Обсуждение: Table : Bloat grow high

P {margin-top:0;margin-bottom:0;}

Hi all, 

In postgresql 13.5 

I have a table (size 3.1 GB) and in this table occurs near to 200 updates per second, after 2 days the size table is 7 GB and bloat grow to 45% and 

 the query operations are degraded. vacuum runs every 5 seconds over this. but the bloat growth continues, to solve the 

problem quickly, we have made a replica of the table with a trigger, then a copy of the data and in a 

transaction we rename the table, but it would not be the best solution.

Some suggestion about stop this size increase or parameter to setting up?

On Fri, 2022-11-11 at 17:09 +0000, Alexis Zapata wrote:
> In postgresql 13.5 I have a table (size 3.1 GB) and in this table occurs near to
> 200 updates per second, after 2 days the size table is 7 GB and bloat grow to 45% and 
>  the query operations are degraded. vacuum runs every 5 seconds over this. but the
> bloat growth continues, to solve the 
> problem quickly, we have made a replica of the table with a trigger, then a copy
> of the data and in a 
> transaction we rename the table, but it would not be the best solution.
> Some suggestion about stop this size increase or parameter to setting up?

You'd be most happy with HOT updates.  Make sure that there is no index on any of
the columns you update, and change the table to have a "fillfactor" less than
100.  Then you can get HOT updates which don't require VACUUM for cleaning up.

https://www.cybertec-postgresql.com/en/hot-updates-in-postgresql-for-better-performance/

Yours,
Laurenz Albe
-- 
Cybertec | https://www.cybertec-postgresql.com

On Fri, 2022-11-11 at 17:09 +0000, Alexis Zapata wrote:

In postgresql 13.5 I have a table (size 3.1 GB) and in this table occurs near to
200 updates per second, after 2 days the size table is 7 GB and bloat grow to 45% and 
 the query operations are degraded. vacuum runs every 5 seconds over this. but the
bloat growth continues, to solve the 
problem quickly, we have made a replica of the table with a trigger, then a copy
of the data and in a 
transaction we rename the table, but it would not be the best solution.
Some suggestion about stop this size increase or parameter to setting up?

You'd be most happy with HOT updates.  Make sure that there is no index on any of
the columns you update, and change the table to have a "fillfactor" less than
100.  Then you can get HOT updates which don't require VACUUM for cleaning up.

https://www.cybertec-postgresql.com/en/hot-updates-in-postgresql-for-better-performance/

On 2022-11-12 08:24:23 -0600, Ron wrote:
> On 11/12/22 08:05, Laurenz Albe wrote:
>     You'd be most happy with HOT updates.  Make sure that there is no index on any of
>     the columns you update, and change the table to have a "fillfactor" less than
>     100.  Then you can get HOT updates which don't require VACUUM for cleaning up.
>
>     https://www.cybertec-postgresql.com/en/hot-updates-in-postgresql-for-better-performance/
>
>
> To clarify: do HOT updates automatically happen if there's enough space on the
> page AND you don't update an indexed field (which should be minimized anyway)?

Yes.

> If that is true, what happens if someone then updates an indexed field?

Then that update won't be a HOT update.

>   Does PG keep doing HOT updates on the other tuples,

Yes.

> or does it stop HOT updates altogether until you recluster or full
> vacuum it?

No. The decision on whether an update is a done per tuple update. Within
a single update query, some tuples may be HOT updated and some not. And
for the next update query, each affected tuple is again considered for
HOT updates, so again some may be HOT updated and some not. If you are
updating the same tupel several times, you may get a few HOT updates
first, then a non-HOT update, then HOT updates again.

        hp

--
   _  | Peter J. Holzer    | Story must make more sense than reality.
|_|_) |                    |
| |   | hjp@hjp.at         |    -- Charles Stross, "Creative writing
__/   | http://www.hjp.at/ |       challenge!"

On Sun, 2022-11-13 at 14:50 +0000, Alexis Zapata wrote:
> I found that xmin does not change when running the vacuum.

Which xmin?

Yours,
Laurenz Albe