librelist archives

two servers, one host (causing a crash)

From:

Ed Marshall

Date:

2011-06-14 @ 23:49

I have a fairly simple configuration here that seems to be causing an
SSL-related crash (UUIDs stripped):

hgwebdir = Handler(send_spec='tcp://127.0.0.1:9997',
                   send_ident='hgwebdir-send-ident-uuid',
                   recv_spec='tcp://127.0.0.1:9996', recv_ident='')

redmine = Proxy(addr='127.0.0.1', port=3000)

devhost = Host(name="lens.logic.net", routes={
              '/': Dir(base='docroot/', index_file='index.html',
                       default_ctype='text/plain'),
              '/hg': hgwebdir,
              '/redmine': redmine
          })

dev = Server(
    uuid="dev-server-uuid",
    access_log="/logs/access.log",
    error_log="/logs/error.log",
    chroot="/srv/mongrel2/",
    default_host="lens.logic.net",
    name="dev",
    pid_file="/run/m2-dev.pid",
    port=80,
    hosts = [devhost]
)

devssl = Server(
    uuid="devssl-server-uuid",
    access_log="/logs/access-ssl.log",
    error_log="/logs/error-ssl.log",
    chroot="/srv/mongrel2/",
    default_host="lens.logic.net",
    name="devssl",
    pid_file="/run/m2-dev-ssl.pid",
    port=443,
    hosts=[devhost],
    use_ssl=1
)

settings = {
    "upload.temp_store": "/tmp/mongrel2.upload.XXXXXX",
    "certdir": "./certs/"
}

servers = [dev,devssl]


Accessing /hg on "dev" works exactly as expected, but accessing it through
"devssl" crashes with:

*** glibc detected *** mongrel2: double free or corruption (fasttop):
0x0000000000ae6030 ***
======= Backtrace: =========
/lib64/libc.so.6[0x3d00076fea]
mongrel2(Connection_send_to_handler+0x1f6)[0x4591d6]
mongrel2(connection_http_to_handler+0x1ec)[0x45948c]
mongrel2(State_exec+0x30f)[0x45c16f]
mongrel2(Connection_task+0x7c)[0x45965c]
mongrel2[0x4178bd]
/lib64/libc.so.6[0x3d00046210]
======= Memory map: ========


Running mongrel2 1.7.3 with zeromq 2.1.4 (as shipped with Fedora 15
x86_64). I'll try and grab a backtrace later, if you'd find it useful.

(And if I'm doing something terribly foolish above, I'd love to hear that
too. :)

-- 
Ed Marshall <esm@logic.net>
Felix qui potuit rerum cognoscere causas.
http://esm.logic.net/

On Tue, Jun 14, 2011 at 04:49:20PM -0700, Ed Marshall wrote:
> I have a fairly simple configuration here that seems to be causing an
> SSL-related crash (UUIDs stripped):

Ok, I'll try that out tonight.  If you get a chance, can you run it 
under valgrind and see what valgrind says is the offending line that caused it?

Zed