Re: pgsql: Generational memory allocator

On 11/23/2017 09:06 PM, Tom Lane wrote:
> Andres Freund <andres(at)anarazel(dot)de> writes:
>> On 2017-11-23 20:57:10 +1100, Simon Riggs wrote:
>>> On 23 November 2017 at 11:16, Andres Freund <andres(at)anarazel(dot)de> wrote:
>>>> Looks like it's not quite valgrind clean:
>>>> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=skink&dt=2017-11-22%2022%3A30%3A01
>
>>> It doesn't report anything useful that would allow me to fix this.
>
>> Uh. It reports a failure, and you can go from there. The error output
>> actually is in postmaster.log but for some reason the buildfarm code
>> didn't display that in this case.
>

It may not be immediately obvious that the failure is due to valgrind,
but otherwise I agree it's up to us to investigate.

> I think it's a legitimate complaint that postmaster.log wasn't captured
> in this failure, but that's a buildfarm script oversight and hardly
> Andres' fault.
>

Are the valgrind errors really written to postmaster log? I'm assuming
it failed because valgrind ran into an issue and killed the process.

> In any case, valgrind failures are generally easy enough to reproduce
> locally.
>

Right.

> Meanwhile, over on
> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=snapper&dt=2017-11-23%2013%3A56%3A17
>
> we have
>
> ccache gcc-4.7 -Wall -Wmissing-prototypes -Wpointer-arith -Wdeclaration-after-statement -Wendif-labels -Wmissing-format-attribute -Wformat-security -fno-strict-aliasing -fwrapv -fexcess-precision=standard -g -g -O2 -fstack-protector --param=ssp-buffer-size=4 -Wformat -Werror=format-security -I../../../../src/include -D_FORTIFY_SOURCE=2 -D_GNU_SOURCE -I/usr/include/libxml2 -I/usr/include/mit-krb5 -c -o generation.o generation.c
> generation.c: In function ‘GenerationContextCreate’:
> generation.c:206:7: error: static assertion failed: "padding calculation in GenerationChunk is wrong"
> make[4]: *** [generation.o] Error 1
>
> Looks to me like GenerationChunk imagines that 3*sizeof(pointer)
> must be a maxaligned quantity, which is wrong on platforms where
> MAXALIGN is bigger than sizeof(pointer).
>

Hmmm, I see. Presumably adding this to GenerationChunk (similarly to
what we do in AllocChunkData) would address the issue:

#if MAXIMUM_ALIGNOF > 4 && SIZEOF_VOID_P == 4
Size padding;
#endif

but I have no way to verify that (no access to such machine). I wonder
why SlabChunk doesn't need to do that (perhaps a comment explaining that
would be appropriate?).

regards
Tomas