On Tue, Jun 22, 2021 at 08:54:31AM -0400, Greg Troxel wrote: > it looks like a normal stat syscatl, but also like there is some > invocation of the emulation layer.
stat here is only the trigger, but not the cause. The kernel ran out of memory (the kernel pmap can not be enlarged) - which is a very strange scenario to hit on amd64 machines. Does vmstat -s or top output show some memory usage grow (slowly) before the crash? Maybe the new version of postgres uses some syscall that leaks kernel memory and nobody noticed so far. Do you have any special tuning sysctls in effect? Martin
