[Time] Name | Message |
[00:23] count
|
So, I'm using a zmq.QUEUE in python
|
[00:23] count
|
And everything is fantastic and full of wonderfullness
|
[00:23] count
|
except when all of my REP sockets quit
|
[00:23] count
|
the queue stalls
|
[00:24] count
|
(when they join back up, it's still stuck waiting for a recieve from one that died without talking)
|
[00:24] count
|
the REQ does:
|
[00:24] count
|
socket.send("blah")
|
[00:24] count
|
msg = socket.recv()
|
[00:24] count
|
(in a loop)
|
[00:25] count
|
How do I handle the disconnected clients properly?
|
[00:25] count
|
ie: that socket.recv() for that particular message is gone now
|
[00:25] count
|
I catch termination of the interpreter in the REP socket, and issue a socket.close/context.term
|
[00:25] count
|
but that doesn't seem to help
|
[00:26] count
|
Is there a way to 'flush' the REP ?
|
[03:33] count
|
hello?
|
[03:33] jsimmons
|
hi
|
[03:36] count
|
hm, nevermind, I thinkn I need to reverse my req/rep
|
[03:39] count
|
YAYA that fixed it
|
[03:39] count
|
I was htinking about things backwards
|
[03:39] count
|
0mq hurts my brain
|
[04:33] count
|
anybody got a minute to help with a wierd problem with a matched pair of req/rep sockets?
|
[04:33] count
|
I'm losing the very last message
|
[04:33] count
|
er, the response for the very last message
|
[04:38] count
|
nm, figured that out !
|
[05:33] sustrik
|
jefferai: which manpage is that?
|
[05:41] CIA-32
|
libzmq: 03Martin Sustrik 07master * r9fb9fea 10/ (src/tcp_connecter.cpp src/xrep.cpp): Improve error reporting in a minor way ...
|
[07:03] CIA-32
|
libzmq: 03Pieter Hintjens 07master * r4138aca 10/ (6 files): Fixed doc to clarify how label parts work ...
|
[08:20] CIA-32
|
libzmq: 03Martin Sustrik 07master * r193fa1c 10/ src/tcp_address.cpp : Accept square brackets around IPv6 address ...
|
[12:09] mikko
|
sustrik: there?
|
[12:10] sustrik
|
mikko: hi
|
[12:10] mikko
|
http://valokuva.org/~mikko/poller.patch
|
[12:10] mikko
|
can you test that when you got time?
|
[12:10] mikko
|
it's the poller selection
|
[12:10] sustrik
|
sure
|
[12:10] mikko
|
you can do --with-poller=something to force polling system
|
[12:10] mikko
|
or let it detect automatically
|
[12:11] sustrik
|
ok, i'll test it with select, poll and epoll
|
[12:11] mikko
|
i wonder if the --with-poller=xyz should also check if that polling system actually works
|
[12:11] mikko
|
currently it just sets -DZMQ_FORCE_XYZ
|
[12:12] mikko
|
and assumes that user knows what they are doing
|
[12:12] mikko
|
so on linux --with-poller=kqueue will result to build time error
|
[12:14] sustrik
|
that's ok, i would say
|
[12:15] sustrik
|
it does exactly what the user asked for
|
[12:19] mikko
|
seems to work ok on linux and mac
|
[12:20] mikko
|
kqueue detected on mac and epoll on linux
|
[12:23] sustrik
|
nice
|
[12:39] sustrik
|
mikko: the test programs are actually run or just compiled?
|
[12:40] mikko
|
linked
|
[12:45] sustrik
|
ok
|
[12:58] sustrik
|
mikko: looks like it's working ok
|
[12:58] sustrik
|
can you send it as a signed-off patch to the mailing list?
|
[17:47] jefferai
|
sustrik: it's the zmq-ipc page at http://api.zeromq.org/2-1-1:zmq-ipc
|
[21:03] mikko
|
pieterh: here?
|
[21:03] mikko
|
have you broken 2-1 and 2-2 windows builds lately?
|
[21:04] mikko
|
ip.cpp: In function 'int zmq::resolve_local_path(sockaddr_storage*, socklen_t*, const char*)':
|
[21:04] mikko
|
ip.cpp:353:13: error: 'EPROTONOSUPPORT' was not declared in this scope
|
[21:49] ahbritto
|
When using SUB and setting a filter, what is it filtering on?
|
[21:51] ahbritto
|
Never mind. :)
|
[21:52] mikko
|
ahbritto: it's effectively a prefix match
|
[21:52] ahbritto
|
mikko: Thanks, I refound the section on message envelopes.
|
[21:55] ahbritto
|
If I want to have multiple things publishing and multiple things subscribing, I must have a stable device in the middle?
|
[22:02] mikko
|
ahbritto: that would be the best way
|
[22:02] mikko
|
you can also have all subscribers connect to all publishers
|
[22:02] mikko
|
or use pgm
|
[22:03] mikko
|
but usually device is the least painful way
|
[22:03] ahbritto
|
The doc is a little unclear. I assume multiple publishers can't bind to the same address?
|
[22:04] mikko
|
no, they can't
|
[22:04] ahbritto
|
Thank you. :)
|