Re: [PERFORM] encouraging index-only scans

On 09/06/2013 03:12 PM, Andres Freund wrote:
> On 2013-09-06 13:38:56 +0200, Hannu Krosing wrote:
>> On 09/06/2013 09:23 AM, Dimitri Fontaine wrote:
>>> Alvaro Herrera <alvherre(at)2ndquadrant(dot)com> writes:
>>>> I'm not sure if we need to expose all these new maintenance actions as
>>>> SQL commands.
>>> I strongly think we should, if only for diagnostic purposes.
>> It would be much easier and more flexible to expose them
>> as pg_*() function calls, not proper "commands".
> I don't think that's as easy as you might imagine. For much of what's
> done in that context you cannot be in a transaction, you even need to be
> in a toplevel statement (since we internally
> CommitTransactionCommand/StartTransactionCommand).
>
> So those pg_* commands couldn't be called (except possibly via the
> fastpath function call API ...) which might restrict their usefulnes a
> teensy bit ;)
>
> So, I think extending the options passed to VACUUM - since it can take
> pretty generic options these days - is a more realistic path.
Might be something convoluted like

VACUUM indexname WITH (function = "pg_cleanup_gin($1)");

:)
>
>>> Also to
>>> adapt to some well defined workloads that the automatic system is not
>>> designed to handle.
>> +1
> What would you like to expose individually?
>
> Greetings,
>
> Andres Freund
>