def api_runCancel():
data = _parse_data_input()
jid = simulation_db.job_id(data)
if feature_config.cfg.runner_daemon:
jhash = template_common.report_parameters_hash(data)
run_dir = simulation_db.simulation_run_dir(data)
runner_client.cancel_report_job(run_dir, jhash)
# Always true from the client's perspective
return http_reply.gen_json({'state': 'canceled'})
else:
# TODO(robnagler) need to have a way of listing jobs
# Don't bother with cache_hit check. We don't have any way of canceling
# if the parameters don't match so for now, always kill.
#TODO(robnagler) mutex required
if runner.job_is_processing(jid):
run_dir = simulation_db.simulation_run_dir(data)
# Write first, since results are write once, and we want to
# indicate the cancel instead of the termination error that
# will happen as a result of the kill.
simulation_db.write_result({'state': 'canceled'}, run_dir=run_dir)
runner.job_kill(jid)
# TODO(robnagler) should really be inside the template (t.cancel_simulation()?)
# the last frame file may not be finished, remove it
t = sirepo.template.import_module(data)
if hasattr(t, 'remove_last_frame'):
t.remove_last_frame(run_dir)
# Always true from the client's perspective
return http_reply.gen_json({'state': 'canceled'})
def api_simulationFrame(frame_id):
#TODO(robnagler) startTime is reportParametersHash; need version on URL and/or param names in URL
keys = ['simulationType', 'simulationId', 'modelName', 'animationArgs', 'frameIndex', 'startTime']
data = dict(zip(keys, frame_id.split('*')))
template = sirepo.template.import_module(data)
data['report'] = template.get_animation_name(data)
run_dir = simulation_db.simulation_run_dir(data)
model_data = simulation_db.read_json(run_dir.join(template_common.INPUT_BASE_NAME))
if feature_config.cfg.runner_daemon:
# XX TODO: it would be better if the frontend passed the jhash to this
# call. Since it doesn't, we have to read it out of the run_dir, which
# creates a race condition -- we might return a frame from a different
# version of the report than the one the frontend expects.
jhash = template_common.report_parameters_hash(model_data)
frame = runner_client.run_extract_job(
run_dir, jhash, 'get_simulation_frame', data,
)
else:
frame = template.get_simulation_frame(run_dir, data, model_data)
resp = http_reply.gen_json(frame)
if 'error' not in frame and template.WANT_BROWSER_FRAME_CACHE:
now = datetime.datetime.utcnow()
expires = now + datetime.timedelta(365)
resp.headers['Cache-Control'] = 'public, max-age=31536000'
resp.headers['Expires'] = expires.strftime("%a, %d %b %Y %H:%M:%S GMT")
resp.headers['Last-Modified'] = now.strftime("%a, %d %b %Y %H:%M:%S GMT")
else:
_no_cache(resp)
return resp
def report_info(data):
"""Read the run_dir and return cached_data.
Only a hit if the models between data and cache match exactly. Otherwise,
return cached data if it's there and valid.
Args:
data (dict): parameters identifying run_dir and models or reportParametersHash
Returns:
Dict: report parameters and hashes
"""
# Sets data['reportParametersHash']
rep = pkcollections.Dict(
cache_hit=False,
cached_data=None,
cached_hash=None,
job_id=job_id(data),
model_name=data['report'],
parameters_changed=False,
run_dir=simulation_run_dir(data),
)
rep.input_file = json_filename(template_common.INPUT_BASE_NAME,
rep.run_dir)
rep.job_status = read_status(rep.run_dir)
rep.req_hash = template_common.report_parameters_hash(data)
if not rep.run_dir.check():
return rep
#TODO(robnagler) Lock
try:
cd = read_json(rep.input_file)
rep.cached_hash = template_common.report_parameters_hash(cd)
rep.cached_data = cd
if rep.req_hash == rep.cached_hash:
rep.cache_hit = True
return rep
rep.parameters_changed = True
except IOError as e:
pkdlog('{}: ignore IOError: {} errno={}', rep.run_dir, e, e.errno)
except Exception as e:
pkdlog('{}: ignore other error: {}', rep.run_dir, e)
# No idea if cache is valid or not so throw away
return rep
def report_info(data):
"""Read the run_dir and return cached_data.
Only a hit if the models between data and cache match exactly. Otherwise,
return cached data if it's there and valid.
Args:
data (dict): parameters identifying run_dir and models or reportParametersHash
Returns:
Dict: report parameters and hashes
"""
# Sets data['reportParametersHash']
rep = pkcollections.Dict(
cache_hit=False,
cached_data=None,
cached_hash=None,
job_id=job_id(data),
model_name=_report_name(data),
parameters_changed=False,
run_dir=simulation_run_dir(data),
)
rep.input_file = json_filename(template_common.INPUT_BASE_NAME, rep.run_dir)
rep.job_status = read_status(rep.run_dir)
rep.req_hash = template_common.report_parameters_hash(data)
if not rep.run_dir.check():
return rep
#TODO(robnagler) Lock
try:
cd = read_json(rep.input_file)
rep.cached_hash = template_common.report_parameters_hash(cd)
rep.cached_data = cd
if rep.req_hash == rep.cached_hash:
rep.cache_hit = True
return rep
rep.parameters_changed = True
except IOError as e:
pkdlog('{}: ignore IOError: {} errno={}', rep.run_dir, e, e.errno)
except Exception as e:
pkdlog('{}: ignore other error: {}', rep.run_dir, e)
# No idea if cache is valid or not so throw away
return rep
def api_runSimulation():
from pykern import pkjson
data = _parse_data_input(validate=True)
# if flag is set
# - check status
# - if status is bad, rewrite the run dir (XX race condition, to fix later)
# - then request it be started
if feature_config.cfg.runner_daemon:
jhash = template_common.report_parameters_hash(data)
run_dir = simulation_db.simulation_run_dir(data)
status = runner_client.report_job_status(run_dir, jhash)
already_good_status = [
runner_client.JobStatus.RUNNING, runner_client.JobStatus.COMPLETED
]
if status not in already_good_status:
data['simulationStatus'] = {
'startTime': int(time.time()),
'state': 'pending',
}
tmp_dir = run_dir + '-' + jhash + '-' + uuid.uuid4(
) + srdb.TMP_DIR_SUFFIX
cmd, _ = simulation_db.prepare_simulation(data, tmp_dir=tmp_dir)
runner_client.start_report_job(run_dir, jhash, cfg.backend, cmd,
tmp_dir)
res = _simulation_run_status_runner_daemon(data, quiet=True)
return http_reply.gen_json(res)
else:
res = _simulation_run_status(data, quiet=True)
if ((not res['state'] in _RUN_STATES and
(res['state'] != 'completed' or data.get('forceRun', False)))
or res.get('parametersChanged', True)):
try:
_start_simulation(data)
except runner.Collision:
pkdlog('{}: runner.Collision, ignoring start',
simulation_db.job_id(data))
res = _simulation_run_status(data)
return http_reply.gen_json(res)
def _simulation_run_status_runner_daemon(data, quiet=False):
"""Look for simulation status and output
Args:
data (dict): request
quiet (bool): don't write errors to log
Returns:
dict: status response
"""
try:
run_dir = simulation_db.simulation_run_dir(data)
jhash = template_common.report_parameters_hash(data)
status = runner_client.report_job_status(run_dir, jhash)
is_running = status is runner_client.JobStatus.RUNNING
rep = simulation_db.report_info(data)
res = {'state': status.value}
if not is_running:
if status is not runner_client.JobStatus.MISSING:
res, err = runner_client.run_extract_job(
run_dir,
jhash,
'result',
data,
)
if err:
return _simulation_error(err, 'error in read_result',
run_dir)
if simulation_db.is_parallel(data):
new = runner_client.run_extract_job(
run_dir,
jhash,
'background_percent_complete',
is_running,
)
new.setdefault('percentComplete', 0.0)
new.setdefault('frameCount', 0)
res.update(new)
res['parametersChanged'] = rep.parameters_changed
if res['parametersChanged']:
pkdlog(
'{}: parametersChanged=True req_hash={} cached_hash={}',
rep.job_id,
rep.req_hash,
rep.cached_hash,
)
#TODO(robnagler) verify serial number to see what's newer
res.setdefault('startTime', _mtime_or_now(rep.input_file))
res.setdefault('lastUpdateTime', _mtime_or_now(rep.run_dir))
res.setdefault('elapsedTime', res['lastUpdateTime'] - res['startTime'])
if is_running:
res['nextRequestSeconds'] = simulation_db.poll_seconds(
rep.cached_data)
res['nextRequest'] = {
'report': rep.model_name,
'reportParametersHash': rep.cached_hash,
'simulationId': rep.cached_data['simulationId'],
'simulationType': rep.cached_data['simulationType'],
}
pkdc(
'{}: processing={} state={} cache_hit={} cached_hash={} data_hash={}',
rep.job_id,
is_running,
res['state'],
rep.cache_hit,
rep.cached_hash,
rep.req_hash,
)
except Exception:
return _simulation_error(pkdexc(), quiet=quiet)
return res