[Bizgres-general] Statement Queuing take II - Resource Scheduling

[Luke Lonergan]

Jim, Mark,

On 6/19/06 9:29 AM, "Jim C. Nasby" <jnasby at pervasive.com> wrote:

> On Sun, Jun 18, 2006 at 02:07:49PM +1200, Mark Kirkwood wrote:
>> I do agree that being able to send a "query is queued" client message
>> type would be good, but the level of impact on existing client
>> libraries/drivers makes that not so nice.
> 
> Plus, it can always be added later.

I'll bring this up now at the risk of having a prolonged discussion about
it, but the logical next step here might be to introduce time-sliced
scheduling after the queuing is working.

What I mean by this is that the backends launched to process connections
would execute in time slices subject to the management of a resource
scheduler.  Different queues would have different time slice sizes, all time
slices would probably run in the range of seconds or tens of seconds and
could be preemptively released early.

This is basically what the OS kernels do, and in our case would be done for
the same reasons but at the DBMS level.

- Luke