On Sun, May 13, 2012 at 9:01 AM, Craig James <cjames(at)emolecules(dot)com> wrote:
> In my experience (PG 8.4.x), the system can handle in the neighborhood of
> 100,000 relations pretty well. Somewhere over 1,000,000 relations, the
> system becomes unusable. It's not that it stops working -- day-to-day
> operations such as querying your tables and running your applications
> continue to work. But system operations that have to scan for table
> information seem to freeze (maybe they run out of memory, or are
> encountering an O(N^2) operation and simply cease to complete).
> For example, pg_dump fails altogether. After 24 hours, it won't even start
> writing to its output file. The auto-completion in psql of table and column
> names freezes the system. It takes minutes to drop one table. Stuff like
> that. You'll have a system that works, but can't be backed up, dumped,
> repaired or managed.
> As I said, this was 8.4.x. Things may have changed in 9.x.
I think some of those things might have improved, but enough of them
have not improved, or not by enough.
So I agree with your assessment, under 9.2 having millions of
sequences might technically work, but would render the database
In response to
pgsql-performance by date
|Next:||From: Віталій Тимчишин||Date: 2012-05-15 06:29:11|
|Subject: Re: Maximum number of sequences that can be created|
|Previous:||From: Noah Misch||Date: 2012-05-13 22:00:53|
|Subject: Re: Any disadvantages of using =ANY(ARRAY()) instead of IN?|