Strange system crash?
Posted in 2000
Topics: Installation, Setup & Upgrades, Platform-Specific Issues, Clustering, Grid & MACH11
Over the past few weeks we have had two Sun 450's running Informix
9.20.UC2 when suddenly the system crashes and no error messages show up
in the Informix or system logs (and I'm logging EVERYTHING). There
isn't even a core dump. I was wondering if anyone else has experienced
a problem like this.
The problem has occurred on both servers (which are identical) at random
times, sometimes when the database is working hard, sometimes when it's
idle (but oninit is running.) The systems have run for nearly a week at
a time without crashing but sometimes they'll crash three times in a
day...
The Sun 450 is running with a fresh install of Solaris 7, the most
current patch cluster is installed, I made the recommended kernel
changes, and these servers aren't running anything but the OS and
Informix. Each has 1GB of RAM and 2 processors.
One thing I noticed but haven't found good documentation on, is that the
Informix resident shared memory size was close to 700,000kbytes which
exceeds the kernel shmmax parameter. I have since changed the shmmax
kernel parameter from the default recommended value of 268435456 to
805306368 and the systems didn't crash for over a week. The change
appears to have fixed the problem, but I'm not sure. Does the kernel
shmmax parameter have to be larger than the resident shared memory
size? If it isn't, would that really cause the system to crash without
warning?
Ross
rdberman@yahoo.com
Ross Berman wrote:
> ...
> appears to have fixed the problem, but I'm not sure. Does the kernel
> shmmax parameter have to be larger than the resident shared memory
> size?
No, it doesn't (although its recommended).
> If it isn't, would that really cause the system to crash without
> warning?
Haven't seen crashes, but have seen misbehaviour in accesses to Shared
Memory (e.g onstat commands failing)
Rudy