Ensure a single Faraday client is used per Client

[redox]

Purpose

This PR ensures a single Faraday HTTP client is used per Trino::Client instance.

Overview

For now, each start, resume or kill query would instantiate a new Faraday client by calling Trino::Client.faraday_client at the query level.

This PR refactors the logic of the Trino::Client to ensure a single Faraday client is used per instance of Trino::Client and set all the query (and user) specific HTTP headers at query time. This avoids the need to reconnect to the Trino server dealing with all the SSL handshake processing time at each query + allow the usage of faraday-net_http_persistent to benefit from HTTP keep-alive.

While the benefit of saving a few ms at each Trino statement query might not be that visible for the "long running queries" use-cases (where the actual Trino query anyway takes up to a few seconds to answer), we've seen large improvements for use-cases where the queries are taking a few dozens of ms.

Checklist

• Code compiles correctly
• Created tests which fail without the change (if possible)
• All tests passing
• Extended the README / documentation, if necessary

[yuokada]

@redox Thanks for your contribution. Could you check and fix the errors reported on GitHub Actions?

[redox]

@redox Thanks for your contribution. Could you check and fix the errors reported on GitHub Actions?

Yeah not sure why this started complaining but I'm adding the libcurl4-openssl-dev deps which should solve this. Never easy to iterate as for every single commit I add to the PR it requires your approval to run the CI pipeline :/

[yuokada]

Could you reformat files with the command below?

bundle exec standardrb --fix

[redox]

Could you reformat files with the command below?

Gosh of course!

[p623]

Hi @redox, thank you for the original proposal and implementation.

Since this PR has not been updated for over a year, I opened #168 with a fresh implementation based on the same approach, updated for the current codebase. I also added more extensive test coverage.

I referenced this PR and credited your original work in the description of #168.

@yuokada, would it be okay to continue the discussion and review in #168?