On 01/17/2012 07:09 PM, Jim Nasby wrote:
> On Jan 13, 2012, at 4:15 PM, Christopher Browne wrote:
>> Have two logical tasks:
>> a) A process that manages the list, and
>> b) Child processes doing vacuums.
>>
>> Each time a child completes a table, it asks the parent for another one.
> There is also a middle ground, because having the the scheduling process sounds like a lot more work than what Josh
wasproposing.
>
> CREATE TEMP SEQUENCE s;
> SELECT relname, s mod<number of backends> AS backend_number
> FROM ( SELECT relname
> FROM pg_class
> ORDER BY relpages
> );
>
> Of course, having an actual scheduling process is most likely the most efficient.
We already have a model for this in parallel pg_restore. It would
probably not be terribly hard to adapt to parallel vacuum.
cheers
andrew
