Clarification on when _PG_init() is invoked for extensions - Mailing list pgsql-hackers

From Ayush Vatsa
Subject Clarification on when _PG_init() is invoked for extensions
Date
Msg-id CACX+KaPSTwU50rhTcoKEBoRgnPUA=kTuZA_SgyDZTYKYSwFX5g@mail.gmail.com
Whole thread Raw
List pgsql-hackers

Hello Hackers,

I’m working on a PostgreSQL extension that defines GUCs and initializes hooks inside the _PG_init() function,
and I wanted to confirm my understanding of when _PG_init() is called and whether the library remains loaded
across sessions.

Based on documentation and experimentation, I believe the behavior works as follows, but I would appreciate 
confirmation or corrections:

1/ When the extension is listed in shared_preload_libraries_PG_init() is executed before the server
starts accepting connections, and hooks are initialized for every backend process. The downside is that this enables
hooks globally for all databases in the instance.

2/ When using LOAD 'extension': My understanding is that the library loads only in the backend executing the LOAD
command, and _PG_init() runs just for that backend. I’m unsure whether this affects existing and future backends for
the same database, or if it is strictly per-backend and non-persistent. I also believe this mechanism cannot be used if
the extension needs to register a background worker.

3/ When running CREATE EXTENSION: Here the .so is loaded lazily when a function defined with MODULE_PATHNAME is
first executed or defined. In this model, I’m unsure whether I need to explicitly run LOAD 'extension' to ensure
_PG_init() runs for other existing and upcoming backends. Also wanted to check if I need LOAD extname in my sql script?

My use case:
I would like the extension to initialize hooks automatically, but only for all (existing and upcoming) backends of the specific
database where the extension is installed - not globally across all databases.

If someone could confirm whether my understanding is correct
Thanks in advance for your guidance.
Regards,
Ayush

pgsql-hackers by date:

Previous
From: Amit Kapila
Date:
Subject: Re: DOCS: Missing tags for some SEQUENCE fields
Next
From: John Naylor
Date:
Subject: Re: Proposal for enabling auto-vectorization for checksum calculations