On Sat, Apr 7, 2018 at 1:09 PM, Andres Freund <andres@anarazel.de> wrote:
> Hi,
>
> On 2018-04-07 15:49:54 +1200, David Rowley wrote:
>> Right, I suggest we wait and see if all members go green again as a
>> result of 40e42e1024c, and if they're happy then we could maybe leave
>> it as is with the 2 alternatives output files.
>
> At least the first previously borked animal came back green (termite).
>
>
>> I don't particularly think it matters which hash partition a tuple
>> goes into, as long as the hash function spreads the values out enough
>> and most importantly, the pruning code looks for the tuple in the
>> partition that it was actually inserted into in the first place.
>> Obviously, we also want to ensure we never do anything which would
>> change the matching partition in either minor or major version
>> upgrades too.
>
> +1
+1
Given that the difference only appeared on animals that David pointed
out have big-endian architecture, it seems we'd only need two output
files. It does seem true that the extended hashing functions that
were adding to support partitioning would somehow be affected by
endianness.
Thank you David for creating the patches and Andres for committing it.
Buildfarm seems to be turning green where it had gone red due to the
hashing differences.
> I've also attempted to fix rhinoceros's failure I remarked upon a couple
> hours ago in
> https://postgr.es/m/20180406210330.wmqw42wqgiicktli@alap3.anarazel.de
Thanks Andres.
Regards,
Amit