if options.reboot_before:
self.data['reboot_before'] = options.reboot_before.capitalize()
if options.reboot_after:
self.data['reboot_after'] = options.reboot_after.capitalize()
if options.parse_failed_repair:
self.data['parse_failed_repair'] = (
options.parse_failed_repair == 'true')
if options.noverify:
self.data['run_verify'] = False
if options.timeout_mins:
self.data['timeout_mins'] = options.timeout_mins
if options.max_runtime:
self.data['max_runtime_mins'] = options.max_runtime
self.data['dependencies'] = self.dependencies
if options.synch_count:
self.data['synch_count'] = options.synch_count
if options.server:
self.data['control_type'] = control_data.CONTROL_TYPE_NAMES.SERVER
else:
self.data['control_type'] = control_data.CONTROL_TYPE_NAMES.CLIENT
self.data['require_ssp'] = options.require_ssp
return options, leftover
def execute(self):
if self.ctrl_file_data:
cf_info = self.execute_rpc(op='generate_control_file',
item=self.jobname,
**self.ctrl_file_data)
self.data['control_file'] = cf_info['control_file']
if 'synch_count' not in self.data:
self.data['synch_count'] = cf_info['synch_count']
if cf_info['is_server']:
self.data['control_type'] = control_data.CONTROL_TYPE_NAMES.SERVER
else:
self.data['control_type'] = control_data.CONTROL_TYPE_NAMES.CLIENT
# Get the union of the 2 sets of dependencies
deps = set(self.data['dependencies'])
deps = sorted(deps.union(cf_info['dependencies']))
self.data['dependencies'] = list(deps)
if 'synch_count' not in self.data:
self.data['synch_count'] = 1
return self.create_job()
class job_clone(job_create_or_clone):
"""atest job clone [--priority <int>]
[--mlist </path/to/machinelist>] [--machine <host1 host2 host3>]
[--labels <list of labels of machines to run on>]
[--one-time-hosts <hosts>] [--email <email>]
job_name
Cloning a job is rather different from the other create operations,
so it only uses the __init__() and output() from its superclass.
"""
op_action = 'clone'
usage_action = 'clone'
def __init__(self):
super(job_clone, self).__init__()
self.parser.add_option('-i', '--id', help='Job id to clone',
default=False,
metavar='JOB_ID')
self.parser.add_option('-r', '--reuse-hosts',
help='Use the exact same hosts as the '
'cloned job.',
action='store_true', default=False)
def parse(self):
options, leftover = super(job_clone, self).parse()
self.clone_id = options.id
self.reuse_hosts = options.reuse_hosts
host_specified = self.hosts or self.one_time_hosts or options.labels
if self.reuse_hosts and host_specified:
self.invalid_syntax('Cannot specify hosts and reuse the same '
'ones as the cloned job.')
if not (self.reuse_hosts or host_specified):
self.invalid_syntax('Must reuse or specify at least one '
'machine (-r, -m, -M, -b or '
'--one-time-hosts).')
return options, leftover
def execute(self):
clone_info = self.execute_rpc(op='get_info_for_clone',
id=self.clone_id,
preserve_metahosts=self.reuse_hosts)
# Remove fields from clone data that cannot be reused
for field in ('name', 'created_on', 'id', 'owner'):
del clone_info['job'][field]
# Also remove parameterized_job field, as the feature still is
# incomplete, this tool does not attempt to support it for now,
# it uses a different API function and it breaks create_job()
if clone_info['job'].has_key('parameterized_job'):
del clone_info['job']['parameterized_job']
# Keyword args cannot be unicode strings
self.data.update((str(key), val)
for key, val in clone_info['job'].iteritems())
if self.reuse_hosts:
# Convert host list from clone info that can be used for job_create
for label, qty in clone_info['meta_host_counts'].iteritems():
self.data['meta_hosts'].extend([label]*qty)
self.data['hosts'].extend(host['hostname']
for host in clone_info['hosts'])
return self.create_job()
class job_abort(job, action_common.atest_delete):
"""atest job abort <job(s)>"""
usage_action = op_action = 'abort'
msg_done = 'Aborted'
def parse(self):
job_info = topic_common.item_parse_info(attribute_name='jobids',
use_leftover=True)
options, leftover = super(job_abort, self).parse([job_info],
req_items='jobids')
def execute(self):
data = {'job__id__in': self.jobids}
self.execute_rpc(op='abort_host_queue_entries', **data)
print('Aborting jobs: %s' % ', '.join(self.jobids))
def get_items(self):
return self.jobids