mylar/lib/apscheduler/jobstores/rethinkdb.py

154 lines
5.5 KiB
Python

from __future__ import absolute_import
from apscheduler.jobstores.base import BaseJobStore, JobLookupError, ConflictingIdError
from apscheduler.util import maybe_ref, datetime_to_utc_timestamp, utc_timestamp_to_datetime
from apscheduler.job import Job
try:
import cPickle as pickle
except ImportError: # pragma: nocover
import pickle
try:
import rethinkdb as r
except ImportError: # pragma: nocover
raise ImportError('RethinkDBJobStore requires rethinkdb installed')
class RethinkDBJobStore(BaseJobStore):
"""
Stores jobs in a RethinkDB database. Any leftover keyword arguments are directly passed to
rethinkdb's `RethinkdbClient <http://www.rethinkdb.com/api/#connect>`_.
Plugin alias: ``rethinkdb``
:param str database: database to store jobs in
:param str collection: collection to store jobs in
:param client: a :class:`rethinkdb.net.Connection` instance to use instead of providing
connection arguments
:param int pickle_protocol: pickle protocol level to use (for serialization), defaults to the
highest available
"""
def __init__(self, database='apscheduler', table='jobs', client=None,
pickle_protocol=pickle.HIGHEST_PROTOCOL, **connect_args):
super(RethinkDBJobStore, self).__init__()
if not database:
raise ValueError('The "database" parameter must not be empty')
if not table:
raise ValueError('The "table" parameter must not be empty')
self.database = database
self.table = table
self.client = client
self.pickle_protocol = pickle_protocol
self.connect_args = connect_args
self.conn = None
def start(self, scheduler, alias):
super(RethinkDBJobStore, self).start(scheduler, alias)
if self.client:
self.conn = maybe_ref(self.client)
else:
self.conn = r.connect(db=self.database, **self.connect_args)
if self.database not in r.db_list().run(self.conn):
r.db_create(self.database).run(self.conn)
if self.table not in r.table_list().run(self.conn):
r.table_create(self.table).run(self.conn)
if 'next_run_time' not in r.table(self.table).index_list().run(self.conn):
r.table(self.table).index_create('next_run_time').run(self.conn)
self.table = r.db(self.database).table(self.table)
def lookup_job(self, job_id):
results = list(self.table.get_all(job_id).pluck('job_state').run(self.conn))
return self._reconstitute_job(results[0]['job_state']) if results else None
def get_due_jobs(self, now):
return self._get_jobs(r.row['next_run_time'] <= datetime_to_utc_timestamp(now))
def get_next_run_time(self):
results = list(
self.table
.filter(r.row['next_run_time'] != None) # flake8: noqa
.order_by(r.asc('next_run_time'))
.map(lambda x: x['next_run_time'])
.limit(1)
.run(self.conn)
)
return utc_timestamp_to_datetime(results[0]) if results else None
def get_all_jobs(self):
jobs = self._get_jobs()
self._fix_paused_jobs_sorting(jobs)
return jobs
def add_job(self, job):
job_dict = {
'id': job.id,
'next_run_time': datetime_to_utc_timestamp(job.next_run_time),
'job_state': r.binary(pickle.dumps(job.__getstate__(), self.pickle_protocol))
}
results = self.table.insert(job_dict).run(self.conn)
if results['errors'] > 0:
raise ConflictingIdError(job.id)
def update_job(self, job):
changes = {
'next_run_time': datetime_to_utc_timestamp(job.next_run_time),
'job_state': r.binary(pickle.dumps(job.__getstate__(), self.pickle_protocol))
}
results = self.table.get_all(job.id).update(changes).run(self.conn)
skipped = False in map(lambda x: results[x] == 0, results.keys())
if results['skipped'] > 0 or results['errors'] > 0 or not skipped:
raise JobLookupError(job.id)
def remove_job(self, job_id):
results = self.table.get_all(job_id).delete().run(self.conn)
if results['deleted'] + results['skipped'] != 1:
raise JobLookupError(job_id)
def remove_all_jobs(self):
self.table.delete().run(self.conn)
def shutdown(self):
self.conn.close()
def _reconstitute_job(self, job_state):
job_state = pickle.loads(job_state)
job = Job.__new__(Job)
job.__setstate__(job_state)
job._scheduler = self._scheduler
job._jobstore_alias = self._alias
return job
def _get_jobs(self, predicate=None):
jobs = []
failed_job_ids = []
query = (self.table.filter(r.row['next_run_time'] != None).filter(predicate) if
predicate else self.table)
query = query.order_by('next_run_time', 'id').pluck('id', 'job_state')
for document in query.run(self.conn):
try:
jobs.append(self._reconstitute_job(document['job_state']))
except:
self._logger.exception('Unable to restore job "%s" -- removing it', document['id'])
failed_job_ids.append(document['id'])
# Remove all the jobs we failed to restore
if failed_job_ids:
r.expr(failed_job_ids).for_each(
lambda job_id: self.table.get_all(job_id).delete()).run(self.conn)
return jobs
def __repr__(self):
connection = self.conn
return '<%s (connection=%s)>' % (self.__class__.__name__, connection)