cc-4105: fixed a ton of formatting
This commit is contained in:
parent
4f29301334
commit
a8b9b300e5
4 changed files with 56 additions and 42 deletions
|
@ -6,7 +6,6 @@ import traceback
|
|||
|
||||
from media.monitor.handler import ReportHandler
|
||||
from media.monitor.log import Loggable
|
||||
#from media.monitor.listeners import FileMediator
|
||||
from media.monitor.exceptions import BadSongFile
|
||||
from media.monitor.pure import LazyProperty
|
||||
|
||||
|
@ -25,7 +24,8 @@ class RequestSync(threading.Thread,Loggable):
|
|||
return ac.AirtimeApiClient.create_right_config()
|
||||
|
||||
def run(self):
|
||||
self.logger.info("Attempting request with %d items." % len(self.requests))
|
||||
self.logger.info("Attempting request with %d items." %
|
||||
len(self.requests))
|
||||
# Note that we must attach the appropriate mode to every response. Also
|
||||
# Not forget to attach the 'is_record' to any requests that are related
|
||||
# to recorded shows
|
||||
|
@ -43,7 +43,6 @@ class RequestSync(threading.Thread,Loggable):
|
|||
except Exception as e:
|
||||
self.logger.info("An evil exception occured")
|
||||
self.logger.error( traceback.format_exc() )
|
||||
import ipdb; ipdb.set_trace()
|
||||
def make_req():
|
||||
self.apiclient.send_media_monitor_requests( packed_requests )
|
||||
# Is this retry shit even necessary? Consider getting rid of this.
|
||||
|
@ -54,15 +53,16 @@ class RequestSync(threading.Thread,Loggable):
|
|||
self.logger.info("Api Controller is a piece of shit\n \
|
||||
it's not returning json when it should\n \
|
||||
... will fix once I setup the damn debugger")
|
||||
self.logger.info("Trying again after %f seconds" % self.request_wait)
|
||||
self.logger.info("Trying again after %f seconds" %
|
||||
self.request_wait)
|
||||
time.sleep( self.request_wait )
|
||||
except Exception as e:
|
||||
self.unexpected_exception(e)
|
||||
except Exception as e: self.unexpected_exception(e)
|
||||
else:
|
||||
self.logger.info("Request worked on the '%d' try" % (try_index + 1))
|
||||
self.logger.info("Request worked on the '%d' try" %
|
||||
(try_index + 1))
|
||||
break
|
||||
else: self.logger.info("Failed to send request after '%d' tries..." % self.retries)
|
||||
#self.logger.info("Now ignoring: %d files" % len(FileMediator.ignored_set))
|
||||
else: self.logger.info("Failed to send request after '%d' tries..." %
|
||||
self.retries)
|
||||
self.watcher.flag_done()
|
||||
|
||||
class TimeoutWatcher(threading.Thread,Loggable):
|
||||
|
@ -86,7 +86,8 @@ class TimeoutWatcher(threading.Thread,Loggable):
|
|||
self.watcher.request_do()
|
||||
# Same for events, this behaviour is mandatory however.
|
||||
if self.watcher.events_in_queue():
|
||||
self.logger.info("We got %d events that are unflushed" % self.watcher.events_left_count())
|
||||
self.logger.info("We got %d events that are unflushed" %
|
||||
self.watcher.events_left_count())
|
||||
self.watcher.flush_events()
|
||||
|
||||
class WatchSyncer(ReportHandler,Loggable):
|
||||
|
@ -146,7 +147,8 @@ class WatchSyncer(ReportHandler,Loggable):
|
|||
self.request_do()
|
||||
|
||||
def events_in_queue(self):
|
||||
"""returns true if there are events in the queue that haven't been processed yet"""
|
||||
"""returns true if there are events in the queue that haven't been
|
||||
processed yet"""
|
||||
return len(self.__queue) > 0
|
||||
|
||||
def requests_in_queue(self):
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue