A few quick things:
Have you tried the latest 1.3.3? 1.4 release is imminent and a lot of work has been done there. Can you run your test with the memcached-debug binary? It should provide more useful results. Most importantly, can you provide a test program we can use to recreate this failure, understand it well, and be sure it doesn't happen again? On May 23, 5:40 am, Gavin Hamill <g...@acentral.co.uk> wrote: > Hi, we've been using 1.2.6 on amd64 for months without a single problem, > but we want to migrate from 2 x 16GB machines to 16 x 4G 32-bit machines > for better survivability. > > Unfortunately we're getting a lot of segfaults both with 1.2.6 and > 1.2.8 :( > > This is the backtrace I get from a 1.2.6-debug instance (I have since > started gdb on a 1.2.8-debug instance to get the 'latest and greatest) > > Hopefully this will mean something to a dev, because I'm gradually > losing my sanity. > > calloc()Starting program: /usr/local/sbin/memcached-debug -m 3200 -p > 11211 -u nobody -c 4096 -t 10 > [Thread debugging using libthread_db enabled] > [New Thread 0xb7da96b0 (LWP 24417)] > [New Thread 0xb7d47b90 (LWP 24420)] > [New Thread 0xb7546b90 (LWP 24421)] > [New Thread 0xb6d45b90 (LWP 24422)] > [New Thread 0xb6544b90 (LWP 24423)] > [New Thread 0xb5d43b90 (LWP 24424)] > [New Thread 0xb5542b90 (LWP 24425)] > [New Thread 0xb4d41b90 (LWP 24426)] > [New Thread 0xb4540b90 (LWP 24427)] > [New Thread 0xb3d3fb90 (LWP 24428)] > > calloc() > calloc() > [ maybe 100 of these...] > calloc() > > Program received signal SIGSEGV, Segmentation fault. > [Switching to Thread 0xb6544b90 (LWP 24423)] > 0x0804a069 in add_iov (c=0x9b69cf0, buf=0x9ea32e0, len=44) at > memcached.c:683 > 683 memcached.c: No such file or directory. > in memcached.c > (gdb) > (gdb) bt > #0 0x0804a069 in add_iov (c=0x9b69cf0, buf=0x9ea32e0, len=44) at > memcached.c:683 > #1 0x0804ded3 in event_handler (fd=295, which=2, arg=0x9b69cf0) at > memcached.c:2317 > #2 0xb7f5b9e2 in event_base_loop () from /usr/lib/libevent-1.3e.so.1 > #3 0x08051807 in worker_libevent (arg=0x993a3a8) at thread.c:324 > #4 0xb7de74c0 in start_thread () from /lib/i686/cmov/libpthread.so.0 > #5 0xb7edb6de in clone () from /lib/i686/cmov/libc.so.6 > (gdb) > > The servers hosting these are little low-power VIA EPIA / Intel Atom > boxes.. load average is zero and they are dedicated to memcached, > running Debian lenny. > > We typically have 10 clients making a total of 500-1000 simultaneous TCP > connections (shown by netstat) to each mc server. > > libc6 is 2.7-18, linux-image 2.6.26-13 > > Cheers, > Gavin