Обсуждение: Transactions and insertion ordering

Hi,

I'm trying to work out how to make sure things are read from a table in
a consistent order. The table represents a queue of items and also the
history of those items.

Even with "serializable" transaction isolation I can begin two
transactions, insert a record in each, commit the second transaction
first. This second record is now visible and can be read from the queue.
But when I commit the first this one appears before the second one. This
could then be read from the queue second but when I rescan the table to
view history it looks like it was read first.

Are there any ways to make this work a little more intuitively?
Basically I guess I want to be able to model a queue effectively.

James

nextval() should return value in the order they were called, rather
than commit time. I hope you're not relying on the unordered results of
a table scan remaining stable. Tables have no intrinsic "order", only
one inposed by an external sequence.

Hope this helps,

On Thu, Jun 10, 2004 at 09:28:50AM +0100, James Pharaoh wrote:
> Hi,
>
> I'm trying to work out how to make sure things are read from a table in
> a consistent order. The table represents a queue of items and also the
> history of those items.
>
> Even with "serializable" transaction isolation I can begin two
> transactions, insert a record in each, commit the second transaction
> first. This second record is now visible and can be read from the queue.
> But when I commit the first this one appears before the second one. This
> could then be read from the queue second but when I rescan the table to
> view history it looks like it was read first.
>
> Are there any ways to make this work a little more intuitively?
> Basically I guess I want to be able to model a queue effectively.
>
> James
>
>
> ---------------------------(end of broadcast)---------------------------
> TIP 4: Don't 'kill -9' the postmaster

--
Martijn van Oosterhout   <kleptog@svana.org>   http://svana.org/kleptog/
> Patent. n. Genius is 5% inspiration and 95% perspiration. A patent is a
> tool for doing 5% of the work and then sitting around waiting for someone
> else to do the other 95% so you can sue them.

On Thu, Jun 10, 2004 at 20:15:53 +1000,
  Martijn van Oosterhout <kleptog@svana.org> wrote:
> nextval() should return value in the order they were called, rather
> than commit time. I hope you're not relying on the unordered results of
> a table scan remaining stable. Tables have no intrinsic "order", only
> one inposed by an external sequence.

This isn't really guarenteed with nextval. I think it will work if you
are just reserving one value at a time (which is the default).

I think the real problem is that the original poster needs to precisely
define what determines order. If the precise definition is transaction
commit order, I think that is going to be hard to do exactly right.

On Thu, 2004-06-10 at 14:47, Bruno Wolff III wrote:
> I think the real problem is that the original poster needs to precisely
> define what determines order. If the precise definition is transaction
> commit order, I think that is going to be hard to do exactly right.

Yes, that is what I want. So I can guarantee that the order of the IDs
in the database will be the same as the order in which they are taken
out of the queue.

I think I've come up with a reasonable solution now though. I can lock
the record representing the queue in another table FOR UPDATE and then
do the insert, then no other process will be able to gain that lock
until I complete. Best bit is other items can still insert concurrently,
but only one per queue, which is exactly what I was after.

Thanks for the help anyway ;-)

James