Fix rdb when sockets default to nonblocking

In some celery configurations (we run with the process pool and have
the issue) by default socket is configured to have a timeout of 5
seconds. The rdb plugin assumes non-blocking sockets so this led to
unexpected termination of debugging sessions with an exception that
could take the entire worker down if invoked using the SIGUSR2
mechanism.

This patch makes both the accepting and communicating sockets for rdb
explicitly blocking to avoid this issue.

celery/contrib/rdb.py

@@ -74,6 +74,7 @@ class Rdb(Pdb):
         self._sock, this_port = self.get_avail_port(
             host, port, port_search_limit, port_skew,
         )
+        self._sock.setblocking(1)
         self._sock.listen(1)
         me = '%s:%s' % (self.me, this_port)
         context = self.context = {'me': me, 'host': host, 'port': this_port}
@@ -82,6 +83,7 @@ class Rdb(Pdb):
         self.say('%(me)s: Waiting for client...' % context)
 
         self._client, address = self._sock.accept()
+        self._client.setblocking(1)
         context['remote_addr'] = ':'.join(str(v) for v in address)
         self.say('%(me)s: In session with %(remote_addr)s' % context)
         self._handle = sys.stdin = sys.stdout = self._client.makefile('rw')