Re: should we document an example to set multiple libraries in shared_preload_libraries?

On Thu, Dec 9, 2021 at 7:42 AM Bharath Rupireddy
<bharath(dot)rupireddyforpostgres(at)gmail(dot)com> wrote:
> How about ALTER SYSTEM SET shared_preload_libraries command itself
> checking for availability of the specified libraries (after fetching
> library names from the parsed string value) with stat() and then
> report an error if any of the libraries doesn't exist? Currently, it
> just accepts if the specified value passes the parsing.

That certainly would have helped me. I guess it technically breaks the
theoretical use case of "first change the shared_preload_libraries
setting, then install those libraries, then restart Postgres," but I
don't see why anyone would do that in practice.

For what it's worth, I don't even feel strongly that this needs to be
an error—just that the current flow around this is error-prone due to
several sharp edges and could be improved. I would be happy with an
error, but a warning or other mechanism could work, too. I do think
better documentation is not enough: the differences between a working
setting value and a broken one are pretty subtle.

Thanks,
Maciek