Csaba Nagy wrote:
> On Tue, 2007-01-09 at 17:36, Csaba Nagy wrote:
>
>> On Tue, 2007-01-09 at 17:31, Matthew T. O'Connor wrote:
>>
>>> Without getting into all the details, the autovacuum naptime is a GUC
>>> variable right now, so it can be much more frequent than the current
>>> default which is 60 seconds.
>>>
>> Hmm, for some reason I thought the granularity is minutes, but it is
>> indeed in seconds... one more thing learned.
>>
>
> OK, so after checking my config, it is still not optimal because it
> refers to all the data bases in the cluster. I have setups where I have
> multiple data bases in the same cluster, with various degrees of
> activity... some of them should be checked continuously, some rarely...
> so now if I let the default 60 seconds, each data base will be checked
> in db_count * (60 + vacuum time) seconds. This is not optimal... some of
> the DBs have way much activity than others. Those I would like to be
> checked say each 10 seconds, the rest each 5 minutes...
>
Agreed, this is the point of this whole thread that there are lots of
setups where autovacuum could do better. My point was only that as we
move forward with these multiple queue / multiple worker process setups
etc, that we already have some infrastructure to make things go faster.
