On 25/08/17 10:28, Chris Travers wrote:
>
>
> On Thu, Aug 24, 2017 at 9:44 PM, Andres Freund <andres@anarazel.de
> <mailto:andres@anarazel.de>> wrote:
>
> Hi,
>
> On 2017-08-18 12:12:58 +0300, Ildar Musin wrote:
> > While we've been developing pg_pathman extension one of the most frequent
> > questions we got from our users was about global index support. We cannot
> > provide it within an extension. And I couldn't find any recent discussion
> > about someone implementing it. So I'm thinking about giving it a shot and
> > start working on a patch for postgres.
>
> FWIW, I personally think for constraints the better approach is to make
> the constraint checking code cope with having to check multiple
> indexes. Initially by just checking all indexes, over the longer term
> perhaps pruning the set of to-be-checked indexes based on the values in
> the partition key if applicable. The problem with creating huge global
> indexes is that you give away some the major advantages of partitioning:
> - dropping partitions now is slow / leaves a lof of garbage again
> - there's no way you can do this with individual partitions being remote
> or such
> - there's a good chunk of locality loss in global indexes
>
> The logic we have for exclusion constraints checking can essentially be
> extended to do uniqueness checking over multiple partitions. Depending
> on the desired deadlock behaviour one might end up doing speculative
> insertions in addition. The foreign key constraint checking is fairly
> simple, essentially one "just" need to remove the ONLY from the
> generated check query.
>
+1 (or +as much as I am allowed to get away with really ;) )
>
> To be clear, this would still require a high-level concept of a global
> index and the only question is whether it gets stored as multiple
> partitions against partitioned tables vs stored in one giant index, right?
>
No, just global constraints. For example, if you consider unique index
to be implementation detail of a unique constraint, there is nothing
stopping us to use multiple such indexes (one per partition) as
implementation detail to single global unique constraint. No need for
global index at all.
-- Petr Jelinek http://www.2ndQuadrant.com/ PostgreSQL Development, 24x7 Support, Training &
Services