Re: [ADMIN] Autovacuum after bulk data insert(millions!)

On Tue, Oct 17, 2017 at 12:59 AM, Günce Kaya <guncekaya14@gmail.com> wrote:
> Hi all,
>
> We have a database that is used for CRM. During day there are many process
> like bulk loading large amounts of data(not thousand like millions!). And
> all day we observe auto vacuum in server status. It takes long time. I'm
> curious what happens if we disable autovacuum for a database that process
> bulk insert? After insert data a table, what autovacuum will do for that
> table?
>
> I've read some article like; "The autovacuum process takes care of several
> maintenance chores inside your database that you really need. Generally, if
> you think you need to turn regular vacuuming off because it's taking too
> much time or resources, that means you're doing it wrong. The answer to
> almost all vacuuming problems is to vacuum more often, not less, so that
> each individual vacuum operation has less to clean up.
> However, it's acceptable to disable autovacuum for short periods of time,
> for instance when bulk loading large amounts of data."
>
> The reason why I'm asking, millions of data is inserting to CRM database all
> day. What is your advice for that scenario? Should I disable autovacuum
> during bulk process or disable autovacuum totally and autovacuum tables in a
> some period?

I'm not sure the exact problem you're trying to solve here. Do you
want vacuum to finish more quickly? Is autovacuum impacting
performance of the cluster? Or is autovacuum kicking in partway
through the bulk load and causing that to be slow?

If you have a fast IO subsystem and you just want autovacuum to finish
more quickly, then look at making it more aggressive by lowering the
autovacuum_vacuum_cost_delay from the default 20ms to something lower
like 5 or 10ms.

If autovacuum is kicking in part way and affecting performance then
you can look at disabling autovacuum on those tables and then
vacuuming after the bulk upload. Set things like vacuum_cost_delay to
keep vacuum from taking up too much IO. Note that the default for
vacuum_cost_delay is 0, while the default for
autovacuum_vacuum_cost_delay is 20ms. These two values are worlds
apart in terms of impact on the IO of your db server.

Of the cuff I'd guess what you want is for autovacuum to keep up, and
it's taking too long. In that instance, just try lowering
autovacuum_vacuum_cost_delay from 20ms down to 10ms at first and check
both to see if autovacuum is flooding your IO and if autovacuum now
finishes "soon enough" for your needs.