def harvest_jobs_run(context, data_dict):
'''
Runs scheduled jobs, checks if any jobs need marking as finished, and
resubmits queue items if needed.
This should be called every few minutes (e.g. by a cron), or else jobs
will never show as finished.
This used to also 'run' new jobs created by the web UI, putting them onto
the gather queue, but now this is done by default when you create a job. If
you need to send do this explicitly, then use
``harvest_send_job_to_gather_queue``.
:param source_id: the id of the harvest source, if you just want to check
for its finished jobs (optional)
:type source_id: string
'''
log.info('Harvest job run: %r', data_dict)
check_access('harvest_jobs_run', context, data_dict)
session = context['session']
source_id = data_dict.get('source_id')
# Scheduled jobs
if not source_id:
_make_scheduled_jobs(context, data_dict)
context['return_objects'] = False
# Flag finished jobs as such
jobs = harvest_job_list(context, {
'source_id': source_id,
'status': u'Running'
})
if len(jobs):
for job in jobs:
if job['gather_finished']:
num_objects_in_progress = \
session.query(HarvestObject.id) \
.filter(HarvestObject.harvest_job_id == job['id']) \
.filter(and_((HarvestObject.state != u'COMPLETE'),
(HarvestObject.state != u'ERROR'))) \
.count()
if num_objects_in_progress == 0:
job_obj = HarvestJob.get(job['id'])
job_obj.status = u'Finished'
log.info('Marking job as finished %s %s',
job_obj.source.url, job_obj.id)
# save the time of finish, according to the last running
# object
last_object = session.query(HarvestObject) \
.filter(HarvestObject.harvest_job_id == job['id']) \
.filter(
HarvestObject.import_finished != None # noqa: E711
).order_by(HarvestObject.import_finished.desc()) \
.first()
if last_object:
job_obj.finished = last_object.import_finished
else:
job_obj.finished = job['gather_finished']
job_obj.save()
# Reindex the harvest source dataset so it has the latest
# status
get_action('harvest_source_reindex')(
context, {
'id': job_obj.source.id
})
status = get_action('harvest_source_show_status')(
context, {
'id': job_obj.source.id
})
if config.get('ckan.harvest.status_mail') == 'errors' \
and status['last_job']['stats']['errored']:
subject, body = prepare_error_mail(
context, job_obj.source_id, status,
'emails/error_email.txt')
log.info('Sending error mail')
send_mail(context, job_obj.source.id, subject, body)
if config.get('ckan.harvest.status_mail') == 'all':
subject, body = prepare_summary_mail(
context, job_obj.source.id, status,
'emails/summary_email.txt')
log.info('Sending summary email')
send_mail(context, job_obj.source.id, subject, body)
else:
log.debug('Ongoing job:%s source:%s', job['id'],
job['source_id'])
log.debug('No jobs to send to the gather queue')
# Resubmit old redis tasks
resubmit_jobs()
# Resubmit pending objects missing from Redis
resubmit_objects()
return [] # merely for backwards compatibility
def harvest_jobs_run(context, data_dict):
'''
Runs scheduled jobs, checks if any jobs need marking as finished, and
resubmits queue items if needed.
If ckanext.harvest.timeout is set:
Check if the duration of the job is longer than ckanext.harvest.timeout,
then mark that job as finished as there is probably an underlying issue with the harvest process.
This should be called every few minutes (e.g. by a cron), or else jobs
will never show as finished.
This used to also 'run' new jobs created by the web UI, putting them onto
the gather queue, but now this is done by default when you create a job. If
you need to send do this explicitly, then use
``harvest_send_job_to_gather_queue``.
:param source_id: the id of the harvest source, if you just want to check
for its finished jobs (optional)
:type source_id: string
'''
log.info('Harvest job run: %r', data_dict)
check_access('harvest_jobs_run', context, data_dict)
timeout = config.get('ckan.harvest.timeout')
session = context['session']
source_id = data_dict.get('source_id')
# Scheduled jobs
if not source_id:
_make_scheduled_jobs(context, data_dict)
context['return_objects'] = False
# Flag finished jobs as such
jobs = harvest_job_list(context, {
'source_id': source_id,
'status': u'Running'
})
if len(jobs):
for job in jobs:
if timeout:
created = datetime.datetime.strptime(job['created'],
'%Y-%m-%d %H:%M:%S.%f')
now = datetime.datetime.now()
if now - created > datetime.timedelta(minutes=int(timeout)):
msg = 'Job timeout: %s is taking longer than %s minutes' % (
job['id'], timeout)
log.error(msg)
job_obj = HarvestJob.get(job['id'])
job_obj.status = u'Finished'
job_obj.finished = now
job_obj.save()
err = HarvestGatherError(message=msg, job=job_obj)
err.save()
log.info('Marking job as finished due to error: %s %s',
job_obj.source.url, job_obj.id)
continue
if job['gather_finished']:
num_objects_in_progress = \
session.query(HarvestObject.id) \
.filter(HarvestObject.harvest_job_id == job['id']) \
.filter(and_((HarvestObject.state != u'COMPLETE'),
(HarvestObject.state != u'ERROR'))) \
.count()
if num_objects_in_progress == 0:
job_obj = HarvestJob.get(job['id'])
job_obj.status = u'Finished'
log.info('Marking job as finished %s %s',
job_obj.source.url, job_obj.id)
# save the time of finish, according to the last running
# object
last_object = session.query(HarvestObject) \
.filter(HarvestObject.harvest_job_id == job['id']) \
.filter(
HarvestObject.import_finished != None # noqa: E711
).order_by(HarvestObject.import_finished.desc()) \
.first()
if last_object:
job_obj.finished = last_object.import_finished
else:
job_obj.finished = job['gather_finished']
job_obj.save()
# Reindex the harvest source dataset so it has the latest
# status
get_action('harvest_source_reindex')(
context, {
'id': job_obj.source.id
})
status = get_action('harvest_source_show_status')(
context, {
'id': job_obj.source.id
})
if toolkit.asbool(config.get('ckan.harvest.status_mail.errored'))\
and (status['last_job']['stats']['errored']):
send_error_mail(context, job_obj.source.id, status)
else:
log.debug('Ongoing job:%s source:%s', job['id'],
job['source_id'])
log.debug('No jobs to send to the gather queue')
# Resubmit old redis tasks
resubmit_jobs()
# Resubmit pending objects missing from Redis
resubmit_objects()
return [] # merely for backwards compatibility
def test_resubmit_objects(self):
'''
Test that only harvest objects re-submitted which were not be present in the redis fetch queue.
'''
if config.get('ckan.harvest.mq.type') != 'redis':
pytest.skip()
# make sure that there are no old elements in the redis db
redis = queue.get_connection()
fetch_routing_key = queue.get_fetch_routing_key()
redis.flushdb()
try:
# make sure queues/exchanges are created first and are empty
consumer = queue.get_gather_consumer()
consumer_fetch = queue.get_fetch_consumer()
consumer.queue_purge(queue=queue.get_gather_queue_name())
consumer_fetch.queue_purge(queue=queue.get_fetch_queue_name())
user = toolkit.get_action('get_site_user')(
{'model': model, 'ignore_auth': True}, {}
)['name']
context = {'model': model, 'session': model.Session,
'user': user, 'api_version': 3, 'ignore_auth': True}
harvest_source, job_id = self._create_harvest_job_and_finish_gather_stage(consumer, context)
assert redis.llen(fetch_routing_key) == 3
# do only one time for the first harvest object
reply = consumer_fetch.basic_get(queue='ckan.harvest.fetch')
queue.fetch_callback(consumer_fetch, *reply)
count = model.Session.query(model.Package) \
.filter(model.Package.type == 'dataset') \
.count()
assert count == 1
all_objects = model.Session.query(HarvestObject).order_by(HarvestObject.state.asc()).all()
assert len(all_objects) == 3
assert all_objects[0].state == 'COMPLETE'
assert all_objects[0].report_status == 'added'
assert all_objects[0].current is True
assert all_objects[1].state == 'WAITING'
assert all_objects[1].current is False
assert all_objects[2].state == 'WAITING'
assert all_objects[2].current is False
assert len(redis.keys(fetch_routing_key + ':*')) == 0
assert redis.llen(fetch_routing_key) == 2
# Remove one object from redis that should be re-sent to the fetch queue
reply = consumer_fetch.basic_get(queue='ckan.harvest.fetch')
fetch_queue_items = redis.lrange(fetch_routing_key, 0, 10)
assert len(fetch_queue_items) == 1
harvest_object_id = reply[2]
assert fetch_queue_items[0] != harvest_object_id
queue.resubmit_objects()
assert redis.llen(fetch_routing_key) == 2
fetch_queue_items = redis.lrange(fetch_routing_key, 0, 10)
assert harvest_object_id in fetch_queue_items
assert redis.dbsize() == 1
finally:
redis.flushdb()
def harvest_jobs_run(context, data_dict):
'''
Runs scheduled jobs, checks if any jobs need marking as finished, and
resubmits queue items if needed.
If ckanext.harvest.timeout is set:
Check if the duration of the job is longer than ckanext.harvest.timeout,
then mark that job as finished as there is probably an underlying issue with the harvest process.
This should be called every few minutes (e.g. by a cron), or else jobs
will never show as finished.
This used to also 'run' new jobs created by the web UI, putting them onto
the gather queue, but now this is done by default when you create a job. If
you need to send do this explicitly, then use
``harvest_send_job_to_gather_queue``.
:param source_id: the id of the harvest source, if you just want to check
for its finished jobs (optional)
:type source_id: string
'''
log.info('Harvest job run: %r', data_dict)
check_access('harvest_jobs_run', context, data_dict)
timeout = config.get('ckan.harvest.timeout')
session = context['session']
source_id = data_dict.get('source_id')
# Scheduled jobs
if not source_id:
_make_scheduled_jobs(context, data_dict)
context['return_objects'] = False
# Flag finished jobs as such
jobs = harvest_job_list(context, {
'source_id': source_id,
'status': u'Running'
})
if len(jobs):
for job in jobs:
job_obj = HarvestJob.get(job['id'])
if timeout:
last_time = job_obj.get_last_action_time()
now = datetime.datetime.utcnow()
if now - last_time > datetime.timedelta(minutes=int(timeout)):
msg = 'Job {} timeout ({} minutes)\n'.format(
job_obj.id, timeout)
msg += '\tJob created: {}\n'.format(job_obj.created)
msg += '\tJob gather finished: {}\n'.format(
job_obj.created)
msg += '\tJob last action time: {}\n'.format(last_time)
job_obj.status = u'Finished'
job_obj.finished = now
job_obj.save()
err = HarvestGatherError(message=msg, job=job_obj)
err.save()
log.info('Marking job as finished due to error: %s %s',
job_obj.source.url, job_obj.id)
continue
if job['gather_finished']:
num_objects_in_progress = \
session.query(HarvestObject.id) \
.filter(HarvestObject.harvest_job_id == job['id']) \
.filter(and_((HarvestObject.state != u'COMPLETE'),
(HarvestObject.state != u'ERROR'))) \
.count()
if num_objects_in_progress == 0:
job_obj.status = u'Finished'
log.info('Marking job as finished %s %s',
job_obj.source.url, job_obj.id)
# save the time of finish, according to the last running
# object
last_object = session.query(HarvestObject) \
.filter(HarvestObject.harvest_job_id == job['id']) \
.filter(
HarvestObject.import_finished != None # noqa: E711
).order_by(HarvestObject.import_finished.desc()) \
.first()
if last_object:
job_obj.finished = last_object.import_finished
else:
job_obj.finished = job['gather_finished']
job_obj.save()
# Reindex the harvest source dataset so it has the latest
# status
get_action('harvest_source_reindex')(
context, {
'id': job_obj.source.id
})
status = get_action('harvest_source_show_status')(
context, {
'id': job_obj.source.id
})
notify_all = toolkit.asbool(
config.get('ckan.harvest.status_mail.all'))
notify_errors = toolkit.asbool(
config.get('ckan.harvest.status_mail.errored'))
last_job_errors = status['last_job']['stats'].get(
'errored', 0)
log.debug(
'Notifications: All:{} On error:{} Errors:{}'.format(
notify_all, notify_errors, last_job_errors))
if last_job_errors > 0 and (notify_all or notify_errors):
# send_error_mail_ncar(context, job_obj)
# get_mail_extra_vars(context, job_obj.source.id, status)
send_error_email(context, job_obj.source.id, status)
elif notify_all:
send_summary_email(context, job_obj.source.id, status)
else:
log.debug('%d Ongoing jobs for %s (source:%s)',
num_objects_in_progress, job['id'],
job['source_id'])
log.debug('No jobs to send to the gather queue')
# Resubmit old redis tasks
resubmit_jobs()
# Resubmit pending objects missing from Redis
resubmit_objects()
# log.debug('Start of commit and close')
# session.commit()
# log.debug(' (Start of close)')
# session.close()
# log.debug('End of commit and close')
return [] # merely for backwards compatibility
def test_raise_child_error_and_retry(self):
""" if a harvest job for a child fails because
parent still not exists we need to ensure
this job will be retried.
This test emulate the case we harvest children first
(e.g. if we have several active queues).
Just for CKAN 2.8 env"""
# start harvest process with gather to create harvest objects
url = 'http://127.0.0.1:%s/collection-1-parent-2-children.data.json' % self.mock_port
self.run_gather(url=url)
assert_equal(len(self.harvest_objects), 3)
# create a publisher to send this objects to the fetch queue
publisher = queue.get_fetch_publisher()
for ho in self.harvest_objects:
ho = harvest_model.HarvestObject.get(ho.id) # refresh
ho_data = json.loads(ho.content)
assert_equal(ho.state, 'WAITING')
log.info('HO: {}\n\tCurrent: {}'.format(ho_data['identifier'],
ho.current))
assert_equal(ho.retry_times, 0)
publisher.send({'harvest_object_id': ho.id})
log.info('Harvest object sent to the fetch queue {} as {}'.format(
ho_data['identifier'], ho.id))
publisher.close()
# run fetch for elements in the wrong order (first a child, the a parent)
class FakeMethod(object):
''' This is to act like the method returned by AMQP'''
def __init__(self, message):
self.delivery_tag = message
# get the fetch
consumer_fetch = queue.get_fetch_consumer()
qname = queue.get_fetch_queue_name()
# first a child and assert to get an error
r2 = json.dumps({"harvest_object_id": self.harvest_objects[1].id})
r0 = FakeMethod(r2)
with assert_raises(ParentNotHarvestedException):
queue.fetch_callback(consumer_fetch, r0, None, r2)
assert_equal(self.harvest_objects[1].retry_times, 1)
assert_equal(self.harvest_objects[1].state, "ERROR")
# run the parent later, like in a different queue
r2 = json.dumps({"harvest_object_id": self.harvest_objects[0].id})
r0 = FakeMethod(r2)
queue.fetch_callback(consumer_fetch, r0, None, r2)
assert_equal(self.harvest_objects[0].retry_times, 1)
assert_equal(self.harvest_objects[0].state, "COMPLETE")
# Check status on harvest objects
# We expect one child with error, parent ok and second child still waiting
for ho in self.harvest_objects:
ho = harvest_model.HarvestObject.get(ho.id) # refresh
ho_data = json.loads(ho.content)
idf = ho_data['identifier']
log.info(
'\nHO2: {}\n\tState: {}\n\tCurrent: {}\n\tGathered {}'.format(
idf, ho.state, ho.current, ho.gathered))
if idf == 'OPM-ERround-0001':
assert_equal(ho.state, 'COMPLETE')
elif idf == 'OPM-ERround-0001-AWOL':
assert_equal(ho.state, 'ERROR')
ho_awol_id = ho.id
elif idf == 'OPM-ERround-0001-Retire':
assert_equal(ho.state, 'WAITING')
ho_retire_id = ho.id
else:
raise Exception('Unexpected identifier: "{}"'.format(idf))
# resubmit jobs and objects as harvest_jobs_run does
# we expect the errored harvest object is in this queue
queue.resubmit_jobs()
queue.resubmit_objects()
# iterate over the fetch consumer queue again and check pending harvest objects
harvest_objects = []
while True:
method, header, body = consumer_fetch.basic_get(queue=qname)
if body is None:
break
body_data = json.loads(body)
ho_id = body_data.get('harvest_object_id', None)
log.info('Adding ho_id {}'.format(ho_id))
if ho_id is not None:
ho = harvest_model.HarvestObject.get(ho_id)
if ho is not None:
harvest_objects.append(ho)
content = json.loads(ho.content)
log.info('Harvest object found {}: {} '.format(
content['identifier'], ho.state))
else:
log.info('Harvest object not found {}'.format(ho_id))
ho_ids = [ho.id for ho in harvest_objects]
# Now, we expect the waiting child and the errored one to be in the fetch queue
log.info('Searching wainting object "Retire ID"')
assert_in(ho_retire_id, ho_ids)
log.info('Searching errored object "Awol ID"')
assert_in(ho_awol_id, ho_ids)
