[squid-users] accelerating with 2.6RC1

From: Bryan Richter <bryanr@dont-contact.us>
Date: Wed, 21 Jun 2006 15:38:28 -0700

Hi everyone,

We just recently decided to use Squid to accelerate virtual servers hosted
on multiple real servers. Since it sounds like version 2.6 has made
improvements regarding acceleration, I downloaded and installed it and I've
been trying to make it work.

I was able to accelerate a single host using 2.5 (which came with Debian
stable, the OS running the machine I've put Squid on). But I can't seem to
replicate my success with 2.6.

The simplest test case I could think of is accelerating a single virtual
server. To do that, I tried this config (diffed against config.default and
anonymized):

    66c66
    < http_port 3128
    ---
> #http_port 3128
    156a157,160
> https_port 1.1.1.1:443 \
> cert=/etc/ssl/certs/cert.pem \
> key=/etc/ssl/private/key.pem \
> defaultsite=accelerated.host protocol=http
    1366a1371
> url_rewrite_program /tmp/squid-rewriter.sh
    2461c2466,2467
    < http_access deny all
    ---
> #http_access deny all
> http_access allow all

The rewriter program is just 2 lines,

    #!/bin/sh
    tee /tmp/redirects

In this configuration, Squid segfaults on the first request. Here is the bt
(anonymized):

#0 0x4023de09 in strcasecmp () from /lib/tls/libc.so.6
#1 0x08071310 in clientGetPinnedConnection (conn=0x84bf5b0, request=0x849ab00)
    at client_side.c:5022
#2 0x0807f70a in getPinnedFD (request=0x849ab00) at forward.c:464
#3 0x0807fb08 in fwdConnectStart (data=0x84d5018) at forward.c:515
#4 0x080ac66d in peerSelectCallback (psstate=0x84d5068) at peer_select.c:200
#5 0x0808033d in fwdStart (fd=18, e=0x84d4db0, r=0x849ab00) at forward.c:888
#6 0x0806d8e6 in clientProcessMiss (http=0x84d46f0) at client_side.c:3497
#7 0x0806d58b in clientProcessRequest (http=0x84d46f0) at client_side.c:3420
#8 0x08051e73 in aclCheckCallback (checklist=0x84d4538, answer=ACCESS_ALLOWED) at acl.c:2267
#9 0x08051acb in aclCheck (checklist=0x84d4538) at acl.c:2236
#10 0x08066d76 in clientCheckNoCache (http=0x84d46f0) at client_side.c:657
#11 0x08066a42 in clientRedirectDone (data=0x84d46f0,
    result=0x82c5b10 "http://accelerated.host/") at client_side.c:649
#12 0x080ae032 in redirectHandleReply (data=0x84d46b0,
    reply=0x82c5b10 "http://accelerated.host/") at redirect.c:70
#13 0x0808bf08 in helperHandleRead (fd=0, data=0x82c5ab8) at helper.c:711
#14 0x080747d2 in comm_select (msec=901) at comm_poll.c:471
#15 0x080a11fa in main (argc=1076861356, argv=0x1) at main.c:840

I have to assume I'm missing something important, since as far as I can
tell I'm trying to do the simplest possible acceleration. Plus, I've only
been learning Squid for a week, so my stupid-error factor is probably very
high. :)

Even discounting that factor, though, I think there is room for confusion
with the https_port option. Do the vport and vhost options apply as they do
for http_port? (I assume so, but they aren't listed twice like e.g.
protocol is.) Is accelerated an option, as it states in RELEASENOTES.html,
or isn't it, as its absence in the config file implies?

For the record, I have tried quite a few variations on the https_port
option, to no avail. Some variations don't cause segfaults. But, I thought
I'd start with just one and see what you guys think. :)

TIA,

-Bryan

-- 
Bryan Richter
Systems Administrator
-
iRadeon
Received on Wed Jun 21 2006 - 16:38:36 MDT

This archive was generated by hypermail pre-2.1.9 : Sat Jul 01 2006 - 12:00:02 MDT