argh, ugly hack for problem of stale requests poisoning queues

server-ca/daemon.py

@@ -290,6 +290,16 @@ for message in ps.listen():
     populated_queue = message["data"]
     if populated_queue in dispatch:
         queue, function = dispatch[populated_queue]
+        for repetition in (1,2):
+            # This is a potentially inappropriate hack to prevent backlogs
+            # from accumulating in queues if daemons die or if spurious
+            # requests arrived while no daemons were listening.  The idea
+            # is that every daemon process double-checks a queue that it
+            # was told to look at, processing it twice per pubsub message
+            # instead of once.  This causes a pressure on the daemon to
+            # empty the queue over time even if it isn't explicitly asked
+            # to.  For example, if asked 7 times to process a particular
+            # queue, a daemon instance would try 14 times.
             session = r.rpop(queue)
             if session:
                 if debug: print "going to %s for %s" % (populated_queue, session)