def start():
#TODO(robnagler) commands need their own init hook like the server has
job.init()
global cfg
cfg = pkconfig.init(
agent_id=pkconfig.Required(str, 'id of this agent'),
fastcgi_sock_dir=(
pkio.py_path('/tmp'), pkio.py_path,
'directory of fastcfgi socket, must be less than 50 chars'),
start_delay=(0, pkconfig.parse_seconds,
'delay startup in internal_test mode'),
supervisor_uri=pkconfig.Required(
str,
'how to connect to the supervisor',
),
)
pkdlog('{}', cfg)
if pkconfig.channel_in_internal_test() and cfg.start_delay:
pkdlog('start_delay={}', cfg.start_delay)
time.sleep(cfg.start_delay)
i = tornado.ioloop.IOLoop.current()
d = _Dispatcher()
def s(*args):
return i.add_callback_from_signal(_terminate, d)
signal.signal(signal.SIGTERM, s)
signal.signal(signal.SIGINT, s)
i.spawn_callback(d.loop)
i.start()
def default_command(in_file):
"""Reads `in_file` passes to `msg.jobCmd`
Must be called in run_dir
Writes its output on stdout.
Args:
in_file (str): json parsed to msg
Returns:
str: json output of command, e.g. status msg
"""
try:
job.init()
f = pkio.py_path(in_file)
msg = pkjson.load_any(f)
#TODO(e-carlin): find common place to serialize/deserialize paths
msg.runDir = pkio.py_path(msg.runDir)
f.remove()
res = globals()['_do_' + msg.jobCmd](msg,
sirepo.template.import_module(
msg.simulationType))
if res is None:
return
r = PKDict(res).pksetdefault(state=job.COMPLETED)
except Exception as e:
r = PKDict(
state=job.ERROR,
error=e.sr_args.error
if isinstance(e, sirepo.util.UserAlert) else str(e),
stack=pkdexc(),
)
return pkjson.dump_pretty(r, pretty=False)
def init():
global _DB_DIR, cfg, _NEXT_REQUEST_SECONDS, job_driver
if _DB_DIR:
return
job.init()
from sirepo import job_driver
job_driver.init(pkinspect.this_module())
_DB_DIR = sirepo.srdb.root().join(_DB_SUBDIR)
cfg = pkconfig.init(
job_cache_secs=(300, int, 'when to re-read job state from disk'),
max_hours=dict(
analysis=(.04, float, 'maximum run-time for analysis job',),
parallel=(1, float, 'maximum run-time for parallel job (except sbatch)'),
parallel_premium=(2, float, 'maximum run-time for parallel job for premium user (except sbatch)'),
sequential=(.1, float, 'maximum run-time for sequential job'),
),
purge_free_after_days=(1000, int, 'how many days to wait before purging a free users simulation'),
purge_free_start=('02:00:00', str, 'time to first start purging free users simulations (%H:%M:%S)'),
purge_free_period=('01:00:00', str, 'how often to purge free users simulations after start time (%H:%M:%S)'),
sbatch_poll_secs=(15, int, 'how often to poll squeue and parallel status'),
)
_NEXT_REQUEST_SECONDS = PKDict({
job.PARALLEL: 2,
job.SBATCH: cfg.sbatch_poll_secs,
job.SEQUENTIAL: 1,
})
sirepo.auth_db.init(sirepo.srdb.root(), migrate_db_file=False)
if sirepo.simulation_db.user_dir_name().exists():
if not _DB_DIR.exists():
pkdlog('calling upgrade_runner_to_job_db path={}', _DB_DIR)
import subprocess
subprocess.check_call(
(
'pyenv',
'exec',
'sirepo',
'db',
'upgrade_runner_to_job_db',
_DB_DIR,
),
env=PKDict(os.environ).pkupdate(
PYENV_VERSION='py2',
SIREPO_AUTH_LOGGED_IN_USER='******',
),
)
else:
pykern.pkio.mkdir_parent(_DB_DIR)
tornado.ioloop.IOLoop.current().add_callback(
_ComputeJob.purge_free_simulations,
init=True,
)
def init():
global _DB_DIR, cfg, _NEXT_REQUEST_SECONDS, job_driver
if _DB_DIR:
return
job.init()
from sirepo import job_driver
job_driver.init(pkinspect.this_module())
_DB_DIR = sirepo.srdb.root().join(_DB_SUBDIR)
cfg = pkconfig.init(
job_cache_secs=(300, int, 'when to re-read job state from disk'),
max_secs=dict(
analysis=(144, pkconfig.parse_seconds, 'maximum run-time for analysis job',),
parallel=(3600, pkconfig.parse_seconds, 'maximum run-time for parallel job (except sbatch)'),
parallel_premium=(3600*2, pkconfig.parse_seconds, 'maximum run-time for parallel job for premium user (except sbatch)'),
sequential=(360, pkconfig.parse_seconds, 'maximum run-time for sequential job'),
),
purge_non_premium_after_secs=(0, pkconfig.parse_seconds, 'how long to wait before purging non-premium users simulations'),
purge_non_premium_task_secs=(None, pkconfig.parse_seconds, 'when to clean up simulation runs of non-premium users (%H:%M:%S)'),
sbatch_poll_secs=(15, int, 'how often to poll squeue and parallel status'),
)
_NEXT_REQUEST_SECONDS = PKDict({
job.PARALLEL: 2,
job.SBATCH: cfg.sbatch_poll_secs,
job.SEQUENTIAL: 1,
})
sirepo.auth_db.init()
if sirepo.simulation_db.user_path().exists():
if not _DB_DIR.exists():
pkdlog('calling upgrade_runner_to_job_db path={}', _DB_DIR)
import subprocess
subprocess.check_call(
(
'pyenv',
'exec',
'sirepo',
'db',
'upgrade_runner_to_job_db',
_DB_DIR,
),
env=PKDict(os.environ).pkupdate(
SIREPO_AUTH_LOGGED_IN_USER='******',
),
)
else:
pykern.pkio.mkdir_parent(_DB_DIR)
tornado.ioloop.IOLoop.current().add_callback(
_ComputeJob.purge_free_simulations,
)
def init():
global cfg, _DB_DIR, _NEXT_REQUEST_SECONDS, job_driver
if cfg:
return
job.init()
from sirepo import job_driver
job_driver.init(pkinspect.this_module())
cfg = pkconfig.init(
job_cache_secs=(300, int, 'when to re-read job state from disk'),
max_secs=dict(
analysis=(
144,
pkconfig.parse_seconds,
'maximum run-time for analysis job',
),
parallel=(3600, pkconfig.parse_seconds,
'maximum run-time for parallel job (except sbatch)'),
parallel_premium=
(3600 * 2, pkconfig.parse_seconds,
'maximum run-time for parallel job for premium user (except sbatch)'
),
sequential=(360, pkconfig.parse_seconds,
'maximum run-time for sequential job'),
),
purge_non_premium_after_secs=(
0, pkconfig.parse_seconds,
'how long to wait before purging non-premium users simulations'),
purge_non_premium_task_secs=(
None, pkconfig.parse_seconds,
'when to clean up simulation runs of non-premium users (%H:%M:%S)'
),
sbatch_poll_secs=(15, int,
'how often to poll squeue and parallel status'),
)
_DB_DIR = sirepo.srdb.supervisor_dir()
_NEXT_REQUEST_SECONDS = PKDict({
job.PARALLEL: 2,
job.SBATCH: cfg.sbatch_poll_secs,
job.SEQUENTIAL: 1,
})
sirepo.auth_db.init()
tornado.ioloop.IOLoop.current().add_callback(
_ComputeJob.purge_free_simulations, )
def init():
global _DB_DIR, cfg, _NEXT_REQUEST_SECONDS
if _DB_DIR:
return
job.init()
job_driver.init()
_DB_DIR = sirepo.srdb.root().join(_DB_SUBDIR)
cfg = pkconfig.init(
parallel=dict(
max_hours=(1, float,
'maximum run-time for parallel job (except sbatch)'), ),
sbatch_poll_secs=(60, int,
'how often to poll squeue and parallel status'),
sequential=dict(max_hours=(.1, float,
'maximum run-time for sequential job'), ),
)
for k in job.KINDS:
_MAX_RUN_SECS[k] = int(cfg[k].max_hours * 3600)
_NEXT_REQUEST_SECONDS = PKDict({
job.PARALLEL: 2,
job.SBATCH: cfg.sbatch_poll_secs,
job.SEQUENTIAL: 1,
})
if sirepo.simulation_db.user_dir_name().exists():
if not _DB_DIR.exists():
pkdlog('calling upgrade_runner_to_job_db path={}', _DB_DIR)
import subprocess
subprocess.check_call(
(
'pyenv',
'exec',
'sirepo',
'db',
'upgrade_runner_to_job_db',
_DB_DIR,
),
env=PKDict(os.environ).pkupdate(
PYENV_VERSION='py2',
SIREPO_AUTH_LOGGED_IN_USER='******',
),
)
else:
pykern.pkio.mkdir_parent(_DB_DIR)
def start():
#TODO(robnagler) commands need their own init hook like the server has
job.init()
global cfg
cfg = pkconfig.init(
agent_id=pkconfig.Required(str, 'id of this agent'),
supervisor_uri=pkconfig.Required(
str,
'how to connect to the supervisor',
),
)
pkdlog('{}', cfg)
i = tornado.ioloop.IOLoop.current()
d = _Dispatcher()
def s(*args):
return i.add_callback_from_signal(_terminate, d)
signal.signal(signal.SIGTERM, s)
signal.signal(signal.SIGINT, s)
i.spawn_callback(d.loop)
i.start()
