ZeroMq IRC Log

Sunday April 17, 2011

[Time] Name	Message
[07:26] rod	Hi, i'm new to 0mq so forgive me if my question has an obvious answer. What i'm trying to do use use a REQ socket to receive a request, then dispatch a job to worker threads which call back when they're done. But it seems with the REQ socket type you need to send a reply before you can receive another request, is this true? if so is there another solution for my problem? Thanks.
[07:27] Toba	you have to use REP socket for the server side (the one receiving requests) not a
[07:27] Toba	you use REQ on the side making the requests.
[07:27] Toba	this may be your first problem :)
[07:28] rod	ah yes sorry that's what i meant :D
[07:28] rod	(see, i am new to this!)
[07:28] pieterh	rod_: it's explained in the Guide
[07:28] Toba	I don't know a way to work around this, no.
[07:28] Toba	but, the Guide is all knowing, and eternal.
[07:28] pieterh	if you want to handle multiple requests in parallel, use a ROUTER socket
[07:29] pieterh	aka xrep
[07:29] pieterh	there is an example that does exactly what you want, asyncsrv
[07:29] rod	ah perfect. ok i'll give that a go, thanks pieterh!
[07:30] pieterh	http://zguide.zeromq.org/page:all#Asynchronous-Client-Server
[07:30] pieterh	however, before trying that, please do read & understand chapters 1 and 2
[07:32] rod	i've made it to the end of chapter 2 so far, but i'm trying to build some examples to help me learn (so might be jumping ahead too which is the problem here i think)
[08:26] guyvdb	Hi, I was listening to a video of Zed Shaw speaking from the zmq site. He stated something along the lines of do not use zmq over the internet. I.e. keep it within your lan for connecting server processes. Is this correct?
[08:26] guido_g	ahhh... salt already made it into the wiki
[08:27] guido_g	guyvdb_: sure, and you can always use a vpn
[08:28] pieterh	guyvdb_: 0MQ sockets tend to crash if you throw garbage at them
[08:29] guyvdb	hmm ... I have a bunch of embedded devices that connect via 3G that I want to keep track of. After Zeds comments my idea is asio between well known address (broker) and clients and then zmq in the back to worker processes. Does this make sense?
[08:31] pieterh	guyvdb_: asio isn't going to be more secure than 0MQ
[08:32] pieterh	well, perhaps it is... yes, this makes sense
[08:32] pieterh	if I was doing it over public Internet I'd use HTTP for the server
[08:32] guido_g	why?
[08:33] guyvdb	hmm I thought of that... but I want allways connected... I.e. server push to client...
[08:33] guido_g	i mean http is ok, but not the answer to everything
[08:33] pieterh	guido_g: several reasons, but mainly it's friendly for sysadmins (ports 80 and 443), has proven security (TLS), and has stacks in clients
[08:33] pieterh	guyvdb_: you can quite easily do server push using long polling
[08:34] guyvdb	would http long poll make sense? This seems like a bit of a hack when i could just use tcp with thrift or protocol buffer
[08:34] guido_g	long polling and mobiles devices isn't that clever, to get a reasonable reaction time you to wake up the device quite often
[08:34] guyvdb	The clients are embedded bits of hardware... no user interface, etc
[08:35] pieterh	guyvdb_: http long poll is nothing special, but you may need some intelligence in the server
[08:35] pieterh	i.e. to not disconnect clients if they're silent
[08:36] guyvdb	but would an ASIO server maintaining say 5000 connections with the clients not be better than long polling
[08:36] pieterh	guyvdb_: afaics it's precisely the same setup except that with HTTP you get a standard answer to parts of the problem
[08:36] guido_g	and find someone who did that to ask him how the carriers behave in this case
[08:36] guyvdb	I also want to implement heartbeat ala AMQP or something like that
[08:37] pieterh	you would want a friendly http server
[08:37] guido_g	and websockets :)
[08:37] guyvdb	any suggestions? mongrel 2? ngnix
[08:37] guyvdb	There is no web browser involved
[08:37] guido_g	mongrel2 is of course the first choice if you want to use Ã¸mq
[08:38] pieterh	guyvdb_: I'd suggest mongrel2, yes, mainly because you can probably tweak it if you need it
[08:38] pieterh	also, take a look at restms.org
[08:38] guido_g	guyvdb_: websockets is an hhtp protocol extension, no browser needed
[08:38] guido_g	guyvdb_: but is was more of a joke
[08:38] pieterh	guyvdb_: do you have security requirements?
[08:39] guyvdb	yes.. i want a pki infrastructure where i am CA and all clients must have valid certs
[08:39] pieterh	so using https gives you this basically for free
[08:39] guido_g	then use a good http kit
[08:39] guyvdb	yea... does mongrel2 use openssl?
[08:39] pieterh	after that you can use various techniques to switch to a bidirectional async protocol on top of http
[08:40] pieterh	guyvdb_: i assume it does
[08:40] guyvdb	hmm any insight into a bidirectional async protocol on http... have you seen any implementations / papers / blogs on this?
[08:40] pieterh	guyvdb_: the very simplest is to open two connections per client, one for reading, one for writing
[08:41] pieterh	reading uses long polling (see restms for some hints)
[08:41] pieterh	writing just posts stuff
[08:41] pieterh	a more efficient approach is to use the http protocol upgrade method
[08:41] pieterh	which lets you start a http session, then switch to your own binary async protocol afterwards
[08:42] pieterh	that requires a cooperative http server, obviously
[08:42] guyvdb	that sounds promising
[08:42] pieterh	i believe there are quite a few such stacks already built
[08:43] pieterh	you could look at http://bwtp.wikidot.com/ for inspiration
[08:43] pieterh	websockets is meant to standardize this area but it's stuck in endless ietf discussions
[08:44] pieterh	an example implementation of bwtp: http://code.google.com/p/serf/
[08:45] guyvdb	Ok... i think that is a lot for me to absorb right now... going to start with lots of reading :) Thanks pieterh you have been very helpful
[08:45] pieterh	guyvdb_: my advice would be to start with a minimal proof of concept using something like restms's transport layer (resttl)
[08:45] pieterh	and if you find http works for you, look at more efficient designs
[08:46] pieterh	let us know how you get on :)
[08:46] guyvdb	will do... thx
[10:25] guido_g	i will not implement this! https://twitter.com/#!/hintjens/status/59561500293083136
[10:28] headzone	good, we're relying on you to implement zeromq over BLOAT (RFC 3252)
[10:28] headzone	that clearly takes priority
[10:37] guido_g	:)
[11:34] wayneeseguin	Is there an example of using zeromq with a C program ?
[11:34] wayneeseguin	I am reading the docs but
[11:34] wayneeseguin	would like a full example to look at
[11:39] wayneeseguin	Perhaps I can glean this from mongrel2...
[11:43] guido_g	wayneeseguin: see the guide
[11:43] guido_g	full of c it is
[11:44] wayneeseguin	guido_g: I've ben looking at http://libzapi.zeromq.org/
[11:45] wayneeseguin	guido_g: But I was looking for a simple example of usage, is there another page with that? or is that not the guidee
[11:47] headzone	wayneeseguin: http://zguide.zeromq.org/
[11:47] headzone	every example is available in c
[11:52] wayneeseguin	awesome
[11:52] wayneeseguin	thank you both
[11:52] wayneeseguin	I was simply looing at the wrong pages :)
[17:14] seangrove	Hey all, what are unnamed sessions, and why would an assertion about them be failing?
[17:18] mikko	seangrove: what is the exact assertion?
[17:19] seangrove	Assertion failed: unnamed_sessions.empty () (socket_base.cpp:507)
[17:20] seangrove	So, I'm probably mixing up a few pieces here
[17:20] seangrove	But trying to get my head around some of the vocabulary
[17:24] seangrove	hmm
[17:24] seangrove	So this seems to happen right around forking
[17:25] seangrove	If I call ZMQ::Context.new(1) after forking, or don't call it at all, I get the assertion failure and a segfault
[17:25] seangrove	If I call ZMQ::Context.new(1) before forking, I only get the assertion failure
[17:28] seangrove	mikko: Any suggestions on where to read up to understand this?
[17:31] mikko	you should have one context per process
[17:32] mikko	i don't think context survices fork
[17:32] mikko	what language is this?
[17:32] seangrove	Ruby
[17:33] seangrove	I know the one-context-per-process
[17:33] seangrove	Should I create a new context in the parent pre-fork, the child-post-fork, or the parent-post-fork?
[17:33] seangrove	I believe the guide says in the parent pre-fork
[17:35] mikko	i think in each child after fork
[17:35] seangrove	Hmm, that gets me a segfault
[17:36] seangrove	I wonder if it's a library version problem
[17:36] seangrove	Certainly possible
[17:37] seangrove	And is that regardless of whether the child process uses anythign zmq-related?
[17:37] seangrove	None of my child processes do anything with zmq
[17:53] sustrik	seangrove: if you create new context in the child if should work ok
[17:53] sustrik	if it does not, it's a bug
[17:53] sustrik	please, do report it
[17:56] seangrove	sustrik: Ok, sounds good
[17:56] seangrove	Thanks for the tip
[17:56] seangrove	I need to refactor zmq out of the frontend of my app right now anyway
[17:57] seangrove	Put it where it really belongs