import arvados
import config
import errors
+from retry import RetryLoop
import logging
import json
+import thread
import threading
import time
import os
self.filters = [[]]
self.on_event_cb = on_event_cb
self.last_log_id = last_log_id
- self.is_closed = False
- self.ec = _EventClient(url, self.filters, self.on_event, last_log_id, self.on_closed)
+ self.is_closed = threading.Event()
+ self._setup_event_client()
- def connect(self):
- self.ec.connect()
-
- def close_connection(self):
- self.ec.close_connection()
+ def _setup_event_client(self):
+ self.ec = _EventClient(self.url, self.filters, self.on_event,
+ self.last_log_id, self.on_closed)
+ self.ec.daemon = True
+ try:
+ self.ec.connect()
+ except Exception:
+ self.ec.close_connection()
+ raise
def subscribe(self, f, last_log_id=None):
self.filters.append(f)
self.ec.unsubscribe(f)
def close(self, code=1000, reason='', timeout=0):
- self.is_closed = True
+ self.is_closed.set()
self.ec.close(code, reason, timeout)
def on_event(self, m):
try:
self.on_event_cb(m)
except Exception as e:
- _logger.warn("Unexpected exception from event callback.", exc_info=e)
+ _logger.exception("Unexpected exception from event callback.")
+ thread.interrupt_main()
def on_closed(self):
- if self.is_closed == False:
+ if not self.is_closed.is_set():
_logger.warn("Unexpected close. Reconnecting.")
- self.ec = _EventClient(self.url, self.filters, self.on_event, self.last_log_id, self.on_closed)
- while True:
- try:
- self.ec.connect()
- break
- except Exception as e:
- _logger.warn("Error '%s' during websocket reconnect. Will retry after 5s.", e, exc_info=e)
- time.sleep(5)
+ for tries_left in RetryLoop(num_retries=25, backoff_start=.1, max_wait=15):
+ try:
+ self._setup_event_client()
+ break
+ except Exception as e:
+ _logger.warn("Error '%s' during websocket reconnect.", e)
+ if tries_left == 0:
+ _logger.exception("EventClient thread could not contact websocket server.")
+ self.is_closed.set()
+ thread.interrupt_main()
+ return
+
+ def run_forever(self):
+ # Have to poll here to let KeyboardInterrupt get raised.
+ while not self.is_closed.wait(1):
+ pass
class PollClient(threading.Thread):
self.id = self.last_log_id
else:
for f in self.filters:
- try:
- items = self.api.logs().list(limit=1, order="id desc", filters=f).execute(num_retries=1000000)['items']
- except Exception as e:
- # Some apparently non-retryable error happened, so log the
- # error and shut down gracefully.
- _logger.error("Got exception from log query: %s", e)
+ for tries_left in RetryLoop(num_retries=25, backoff_start=.1, max_wait=self.poll_time):
+ try:
+ items = self.api.logs().list(limit=1, order="id desc", filters=f).execute()['items']
+ break
+ except errors.ApiError as error:
+ pass
+ else:
+ tries_left = 0
+ break
+ if tries_left == 0:
+ _logger.exception("PollClient thread could not contact API server.")
with self._closing_lock:
self._closing.set()
+ thread.interrupt_main()
return
if items:
if items[0]['id'] > self.id:
max_id = self.id
moreitems = False
for f in self.filters:
- try:
- # If we get a transient error, we really really need to
- # just keep trying over and over with the same query or
- # we'll potentially drop events which would break the event
- # stream contract.
- items = self.api.logs().list(order="id asc", filters=f+[["id", ">", str(self.id)]]).execute(num_retries=1000000)
- except Exception as e:
- # Some apparently non-retryable error happened, so log the
- # error and shut down gracefully.
- _logger.error("Got exception from log query: %s", e)
+ for tries_left in RetryLoop(num_retries=25, backoff_start=.1, max_wait=self.poll_time):
+ try:
+ items = self.api.logs().list(order="id asc", filters=f+[["id", ">", str(self.id)]]).execute()
+ break
+ except errors.ApiError as error:
+ pass
+ else:
+ tries_left = 0
+ break
+ if tries_left == 0:
+ _logger.exception("PollClient thread could not contact API server.")
with self._closing_lock:
self._closing.set()
+ thread.interrupt_main()
return
for i in items["items"]:
if i['id'] > max_id:
try:
self.on_event(i)
except Exception as e:
- _logger.warn("Unexpected exception from event callback.", exc_info=e)
+ _logger.exception("Unexpected exception from event callback.")
+ thread.interrupt_main()
if items["items_available"] > len(items["items"]):
moreitems = True
self.id = max_id
if not endpoint:
raise errors.FeatureNotEnabledError(
"Server does not advertise a websocket endpoint")
+ uri_with_token = "{}?api_token={}".format(endpoint, api.api_token)
try:
- uri_with_token = "{}?api_token={}".format(endpoint, api.api_token)
client = EventClient(uri_with_token, filters, on_event, last_log_id)
- ok = False
- try:
- client.connect()
- ok = True
- return client
- finally:
- if not ok:
- client.close_connection()
- except:
+ except Exception:
_logger.warn("Failed to connect to websockets on %s" % endpoint)
raise
+ else:
+ return client
def subscribe(api, filters, on_event, poll_fallback=15, last_log_id=None):