pg_total_relation_size accuracy guarantee

[Nikita]

Hello, I'am currently trying to implement some simple disc space limitation
mechanism for my database. Database is partition based and insert-only - I
insert new rows to the current partition and drop oldest partition when
disc space limit exceeded.

Algorithm is quite simple and it seems it works but I must ensure two
assumptions to avoid extra VACUUM/ANALYZE calls:

1. pg_total_relation_size return predictable approximately correct table
size (apart from some preallocation) without subsequent
VACUUM/ANALYZE/anything call as long as I don't DELETE/UPDATE rows;

2. DROP/TRUNCATE table reclaim disc space without subsequent VACUUM;

These assumptions are correct? Is it fixed in doc if so?

Re: pg_total_relation_size accuracy guarantee

[Nikita]

Source is good enough in my case and I planned to grep it anyway, so thanks
for saving my time!

2017-11-24 10:15 GMT+03:00 Laurenz Albe :

> Nikita wrote:
> > 1. pg_total_relation_size return predictable approximately correct table
> size (apart from some preallocation) without subsequent
> VACUUM/ANALYZE/anything call as long as I don't DELETE/UPDATE rows;
> >
> > 2. DROP/TRUNCATE table reclaim disc space without subsequent VACUUM;
> >
> > These assumptions are correct? Is it fixed in doc if so?
>
> Sort of, if you consider the source as documentation.
>
> Ad pg_total_relation_size:
>
> See the function definition in backend/utils/adt/dbsize.c and
> the functions that it calls, especially "calculate_relation_size":
>
> PostgreSQL uses the stat(2) system call to determine the actual
> size of the files on disk.
>
> Ad TRUNCATE (from backend/commands/tablecmds.c):
>
>   /*
>* Need the full transaction-safe pushups.
>*
>* Create a new empty storage file for the relation, and assign it
>* as the relfilenode value. The old storage file is scheduled for
>* deletion at commit.
>*/
>
> Yours,
> Laurenz Albe
>