Re: Reducing opr_sanity test's runtime under CLOBBER_CACHE_ALWAYS

On 5/8/21 3:44 PM, Tom Lane wrote:
> Anyway, I propose that we ought to sneak this into HEAD, since
> it's only touching test code and not anything production-critical.
>
> The patch is a bit more invasive than I would have liked, because
> adding the SQL definition of binary_coercible() to create_function_1
> (where the other regress.c functions are declared) didn't work:
> that runs after opr_sanity, and just moving it up to before
> opr_sanity causes the latter to complain about some of the functions
> in it.  So I ended up splitting the create_function_1 test into
> create_function_0 and create_function_1.   It's annoying from a
> parallelism standpoint that create_function_0 runs by itself, but
> the two parallel groups ahead of it are already full.  Maybe we
> should rebalance that by moving a few of those tests to run in
> parallel with create_function_0, but I didn't do that here.
>
> Thoughts?


+1 for doing it now.


You could possibly just move "inet macaddr macaddr8 " to the following
group and so have room for create_function_0. I just tried that and it
seemed happy.


cheers


andrew


--
Andrew Dunstan
EDB: https://www.enterprisedb.com