Re: [PATCH] Refactoring of LWLock tranches

On Fri, Jan 8, 2016 at 11:22 PM, Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> wrote:
> That idea won't work as we need to separately register tranche for
> each process. The other wayout could be to do it in CreateSharedProcArray()
> which will be quite similar to what we do for other tranches and
> it will cover all kind of processes. Attached patch fixes this problem.
>
> I have considered to separately do it in InitProcessPhase2() and
> InitAuxiliaryProcess(), but then the registration will be done twice for
> some
> of the processes like bootstrap and same is true if we do this InitProcess()
> instead of InitProcessPhase2() and I think it won't be similar to what
> we do for other tranches.
>
> I have done the performance testing of the attached patch and the
> results are attached with this mail. The main tests conducted are
> pgbench read-write and read-only tests and the results indicate that
> this patch doesn't introduce any regression, though you will see some
> cases where the performance is better with patch by ~5% and then
> regressed by 2~3%, but I think it is more of a noise, then anything
> else.

Committed.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company