Re: Large number of partitions of a table

Ron wrote:
> On 1/16/22 11:38 PM, Victor Sudakov wrote:
> > Tom Lane wrote:
> > [dd]
> >
> >> The specific point that depesz was responding to in that blog
> >> was the 64K-ish limit on rangetable entries in a query. That is
> >> a thing, as he could have shown by using queries that weren't
> >> amenable to plan-time pruning. (It's also an ex-thing, having
> >> been fixed for v15 [1]; but that doesn't help you today.)
> >> Now, if you use no queries that can't be pruned to a few
> >> partitions, then it's academic for you.
> > The table will be partitioned `BY LIST (customer_id)` which is a unique
> > index. All queries will be using this index
>
> Good.
>
> > so no query should ever have to use more than 1 partition.
>
> I find it hard to believe that you'll *never* run a report against more
> customers than are in a single partition.

Thank you for raising this question, it can be of great interest.

What's the worst thing to happen if someone runs "SELECT COUNT(*) FROM t" where t has 10000 partitions?

1. The backend will crash?

2. The whole cluster will crash?

3. Only this particular query (spanning multiple partitions) will be very slow?

4. ?

Also, what if it is not a SELECT but an UPDATE query spanning multiple partitions? Does it make any difference?

--
Victor Sudakov VAS4-RIPE
http://vas.tomsk.ru/
2:5005/49(at)fidonet