librelist archives

« back to archive

IO Refactor Mostly Done, Please Test

IO Refactor Mostly Done, Please Test

From:
Zed A. Shaw
Date:
2010-10-06 @ 09:19
Alright everyone, I've redone the IO to use this new IOBuf thing.  It's
basically a fancy psuedo-ring buffer that lets us do all the crazy
reading and parsing we need to handle the HTTP.  The code in the fossil
should mostly work, except chrome seemed to really freak out last I
tested.  There will also probably be odd bugs in proxying and SSL is
broken but it should mostly work.

Please test it out and report back to me your successes or failures.
The XML streaming and json message streams should now work reliably
though, and we'll now have a single place to go to improve how IO is
handled.

If you want to see the work, it's in src/io.c

-- 
Zed A. Shaw
http://zedshaw.com/

Re: [mongrel2] IO Refactor Mostly Done, Please Test

From:
snacktime
Date:
2010-10-06 @ 19:53
>
> Please test it out and report back to me your successes or failures.
> The XML streaming and json message streams should now work reliably
> though, and we'll now have a single place to go to improve how IO is
> handled.

So far all the issues I had with the json messages have gone away.

Thanks Zed!

Chris

Re: [mongrel2] IO Refactor Mostly Done, Please Test

From:
Zed A. Shaw
Date:
2010-10-06 @ 22:09
On Wed, Oct 06, 2010 at 12:53:24PM -0700, snacktime wrote:
> >
> > Please test it out and report back to me your successes or failures.
> > The XML streaming and json message streams should now work reliably
> > though, and we'll now have a single place to go to improve how IO is
> > handled.
> 
> So far all the issues I had with the json messages have gone away.

They mostly work, but there's a couple bugs I'm rooting out.  If you
start seeing bizarre responses with disconnect messages sprinkled in
them then you hit the bug.

-- 
Zed A. Shaw
http://zedshaw.com/

Re: [mongrel2] IO Refactor Mostly Done, Please Test

From:
snacktime
Date:
2010-10-06 @ 23:06
On Wed, Oct 6, 2010 at 3:09 PM, Zed A. Shaw <zedshaw@zedshaw.com> wrote:

> They mostly work, but there's a couple bugs I'm rooting out.  If you
> start seeing bizarre responses with disconnect messages sprinkled in
> them then you hit the bug.


Found something.strange. If I create enough connections in the
handler, when I kill the handler mongrel2 just exits and I see this in
dmesg output:

[2524949.457015] mongrel2[19704]: segfault at 10 ip 00007f5d3bb55754
sp 0000000001c69670 error 4 in libzmq.so.0.0.0[7f5d3bb22000+41000]

Everything works fine up until I kill the handler.  No errors in the
mongrel2 logs.

Chris

Re: [mongrel2] IO Refactor Mostly Done, Please Test

From:
Zed A. Shaw
Date:
2010-10-06 @ 23:38
On Wed, Oct 06, 2010 at 04:06:19PM -0700, snacktime wrote:
> On Wed, Oct 6, 2010 at 3:09 PM, Zed A. Shaw <zedshaw@zedshaw.com> wrote:
> 
> > They mostly work, but there's a couple bugs I'm rooting out.  If you
> > start seeing bizarre responses with disconnect messages sprinkled in
> > them then you hit the bug.
> 
> 
> Found something.strange. If I create enough connections in the
> handler, when I kill the handler mongrel2 just exits and I see this in
> dmesg output:

Nice, yep, that's a problem with how 0mq does it's shutdown.  We're
switching that up so it's more graceful.

-- 
Zed A. Shaw
http://zedshaw.com/