Monday January 17, 2011

[Time] NameMessage
[05:18] andrewvc cremes: around?
[06:30] yrashk hmm, after reinstalling osx I am getting absolutely reproducible Bad file descriptor nbytes != -1 (mailbox.cpp:241)
[06:30] yrashk :-\
[07:36] sustrik yrashk: please do report the bug
[07:36] yrashk I need to narrow the scope first
[07:37] yrashk but this is getting interesting
[07:37] sustrik is that with shutdown_stress test?
[07:37] yrashk although I still hope it's my bug
[07:37] yrashk no
[07:37] sustrik ok
[10:33] mikko howdy
[10:46] Evet fine
[11:10] ivan-korobkov Hello. I need to access a sub socket from another thread in Java. Is it enough to synchronize this method?
[11:14] mikko ivan-korobkov: not sure how java handles this
[11:14] mikko ivan-korobkov: i tend to use inproc pipe between the threads in such a scenario
[11:16] ivan-korobkov Ok, thanks.
[11:55] mikko mingw is being a pain
[11:58] mikko sustrik_: i pushed the new tests into a branch in github if you want to take a look
[11:58] mikko test_identity is failing atm
[12:57] lestrrat if you're specifying a char * buffer to zmq_msg_init_data(...), do you need to give it a zmq_free_fn ?
[13:11] Steve-o it takes a zero-copy reference, doesn't matter what the target is
[13:15] stimpie somebody knows an key value pair serialization library?
[13:18] Steve-o google protobufs
[13:18] Steve-o thrift
[13:18] Steve-o xdr
[13:18] Steve-o json, yaml, etc
[13:22] stimpie thanks for the pointers
[13:23] sustrik lestrrat: free_fn is needed only when you pass your buffer to 0mq and you want it to deallocate it once it's sent
[13:23] lestrrat thanks steve, sustrik
[13:26] Evet oh, zeromq built by imatix?
[13:28] Steve-o Evet: they're trying to steer to full flossy project, like apache
[13:29] Evet i have learnt perl in xitami more than ten years ago
[13:30] Steve-o sustrik: guido_g: any feedback on the pgm_receiver assert?
[13:32] sustrik i haven't tried it myself
[13:32] sustrik what about the patch you proposed?
[13:32] sustrik (removing the timer)
[13:32] sustrik it seems that it should be patched anyway, no?
[13:34] Steve-o looks like it
[13:35] sustrik would you submit a patch?
[13:35] sustrik or should I fix it?
[13:35] Steve-o there's one on the list replying to Guido
[13:35] sustrik let me see
[13:37] sustrik ok, got it
[13:37] sustrik when sending a patch use [PATCH] in subject
[13:37] Steve-o I was waiting for Guido to catch it first
[13:37] sustrik i see
[13:38] Steve-o I've only got my Sparc up in the office presently, I can't test too much
[13:38] sustrik but this is something that should be done anyway, right?
[13:38] sustrik irrespective whether it solves guido's problem or not
[13:38] Steve-o well it looks right
[13:39] Steve-o so go ahead
[13:39] sustrik ok, let me apply it
[13:39] Steve-o for autoconf, waiting for any further feedback from mikko
[13:40] sustrik ack
[13:42] Steve-o I'm sure the devs using OSX will shout soon enough as master should be broken for them currently
[13:42] mikko Steve-o: i will start testing today
[13:42] mikko had suprisingly busy weekend
[13:43] mikko what is the current view regarding openpgm and zeromq?
[13:43] Steve-o no worries, I'm mid buying a house
[13:43] mikko do we bundle a version with zeromq2 and allow people to specify directory as well?
[13:44] Steve-o sounds good
[13:44] Steve-o I actually have autoconf making a shared library, but not really recommended
[13:44] mikko so by default the value of --with-pgm is foreign/...
[13:44] Steve-o came free with libtool I think :D
[13:44] mikko but you can specify --with-pgm=/opt/xyz
[13:45] mikko does that sound like a sensible approach?
[13:45] Steve-o sounds good, then distros can choose their packaging preference
[13:45] mikko yeah, distros don't like static linking between two projects they package
[13:46] CIA-21 zeromq2: 03Steven McCoy 07master * r667b247 10/ src/pgm_receiver.cpp :
[13:46] CIA-21 zeromq2: Cancel pending timers when decoder for PGM stalls
[13:46] CIA-21 zeromq2: Otherwise the timer event can raise an input event
[13:46] CIA-21 zeromq2: causing assertion on unprocessed data.
[13:46] CIA-21 zeromq2: Signed-off-by: Steven McCoy <> -
[13:47] Steve-o once I start to see packages created I can copy the build script and then zeromq can be split up
[13:48] Steve-o The only future work I see is on FEC unless some University project picks up congestion control
[13:49] Steve-o the congestion protocol is broken above 10,000 packets-per-second
[13:49] guido_g Steve-o: i used a very high HWM on the subscriber as proposed and it worked
[13:50] Steve-o great, otherwise you would get application induced dataloss
[13:51] CIA-21 zeromq2: 03Martin Sustrik 07master * rb262f2f 10/ doc/zmq_send.txt :
[13:51] CIA-21 zeromq2: Typo fixed in zmq_send(3) man page.
[13:51] CIA-21 zeromq2: Signed-off-by: Martin Sustrik <> -
[13:52] Steve-o definitely need to work on some feedback for admins around that issue in the future
[13:53] Steve-o but then that's the logging system which is rather awkward
[14:11] sustrik funny video here btw
[14:11] sustrik
[14:12] sustrik using 0mq to create a cloud of 1978 vintage computers :)
[14:17] stimpie Iam doing some performanc benchmarks (with the provided perf code) on 1000G ethernet I get 936 Mb/s on infiband (32 Gb/s) this only goes up to 2070 Mb/s
[14:17] stimpie is this an expected resulted or should this be higher?
[14:25] sustrik stimpie: what's your message size?
[14:27] sustrik there are some test results here:
[14:28] stimpie message size used is 100 bytes
[14:29] stimpie my result is similar to those in your link
[14:29] sustrik yes, looks like
[16:55] Guest10403 Hi There, I am just beginning to use ZeroMQ and am experimenting with some of the samples. I am using the C# language bindings and have run into an issue with getting a DllNotFound Exception
[16:56] Guest10403 Has anyone else run into this issue?
[16:56] mikko Guest10403: dll not found sounds like the libzmq.dll can not be found
[16:56] mikko is it in a path where the loader can find it?
[16:57] Guest10403 I can Load on My development machine, but when I move things to other servers, I get the error despite copying the libzmq.dll into both the folder the samples are running from and into Windows/System32
[16:57] Guest10403 Hi Mikko
[16:57] Guest10403 Thanks for your response
[16:57] Guest10403 Windows should be able to find it, but does not seem to be able to
[16:58] mikko hmm
[16:58] mikko not an ACL issue?
[16:59] Guest10403 I have put clzmq.dll into the Folder the sample app is in and also libzmq.dll Also, i put libzmq into c:\Windows/System32
[17:00] Guest10403 I don't understand your question ACL?
[17:00] Guest10403 What does ACL stand for?
[17:00] mikko i mean file permissions on either one of the dll diles
[17:00] mikko files*
[17:00] mikko acl = access control list
[17:00] Guest10403 Good question
[17:01] Guest10403 It should not be that sort of issue, the user I am running as on the server has administrative rights
[17:01] Guest10403 I'll check again
[17:02] Guest10403 I have admin rights and am Running on Windows Server 2003 Service pack2
[17:03] mikko maybe send an email to mailing-lists
[17:03] mikko i dont really run windows anywhere so i cant help much
[17:04] Guest10403 Thanks for the suggestion
[23:05] mikko lestrrat: looks like perl build is failing for some reason
[23:06] lestrrat whoa
[23:08] lestrrat it says build timed out
[23:08] lestrrat does hudson have a timeout for tests?
[23:26] mikko lestrrat: yes
[23:26] mikko lestrrat: it will eventually time out
[23:26] lestrrat what's the timeout?
[23:27] mikko let me check
[23:27] mikko five minutes
[23:27] mikko is that enough for the build?
[23:27] lestrrat hmmm
[23:27] lestrrat should be
[23:27] lestrrat I added new tests that take up some time, but I'm pretty sure it doesn't go over 5 minutes. odd
[23:27] mikko looking at the trends the time is normally around <2min
[23:28] lestrrat I'll make the tests shorter for now, and see what happens
[23:28] mikko i can increase the build timeout as well
[23:28] mikko just to see if it passes with more time
[23:28] lestrrat well, let me change this first, and if it still doesn't work, I might ask you to change the timeout
[23:29] mikko
[23:29] mikko you can see build time trend from there
[23:29] lestrrat oh man. didn't know about that in hudson. that's nice :)
[23:30] mikko all kind of cool toys in hudson
[23:30] mikko like the static analysis plugins:
[23:34] lestrrat well, as of my latest commit, the test should run in about 12 seconds :/
[23:34] lestrrat so if the hudson runs were taking on average 1 min+, presumably most of that time is spent checking out git
[23:37] lestrrat well let's see what the next run brings up
[23:39] mikko pass
[23:39] lestrrat so soon? lol
[23:39] lestrrat thanks!
[23:39] mikko it will run quicker when there is no load
[23:40] mikko usually the build runs alongside with 4 other builds
[23:40] mikko so it takes longer