Обсуждение: Blocksize Config Make Check Failures

I'm attempting to get a little bit more disk performance out of a data warehouse like PG 9.5 RC1 instance I have that
isrunning on ZFS (recordsize=128K).  When trying to make PG 9.5RC1 with --blocksize > 8 four of the make check tests
arereported as failing.  I'm curious as to if this is a problem with the checks or if the errors would cause problems
inusage.  None of the make check errors caused the server to terminate or have serious issues but I just wanted to
checkif anyone else had concerns.   

You can see for yourself the failures with this command:
git checkout tags/REL9_5_RC1; make clean; ./configure --with-blocksize=16; make; make check

For example the select_views regression.diff shows that the explain plan changed which doesn't seem overly concerning.
Thetablesample failure though is more interesting.  I'm not entirely sure that tablesample is working correctly though
interms of the count of data it should be returning but at least it is returning some data. 

See attached for the regression.diffs.

Mike

GMail <mfwilson@gmail.com> writes:
> I'm attempting to get a little bit more disk performance out of a data
> warehouse like PG 9.5 RC1 instance I have that is running on ZFS
> (recordsize=128K).  When trying to make PG 9.5RC1 with --blocksize > 8
> four of the make check tests are reported as failing.

Yeah, that's unsurprising.  You can also get some discrepancies from the
expected output if you, eg, run with nondefault planner parameters.

Thanks for pointing this out though, because I notice that somebody
stuck a bunch of plain EXPLAIN commands into rowsecurity.sql, in clear
violation of the project policy about using COSTS OFF in regression
explain commands.  That's causing some though not all of your diffs,
and might lead to pseudo-failures in other environments that we do
consider supported.  I'll go fix that ...

            regards, tom lane