Ability to query available disk space

[Steve Chavez]

Problem

Indexing a vector column can outgrow available disk space.

Proposal

Expose a function that can query available disk space. Later create a wrapper CREATE INDEX statement that aborts it if the available space is not enough.

Prior art

• https://github.com/greenplum-db/diskquota
• https://github.com/hlinnaka/pg_quota

[Steve Chavez]

Edit: On second thought, I'll just create some functions. It's too hacky to do this on GUCs which are meant to be static.

To not depend on a table or function (thus on a CREATE EXTENSION supautils) to get this info, we can store it as a PGC_INTERNAL GUC. So the interface to get this info would be:

select current_setting('supautils.free_disk_space', true);

Steps for implementation:

• Get the data_directory from pg
• Use https://man7.org/linux/man-pages/man3/statvfs.3.html to get free space.
• Store it on supautils.free_disk_space.

[Oliver Rice]

could we add number of cpus and total memory here too, or would those be better as separate issues?

[Steve Chavez]

Sure, we can track those here.

• The number of cpus is one function call https://linux.die.net/man/3/get_nprocs.
• Total memory can be obtained with https://linux.die.net/man/2/sysinfo.

[Mark Burggraf]

Was this ever implemented?

[Oliver Rice]

bump @steve-chavez ^

[Steve Chavez]

This was done only for the case of constrained extensions, which was the immediate use case: https://github.com/supabase/supautils#constrained-extensions (see disk)

The code for getting disk space is there but not sure if we should add SQL objects to supautils since it would make it harder to upgrade (we'd need migrations). Right now it only works with db settings (GUCs).

@burggraf What's the use case you had in mind?