Hello, On Fri, Feb 21, 2020 at 08:52:05AM +0000, marcel.klooster...@kpn.com wrote: > Hi all, > > At this moment we’re using HA-Proxy version 2.0.0 2019/06/16 - > https://haproxy.org/ > And we’re running solaris 11.4 sru 950 for SPARC and I used developer studio > version 12.6 to compile the newer versions and 12.5 to compile the 2.0.0 > version. > > And because of a known bug in haproxy, we wanted to upgrade to: HA-Proxy > version 2.1.2 2019/12/21 - https://haproxy.org/ > On our test-server we had no problems but on the production server it kept > crashing. > The problem is somehow related to ssl, because as soon as we add “ssl crt > /etc/certs/self.pem” to the bind it crashes also on the test-server. > So then I checked and version 2.1.3 is already available, solving lots of ssl > bugs. So I tried that one also…but without any luck ☹ > > Starting haproxy now gives a Bus Error (with both versons, 2.1.2 and 2.1.3): > > root@sz4203:~ # /usr/local/sbin/haproxy -p /tmp/haproxy.pid -f > /usr/local/etc/haproxy.cfg > Bus Error (core dumped) > > Can you please help me solving this issue ? > > This is the stack-dump of the 2.1.3 version I compiled: > root@s00501:/var/core# pstack core_sz4203_haproxy_0_0_1582273479_18461 > core 'core_sz4203_haproxy_0_0_1582273479_18461' of 18461: > /usr/local/sbin/haproxy -p /tmp/haproxy.pid -f /usr/local/etc/haproxy. > 0000000100240c30 shctx_init (100366170, 4e20, c4, ffffffff, ffffffff7ec00144, > 0) + b0 (shctx.c:368) > 0000000100086e7c ssl_sock_prepare_bind_conf (1005bbb80, ffffffff, > fffffffefffffffc, ffffffff29729948, ffffffff00000000, 100366000) + 1bc > 0000000100135278 check_config_validity (fffffffefffffffc, 1400, 1005bbb80, > 1005bb750, 100357238, 0) + 25b8 (cfgparse.c:3753) > 0000000100173450 init (100357468, 1000596a0, 10005a940, 100357238, 10003aa90, > 0) + 610 (haproxy.c:1884) > 0000000100178058 main (5, ffffffff7ffff688, ffffffff7ffff6b8, 100000, > 10046d680, 0) + 178 (haproxy.c:2858) > 000000010006caa4 _start (0, 0, 0, 0, 0, 0) + 64 >
That's really odd, what is crashing is the shctx_init(), which is the initialization of the shared memory used by the SSL, it didn't changed much in 2 years so I'm surprised that you didn't have this problem with haproxy 2.0. Since its running on SPARC and that the Bus Error appears in the the shctx block access that could be an alignment issue. But if it's working on your test server it might be something else. Do the test server and the production one share the same hardware? (CPU, RAM etc.) Could you display the stacktrace with gdb? The arguments looks weird to me, (too much arguments and weird values) and I don't know how reliable is pstack about this. Did you try to build haproxy 2.0 with your new compiler and do the same tests? Do you have the same issue? -- William Lallemand
