Re: [HACKERS] PoC: custom signal handler for extensions - Mailing list pgsql-hackers

From Arthur Zakirov
Subject Re: [HACKERS] PoC: custom signal handler for extensions
Date
Msg-id 20180127191907.GA3317@arthur.localdomain
Whole thread Raw
In response to Re: [HACKERS] PoC: custom signal handler for extensions  (Maksim Milyutin <milyutinma@gmail.com>)
Responses Re: Re: [HACKERS] PoC: custom signal handler for extensions
List pgsql-hackers
Hello,

On Mon, Jan 22, 2018 at 02:34:58PM +0300, Maksim Milyutin wrote:
> ...
> I have attached a new version of patch and updated version of
> remote_effective_user function implementation that demonstrates the usage of
> custom signals API.

Thank you.

The patch is applied and build.

> +/*
> + * UnregisterCustomProcSignal
> + *      Release slot of specific custom signal.
> + *
> + * This function have to be called in _PG_init or _PG_fini functions of
> + * extensions at the stage of loading shared preloaded libraries. Otherwise it
> + * throws fatal error. Also it throws fatal error if argument is not valid
> + * custom signal.
> + */
> +void
> +UnregisterCustomProcSignal(ProcSignalReason reason)
> +{
> +    if (!process_shared_preload_libraries_in_progress)
> +        ereport(FATAL, (errcode(ERRCODE_INTERNAL_ERROR),
> +                        errmsg("cannot unregister custom signal after startup")));

Is there actual need in UnregisterCustomProcSignal() within _PG_init()?
An extension registers a handler and then unregister it doing
nothing. It seems useless.

Also process_shared_preload_libraries_in_progress within _PG_fini() will
be false I think. _PG_fini() won't be called though, because
implementation of internal_unload_library() is disabled.

Actually, is there need in UnregisterCustomProcSignal() at all? It
unregisters a handler only in current backend, for actual unergistering
we need to call it everywhere, if I'm not mistaken.

-- 
Arthur Zakirov
Postgres Professional: http://www.postgrespro.com
Russian Postgres Company


pgsql-hackers by date:

Previous
From: Dmitry Dolgov
Date:
Subject: Re: Write lifetime hints for NVMe
Next
From: Daniel Gustafsson
Date:
Subject: Re: [HACKERS] Refactoring identifier checks to consistently use strcmp