Subject: Re: perlbal(reverse proxy role) timing out for slow 'reproxy' backends - msg#00019

Thanks for the test case!

I won't have time to look into this for a while, so if someone else on the list can help out, please do!

-Dormando

Hemant Bist wrote:
Hi, I am able to reproduce this with a simple cgi script that keeps on writing to stdout. (Attached). The error happens within 10 sec of the max_idle_timeout. I see the following messages when I set the DEBUG env variables.

Backend Perlbal::BackendHTTP=ARRAY(0x87a13a4) is readable!
write(Perlbal::ClientProxy=ARRAY(0x87a0a08), <886>"
Some Random dat...") from (Perlbal::BackendHTTP, /usr/share/perl5/Perlbal/BackendHTTP.pm, 562)
Client (Perlbal::ClientProxy=ARRAY(0x87a0a08)) closing backend (Perlbal::BackendHTTP=ARRAY(0x87a13a4))


====Sample run with timeout set to 60 sec with perbal running on port 8090=======
time wget http://192.168.1.52:8090/blodio/forever_internal_redirect.modperl
--14:03:17-- http://192.168.1.52:8090/blodio/forever_internal_redirect.modperl <http://192.168.1.52:8090/blodio/forever_internal_redirect.modperl>
=> `forever_internal_redirect.modperl.1'
Connecting to 192.168.1.52:8090 <http://192.168.1.52:8090>... connected.
HTTP request sent, awaiting response... 200 OK
Length: unspecified [text/plain]

[ <=> ] 55,818 771.43B/s
14:04:24 (856.85 B/s) - `forever_internal_redirect.modperl.1' saved [55818]


real 1m7.116s
user 0m0.012s
sys 0m0.012s

HB

On 9/17/07, *dormando* <dormando@xxxxxxxxx <mailto:dormando@xxxxxxxxx>> wrote:

Okay,

I missed it on my first read through that area. Adjusting that variable
should fix it for you. The latest SVN will work better since it's
configurable.

BackendHTTP.pm does the "client writing" of a response. Thinking if it'd
still be desired behavior to ensure the client idle's been updated.

Something else might still be buggy here, since I know I've done
transfers to a client which take much longer? Can you try testing by
reproxying a lot more data with a low timeout? Ten or thirty times more
data than your CGI's presently sending.

-Dormando

Hemant Bist wrote:
> Hi Dormando,
>
> Really appreciate your reply.
>
> In this case, the 'slow server' is a cgi script, so I guess the
> connection from Perlbal to the slow server will not have http
> Connection:keep-alive header.
> And the cgi script does start sending the data within first few
seconds
> and keeps on sending the data.
>
> If you think I should NOT be receiving the timeout, please do let me
> know. I will try to debug it more.
>
> On my system, I could reproduce/fix the problem repeatedly by
changing
> max_idle_time(where you mentioned) and restarting perlbal.(Also
if I hit
> the url to the slow server directly taking out the perlbal from the
> equation, things work fine).
>
>
> [ The reason I updated the max_idle time because I saw 'keep_alive'
> parameter is only updated in the ClientHTTPBase::event_write.
> So my "theory" was that perlbal will kill the connection to
backend slow
> proxy after 30 sec through Socket::_do_cleanup() function.
> event_write will not be called because Perbal will have nothing
to write
> to the slow proxy after it has sent the get request.
>
> ]
>
>
>
> Thanx,
> HB
>