[Varnish] #51: crash in cache_acceptor_epoll.c

[Varnish]

#51: crash in cache_acceptor_epoll.c
----------------------+-----------------------------------------------------
 Reporter:  airmax    |       Owner:  phk
     Type:  defect    |      Status:  new
 Priority:  high      |   Milestone:     
Component:  varnishd  |     Version:  1.0
 Severity:  blocker   |    Keywords:     
----------------------+-----------------------------------------------------
 Hi,

 I'm trying varnishd, but for the moment, I'm very disappointed. Varnish
 doesn't run more than a minute on my setup. I tried to change the param on
 startup (number of worker thread, file, malloc instead of file, etc...),
 but the problem remains the same. varnishd crashes after some seconds,
 sometime just after startup, if you get lucky it may work for 60 ou 90
 seconds. When I say varnishd crashes, I mean the worker process with 1 gig
 of shared memory crashes, the other process then forks again a new worker
 process, which is affected by the bug and does crash. So the reverse proxy
 does never really go down, but it's not very efficient this way

 Running with :
 {{{
 ./varnishd -a '*:80' -b '192.168.4.230:80' -T 127.0.0.1:1664 -d
 -sfile,/tmp/varnish,1g -w20,200,3
 }}}

 I got message like that :
 {{{
 start child pid 25465
 Child said (2, 25465): <<Child starts
 managed to mmap 1073741824 bytes of 1073741824
 Ready
 CLI ready
 socktest: linger=0 sndtimeo=0 rcvtimeo=0
 >>
 Child said (2, 25465): <<Assert error in vca_del(), cache_acceptor_epoll.c
 line 68:
   Condition(i == 0 || errno == EBADF) not true.
   errno = 14 (Bad address)
 >>
 Cache child died pid=25465 status=0x6
 Clean child
 Child cleaned
 }}}

 Test machine : Linux x86 2.6.8, 512 Mo. Backend is apache 1.3.31 on a
 different test machine.

-- 
Ticket URL: <http://varnish.projects.linpro.no/ticket/51>
Varnish <http://varnish.projects.linpro.no/>
The Varnish HTTP Accelerator