A hacky attempt to fix the RPC starvation problems I encountered with the NDB eventloops.

gistfile1.py

# This attempts to force a FIFO ordering on RPC execution in the eventloop (as opposed to random RPC execution)
RPC_ORDERING = True
if RPC_ORDERING:
	import collections
	from google.appengine.ext.ndb import eventloop

	class OrderedEventLoop(eventloop.EventLoop):
		def __init__(self):
			super(OrderedEventLoop, self).__init__()
			self.rpcs = collections.OrderedDict()
	eventloop.EventLoop = OrderedEventLoop

	from google.appengine.api import apiproxy_stub_map
	from google.appengine.api import apiproxy_rpc
	@classmethod
	def __check_one(cls, rpcs):
		rpc = None
		logging.debug('Pending RPCs: %s', rpcs)
		for rpc in rpcs:
			assert isinstance(rpc, cls), repr(rpc)
			# Original:
			# state = rpc.__rpc.state
			# Revised:
			state = rpc.state
			if state == apiproxy_rpc.RPC.FINISHING:
				# Original:
				# rpc.__call_user_callback()
				# Revised:
				rpc.wait()
				logging.debug('Found finished RPC: %s', rpc)
				return rpc, None
			assert state != apiproxy_rpc.RPC.IDLE, repr(rpc)
		logging.debug('No finished RPCs, going to wait for first RPC: %s', rpcs[0])
		# This is the important changed line in this whole function:
		# We want to return the oldest RPC to execute, not a random one.
		return None, rpcs[0]
	assert hasattr(apiproxy_stub_map.UserRPC, '_UserRPC__check_one')
	apiproxy_stub_map.UserRPC._UserRPC__check_one = __check_one