On Fri, Mar 20, 2020 at 3:59 PM Peter J. Holzer <hjp-pg...@hjp.at> wrote:

> On 2020-03-19 16:48:19 -0700, David G. Johnston wrote:
> > First, it sounds like you care about there being no gaps in the records
> you end
> > up saving.  If that is the case then sequences will not work for you.
>
> I think (but I would love to be proven wrong), that *nothing* will work
> reliably, if
>
> 1) you need gapless numbers which are strictly allocated in sequence
>
A little gap is acceptable. We cannot afford a 100 gap though.

2) you have transactions
> 3) you don't want to block
>
> Rationale:
>
> Regardless of how you get the next number, the following scenario is
> always possible:
>
> Session1: get next number
> Session2: get next nummber
> Session1: rollback
> Session2: commit
>
> At this point you have a gap.
>
> If you can afford to block, I think a simple approach like
>
>     create table s(id int, counter int);
>     ...
>     begin;
>     ...
>     update s set counter = counter + 1 where id = $whatever returning
> counter;
>     -- use counter
>     commit;
>
> should work. But that effectively serializes your transactions and may
> cause some to be aborted to prevent deadlocks.
>
>         hp
>
> --
>    _  | Peter J. Holzer    | Story must make more sense than reality.
> |_|_) |                    |
> | |   | h...@hjp.at         |    -- Charles Stross, "Creative writing
> __/   | http://www.hjp.at/ |       challenge!"
>

Reply via email to