ZeroMq IRC Log

Thursday July 28, 2011

[Time] Name	Message
[02:21] kimmy	hi
[02:22] kimmy	how can i solve the problem in java when i use req/rep model, how can i set timeout when the network is bad
[02:28] kimmy	How to troubleshoot network connection is unavailable RESV timeout problem, I use the Java language. thank you
[06:25] CIA-32	libzmq: 03Martin Sustrik 07vtcp * rce7a9a5 10/ (5 files): Setting TCP socket options moved to tcp_engine_t ...
[07:05] CIA-32	libzmq: 03Martin Sustrik 07vtcp * r6e98742 10/ (src/tcp_connecter.cpp src/vtcp_connecter.cpp): Precise reconnect interval randomised ...
[11:21] CIA-32	libzmq: 03Martin Sustrik 07vtcp * r5ac6314 10/ (11 files): Implementations of TCP and IPC transports separated ...
[11:47] CIA-32	libzmq: 03Martin Sustrik 07vtcp * r46b053b 10/ (6 files): Dead code removed from TCP and IPC transports ...
[11:59] pieter_hintjens	Hi folks, just released 2.1.8
[11:59] pieter_hintjens	anyone able to give positive/negative feedback on the package, that'd be most welcome
[12:11] drbobbeaty	pieter_hintjens: I'm interested... 2.1.7 has been OK for me for a while, but we have a nasty multi-part message problem in another group that might be interested in trying 2.1.8 out. I'll let you know.
[12:12] pieter_hintjens	drbobbeaty: if it was one of the asserts listed in NEWS, it should be fixed, but otherwise not
[12:13] drbobbeaty	pieter_hintjens: OK, thanks... I'll pass that along.
[12:13] pieter_hintjens	ack
[12:13] pieter_hintjens	also, if you need help debugging & fixing issues in zeromq, give a shout, we're here
[12:14] drbobbeaty	Thanks.
[13:08] cremes	pieter_hintjens: i'm available to try to repro that PUSH/NOBLOCK issue on windows but
[13:08] cremes	i haven't been able to get it to compile :(
[13:08] pieter_hintjens	ah, of course
[13:08] pieter_hintjens	do you have czmq installed?
[13:08] cremes	i do not
[13:08] pieter_hintjens	simplest then is to split into two programs, client and server
[13:09] cremes	but i'm sure i can dload it and compile it like i did for 2.1.7
[13:09] cremes	yes, i already did that split
[13:09] cremes	and removed the pthread stuff
[13:09] cremes	now gcc bitches out the "-lzmq" and that it can't find the zmq library
[13:09] cremes	i'm out of my element with windows and mingw32
[13:09] pieter_hintjens	sigh, yeah
[13:10] pieter_hintjens	how did you test it yesterday?
[13:10] mikko	cremes: pass in the path to the lib
[13:10] cremes	i used my ruby version... the ffi_lib can find and load zmq without issue
[13:10] mikko	./configure --with-zmq=/path/to
[13:11] cremes	mikko: i have libzmq properly compiled with mingw32
[13:11] cremes	mikko: my issue is with compiling a small C program that can link to it
[13:11] mikko	cremes: ok, use -L flag for the path to the library
[13:11] mikko	and -I for includes
[13:11] cremes	ok, give me a sec...
[13:12] mikko	well, -L as a path the directory where the library is
[13:13] cremes	mikko: undefined reference to 'sleep'
[13:13] cremes	otherwise it got way further than usual
[13:14] mikko	there is no sleep on win
[13:14] cremes	ok, i'll make it wait for a keypress...
[13:14] mikko	you can use Sleep
[13:14] CIA-32	libzmq: 03Martin Sustrik 07vtcp * rb45fec3 10/ (6 files): Tuning of TCP sockets is done at a single place ...
[13:15] mikko	http://msdn.microsoft.com/en-us/library/ms686298%28v=vs.85%29.aspx
[13:15] mikko	it's in milliseconds rather than seconds
[13:16] sustrik	there's also zmq_sleep() convenience function
[13:16] sustrik	#include <zmq_utils.h>
[13:18] cremes	success!
[13:19] cremes	shit
[13:20] cremes	this doesn't repro what i am seeing...
[13:20] mikko	"it compiles.. oh wait, what does segmentation fault mean?"
[13:20] cremes	heh
[13:21] cremes	sending with NOBLOCK in a tight loop gets 99.9% EAGAIN and only a few (235) messages get through
[13:21] cremes	same as linux & osx
[13:21] cremes	which is correct
[13:22] cremes	so, i am left to assume that this is exposing a bug in the ruby bindings that only shows itself on windows
[13:22] cremes	as soon as i can confirm, i'll close the issue
[13:22] pieter_hintjens	cremes: nice
[14:01] cremes	pieter_hintjens: found and fixed my bug
[14:01] cremes	pieter_hintjens: it's a good thing you "force" people to write repros in C :)
[14:01] pieter_hintjens	great!
[14:01] pieterh	minimal C repros are the key to happiness
[14:02] cremes	i agree wholeheartedly
[14:02] pieterh	that, and a swimming pool and a convertible
[14:02] cremes	and a tall glass of beer
[14:02] cremes	and a short glass of whisky
[14:02] pieterh	but apart from all those, minimal C repros
[14:02] cremes	:)
[14:02] cremes	i will close the issue
[14:15] cremes	sustrik, pieterh: i don't have permission to close issues in jira; please close issue 235 (not a bug in libzmq)
[14:15] sustrik	wait a sec
[14:16] sustrik	done
[14:33] CIA-32	libzmq: 03Martin Sustrik 07vtcp * rf63db00 10/ (4 files): Different listener implementations simplified ...
[17:05] ronny	hi
[17:05] cremes	hi
[17:06] ronny	any news wrt security
[17:06] cremes	i saw some discussions on the mailing list... i don't really track that issue (it's not important to me)
[19:13] technoweenie	hey guys, has anyone had issues compiling zeromq on lion? i used to install it through homebrew on snow leopard with no probs
[19:14] technoweenie	v2.1.7
[19:19] technoweenie	well nevermind then! compiled the 2.1.8 tarball with no problems.
[20:43] mdebreceni	I have a quick question about fanout and ZMQ_PUB
[20:43] mdebreceni	I mean ZMQ_REQ
[20:44] mdebreceni	when a ZMQ_REQ socket is connected to multiple ZMQ_REP endpoints, I know that requests will load-share among the ZMQ_REQ endpoints
[20:44] mdebreceni	If one of those endpoints goes down, is there a way to avoid using it until it reconnects?
[20:48] mdebreceni	We've had good results if the server(s) call zmq_connect() and the client calls zmq_bind() - the client load shares only across the servers that have connected
[20:48] mdebreceni	If the client is the one that calls zmq_connect() though, we get stuck as soon as our zmq_send hits a dead server
[20:50] mdebreceni	We've had good results with DEALER/ROUTER and setting ZMQ_HWM to something low (e.g. 1) - in that case, we will still queue up one request for each dead server, but subsequent requests are successful
[20:52] mdebreceni	(client uses DEALER, server uses ROUTER)
[22:02] korch	is 2.2 "stable" for prod use? or should i stick with 2.1?
[22:17] cremes	korch: use 2.1.8; it's the latest stable release for production and just came out today
[23:33] CIA-32	pyzmq: 03MinRK 07master * rb1cefd1 10/ (7 files in 3 dirs): update tests for 3.0 ...
[23:33] CIA-32	pyzmq: 03MinRK 07master * r8ed4b8b 10/ (31 files in 7 dirs): update author/copyright info to be more accurate ...