ZeroMq IRC Log

Tuesday February 1, 2011

[Time] Name	Message
[07:22] stockMQ	Hi.. I am using protobuf to PUB data.. at the SUB end i deserialize it to a protobuf message
[07:23] stockMQ	but the i need to copy the protobuf message to a struct and write the struct to a file..
[07:23] stockMQ	I need to do this because there are other applications consuming the file and they expect a struct
[07:24] stockMQ	right now i copy each field of the message to the corresponding field of the buffer
[07:24] stockMQ	Is this the right way or is there any other optimized way to go about with it
[08:20] mikko	stockMQ: are the other applications expecting a binary struct?
[08:20] mikko	stockMQ: sounds odd
[08:25] stockMQ	yes.. they too are custom applications built in house but to change them is to be avoided
[09:24] CIA-21	zeromq2: 03Martin Sustrik 07sub-forward * raddcc85 10/ devices/zmq_forwarder/zmq_forwarder.cpp :
[09:24] CIA-21	zeromq2: Forwarder device uses XPUB and XSUB sockets.
[09:24] CIA-21	zeromq2: Signed-off-by: Martin Sustrik <sustrik@250bpm.com> - http://bit.ly/eoHfSJ
[09:28] jugg	having a publisher that connects to an endpoint, how would I prevent messages from being queued if the endpoint has not yet been bound? Set the publisher HWM to zero, and call zmq_send(noblock)?
[09:32] sustrik	yes, use hwm
[09:33] jugg	hmm, I guess HWM = 1, as 0 is no limit...
[09:34] jugg	but that'll queue a single message, no?
[09:34] jugg	So a stale message will be delivered when the listener binds the endpoint.
[09:57] sustrik	it that pub/sub?
[10:06] ianbarber	potentially silly bug: if I destroy then recreate a sub socket with an identity too quickly I get an assert "Assertion failed: new_sndbuf > old_sndbuf (mailbox.cpp:182)" from the publisher
[10:06] ianbarber	this may not be something that is sensible to do, but i encountered it in a semi-real world usage (some prototype code)
[10:12] CIA-21	zeromq2: 03Martin Sustrik 07sub-forward * rca34d2a 10/ (5 files):
[10:12] CIA-21	zeromq2: Debug output for subscription forwarding.
[10:12] CIA-21	zeromq2: Signed-off-by: Martin Sustrik <sustrik@250bpm.com> - http://bit.ly/fdwCNa
[10:34] ianbarber	ah, it's because the publisher thinks it has two sockets with the same identity connected at the same time. It would be nice to have that cause a problem on the client side, rather than the server side, but I don't know if that's doable
[10:37] mikko	i dont think two identities should assert
[10:41] ianbarber	mikko: i haven't made a testcase in C yet, but in PHP
[10:42] ianbarber	https://gist.github.com/805688
[10:43] ianbarber	if anything was going to assert, i'd much rather it be the client
[10:43] ianbarber	but i think the best behavior would be a failed connect()
[10:44] ianbarber	not sure how easy it is to compare a new identity against connected identities on connect though, from the pub side
[10:47] ianbarber	ooh, if this works in c, this is worth a point in the 0mq game :)
[10:53] ianbarber	same thing in C: https://gist.github.com/805701
[10:57] ianbarber	yep, and if I split that into client server, it still happens
[11:02] CIA-21	zeromq2: 03Martin Sustrik 07sub-forward * rcc7bed5 10/ (src/trie.cpp src/trie.hpp src/xsub.cpp):
[11:02] CIA-21	zeromq2: Duplicate subscriptions are not forwarded by XSUB socket.
[11:02] CIA-21	zeromq2: Signed-off-by: Martin Sustrik <sustrik@250bpm.com> - http://bit.ly/hGPp1V
[11:24] suzan_shakya	Is it possible to have mulitple publisher with PUB/SUB ?
[11:39] stockMQ	Hi..what would be the equivalent of zmq_close (sender); zmq_term (context);
[11:39] stockMQ	in C++
[11:40] mikko	stockMQ: object going out of scope
[11:40] mikko	its called during destruction
[11:41] stockMQ	ok..So in order to say premptively stop the socket for some time can i just pause the thread for the duration
[15:40] stockMQ	Is there a way we can exit out of the blocking zmq_recv
[15:42] sustrik	send it a message
[15:44] stockMQ	The issue is I have a thread with basic while(1){zmq_recv }..Now in the Main UI thread I have a Disconnect button,clicking on which an event message is sent to the socket thread asking it to quit gracefully.But due to the blocking message of the socket thread that message is never read
[15:45] sustrik	you have to poll on both sockets
[15:45] sustrik	instead of recv()ing on one of them
[15:46] stockMQ	Ohk.. you mean i can have one tcp main socket and apart from that inprc sockets for communication between Main UI thread and child UI thread
[15:48] stockMQ	is that right
[15:50] sustrik	yes
[15:51] stockMQ	ok thanx
[17:46] andrewvc	any ideas what broken pipe, nbytes != 1 errors are caused by?
[17:50] sustrik	what's the assertion?
[18:28] CIA-21	jzmq: 03robin.palotai 07master * r04c2900 10/ src/Poller.cpp : testing patch for jni crashes - http://bit.ly/hlc8vI
[19:05] andrewvc	cremes: around?
[19:05] andrewvc	I was wondering if you noticed issues with pub/sub and ZMQ::NOBLOCK in the past
[19:05] cremes	andrewvc: yep
[19:06] cremes	nope, i have not noticed any specific issues with pub/sub
[19:06] andrewvc	hmmmm
[19:06] andrewvc	so in my work on em-zeromq, push/pull works well, pub/sub however
[19:06] cremes	what's it doing?
[19:06] andrewvc	zmq_recv just doesn't pick anything up
[19:06] cremes	or not doing?
[19:06] andrewvc	the FD gets triggered as readable
[19:06] andrewvc	on the sub
[19:06] cremes	so pub works but sub is busted?
[19:06] andrewvc	yeah
[19:06] andrewvc	i straced it
[19:06] andrewvc	i see the data getting written and read
[19:06] andrewvc	and ZMQ_EVENTS says POLLOUT is enabled
[19:06] andrewvc	errr triggered
[19:07] andrewvc	but recv doesn't work
[19:07] cremes	which socket are you talking about here? sub sockets don't care about POLLOUT
[19:07] andrewvc	errrm, I meant pollin
[19:07] cremes	they need POLLIN
[19:07] andrewvc	sorry
[19:07] cremes	ok
[19:07] andrewvc	wait, actually
[19:07] cremes	damn, thought we had the fix right there...
[19:08] andrewvc	i think they were triggering POLLOUT (according to ZMQ::EVENTS) on a ZMQSUB socket
[19:08] andrewvc	lemme double check that
[19:12] andrewvc	hmmm
[19:12] andrewvc	now I see no events on the socket, which is good and bad
[19:13] andrewvc	i mean, the same exact code works with pub/sub
[19:13] andrewvc	errr push/pull
[19:14] andrewvc	[pid 14619] sendto(22, "\10\0TMsg777", 9, 0, NULL, 0 <unfinished ...>
[19:14] andrewvc	[pid 14619] <... recvfrom resumed> "\10\0TMsg777", 8192, 0, NULL, NULL) = 9
[19:14] andrewvc	from the strace
[19:14] andrewvc	so, internally, the message gets sent/received
[19:14] andrewvc	by zeromq it looks
[19:15] andrewvc	cremes: any ideas on that? I can dig into ffi-rzmq tonight I guess
[19:15] cremes	andrewvc: can you gist your test code for me; i'd like to look at it
[19:17] andrewvc	sure, i mean, I'm a bit short on time now, about to head into a meeting. I can write a minimal test later
[19:17] andrewvc	however
[19:17] andrewvc	if you want you can look at
[19:17] andrewvc	however
[19:17] andrewvc	https://github.com/andrewvc/em-zeromq
[19:17] andrewvc	if you run this spec https://github.com/andrewvc/em-zeromq/blob/master/spec/em-zeromq_spec.rb
[19:17] cremes	i just wonder if i'll see something that you don't... hopefully it's an easy fix
[19:17] andrewvc	and s/ZMQ::PUSH/ZMQ::PUB
[19:17] andrewvc	and s/ZMQ::PULL/ZMQ::PUB
[19:17] andrewvc	you'll see it
[19:18] cremes	ok
[19:18] andrewvc	or I can send something later, I'll be expanding the test suite tonight
[19:52] CIA-21	zeromq2: 03Martin Sustrik 07sub-forward * ra40436b 10/ (5 files):
[19:52] CIA-21	zeromq2: Couple of bug fixes to subscription forwarding mechanism.
[19:52] CIA-21	zeromq2: Signed-off-by: Martin Sustrik <sustrik@250bpm.com> - http://bit.ly/eOAJxP
[20:14] mikko	good evening
[20:19] sustrik	evening