The two span receivers are from separate processes. Basically, the test
consists of a client and server. Both the client and the server create
their own tracer and span receiver object. They're using zeromq to exchange
messages for a fixed number of times.
The problem is, either the server or client hangs during socket_t::send (or
at least it seems so), when both are using their own htraced span receiver.
I've tried what you suggested, using two separate htraced daemons for each
span receiver, and the same problem is still present.
I'm starting to think this is just some problem with ssh, or maybe even the
remote server itself.
I'll try again on my local machine and ask again if the problem persists.
On Tue, Oct 11, 2016 at 1:13 PM, Colin McCabe <cmcc...@apache.org> wrote:
> Hi Jason,
> Hmm, that's a bummer. It sounds like a bug either in htrace-c or
> htrace-htraced. Were the two htrace-c span receivers in the same
> process, or in separate processes? Can you give a little more detail
> about the test and where it is hanging?
> One easy experiment to try is to set up two separate htraced daemons on
> different ports, and target the separate span receivers at distinct
> htraced instances. I think htraced is unlikely to be the problem,
> though, since we've tested it accepting spans from a whole Hadoop
> cluster. My guess would be that it's probably something in htrace-c or
> maybe in the test.
> On Tue, Oct 11, 2016, at 06:00, Jason Jeong wrote:
> > Hi,
> > I'm trying out HTrace for some experiments and having some problems.
> > I'd greatly appreciate any help!
> > Would there be any reason for two separate htraced span receivers (for
> > the
> > C client) on the same machine to interfere with each other, even if they
> > were started from different processes?
> > I expected that the htraced span receivers wouldn't affect each other,
> > but
> > apparently either one side seems to hang somewhere in the code and my
> > test
> > program fails to end.
> > I'm pretty sure there's no problem with my tests, since when I use the
> > local file span receiver or just no receivers then the test runs as
> > expected. The htraced daemon also seems to run well, too.
> > Thanks,
> > Jason