Re: autovacuum next steps, take 2

Tom Lane wrote:
> "Matthew T. O'Connor" <matthew(at)zeut(dot)net> writes:
>> Tom Lane wrote:
>>> I'm inclined to propose an even simpler algorithm in which every worker
>>> acts alike;
>
>> That is what I'm proposing except for one difference, when you catch up
>> to an older worker, exit.
>
> No, that's a bad idea, because it means that any large table starves
> even-larger tables.

True, but the assumption I'm making is that there is a finite amount of
bandwidth available and more concurrent activity will have a net
negative effect the time it takes to vacuum all tables. I'm willing to
pay that price to prevent small hot tables from getting starved, but
less willing to pay the same price for large tables where the percentage
of bloat will be much smaller.

> (Note: in all this I assume we're all using "size" as a shorthand for
> some sort of priority metric that considers number of dirty tuples not
> only size. We don't want every worker insisting on passing over every
> small read-only table every time, for instance.)

I was using size to mean reltuples. The whole concept of sorting by
size was designed to ensure that smaller (more susceptible to bloat)
tables got priority. It might be useful for workers to sort their to-do
lists by some other metric, but I don't have a clear vision of what that
might be.