jmoiron plays the blues

created gaspar, generic eventlet+zmq worker library

6ed7097d to gaspar fix gaspartests (nosetests -s works now) on vanilla eventlet of latest revision; the duplicate fds were related to sockets that had been closed but not yet shutdown; calling socket.shutdown() beforehand eliminates these problems

4d675283 to gaspar use makefile and read(n) instead of recv(n) in gaspar server, which prevents us from reading less than the desired buffersize and truncating messages

@jmoiron wrote a short benchmark of various #python serialization options: http://t.co/FRpeFVSj

9a2d0628 to gaspar fix some comments, put wait time back down to 0.05s

db4e74da to gaspar lower the slow-start sleep phase back to 0.05 from 0.1

3c5ce124 to gaspar make eventlet stfu about threads being bound and then read from different greenlets by making sure that we always bind and read in the same greenlet

6a764127 to gaspar clean up gaspartest, add gaspar.client so the tests run

cd19286b to gaspar do not import msgpack as it's not used and not a dep of gaspar (yet?)

c1d12622 to gaspar gaspar finally works!

0cef1a0b to gaspar eventlet prod/cons wip still; zmq error in producer recv from the pull socket, no idea why

00439c8c to gaspar Merge branch 'eventlet' Conflicts: gaspar/producers.py

fb85885f to gaspar sorta big workaround, using eventlet as the eventlet poc worked and gevent's didn't (there isn't much eventlet-specific stuff i'm using that i can't adapt to gevent), a lot more well thought out producer/consumer model, split forking from consumer and modify ForkTest, which passes again

@jmoiron They are quite proud: "By installing #Java, you will be able to experience the power of Java, by #Oracle." or "3 BILLION DEVICES RUN JAVA"