Add ability to specify Postgres connection credentials

[michaelbautin]

• Allow specifying PostgreSQL credentials when running against an external instance of PostgreSQL not managed by the benchmark.
• Allow avoiding an attempt to start PostgreSQL service with systemctl, because it might be managed in a different way.
• Create the vector PostgreSQL extension automatically if it does not exist.

Fixes: #563

[maumueller]

Thanks for your contribution, @michaelbautin. I'm a bit skeptical about the proposed solution, because it makes this very pronounced from an API point of view, while it only applies to a very small subset of the algorithms.

Can you achieve the same functionality by specifying these credentials in the config file https://github.com/erikbern/ann-benchmarks/blob/main/ann_benchmarks/algorithms/pgvector/config.yml? If I remember correctly, you can provide it in the base_args entry which is propagated to the constructor of the class.

[michaelbautin]

@maumueller Thank you for your feedback. I've reworked the PR to avoid introducing generic command-line parameters, and instead made it controllable by Postgres-specific environment variables. Also I've added automatic pgvector extension creation and a way to avoid attempting to start PostgreSQL service if it is managed in a different way.

Cc @erikbern

[maumueller]

I think that looks good, @michaelbautin. I suppose your use-case is that the implementation is run locally outside of the Docker environment? Otherwise it seems a bit difficult to pass these variables into the container with the current setup.

[michaelbautin]

@maumueller yes, my use case is that the Postgres server is not managed by ann-benchmarks scripts. In my experiments I frequently need a Postgres server set up in a very specific way in a cloud environment.

[maumueller]

Alright, I can merge this if you agree that it provides the necessary functionality. One last thing: Do you think a short comment in the module.py would help others find out about this feature? (And if yes, could you please add one? Thanks!)

[michaelbautin]

@maumueller : sounds great, thank you! I will add the comment shortly.

[michaelbautin]

@maumueller I've added the requested comment. Thank you!

[maumueller]

Thank you for the contribution, @michaelbautin!