Re: Vacuum analyse after a long time without one ... - Mailing list pgsql-admin

Nicolas Michel <nicolas.michel@lemail.be> writes:
> - I have 16Go of RAM on that server (but 32bits OS with bigmem kernel ;
> so I set shared buffer to 350000 (~2,7GB) for a shmmax of 4000000000
> (~3,8GB)

On a 32-bit machine that's just insane.  You've got something like 300MB
left over in the process address space (assuming the typical 1Gb for
kernel split).  No wonder things are falling over.  Try putting
shared_buffers somewhere around 1Gb.  Or switch to 64-bit.

            regards, tom lane