Re: BUG #5235: Segmentation fault under high load through JDBC

Andrew Gierth wrote:
>>>>>> "Robert" == Robert Haas <robertmhaas@gmail.com> writes:
>
>  Robert> How about (3) getrlimit(RLIMIT_STACK) lies through its teeth,
>  Robert> by ignoring the existence of another and lower limit imposed
>  Robert> elsewhere?
>
>  Robert> A little Googling seems to reveal that FreeBSD has a
>  Robert> parameter called MAXSSIZ (and possibly a variant for 64-bit
>  Robert> builds).  I kind find a lot of people talking about needing
>  Robert> to raise it (for MySQL, among other things), but I haven't
>  Robert> been able to determine for certain what the default is.
>  Robert> Perhaps it is set to a really low value on the OP's system?
>
> The default is 64MB on i386, 512MB on amd64; that's where the
> getrlimit value comes from unless it's been explicitly reduced
> somewhere. The kernel MAXSSIZ sets the value of the hard limit for
> RLIMIT_STACK for proc0, and everything else inherits that. All
> setrlimit calls for RLIMIT_STACK are explicitly clamped to MAXSSIZ, so
> there's no way to set that value higher than the kernel limit, and no
> way for getrlimit to report a value higher than the real limit.

I vaguely recall issues in the past with linking of postgresql (or PLs
that require it) against libc_r causing some rather small stack limits
being imposed under some circumstances but I don't recall the details
any more...



Stefan