Automatic reconnection on memcached

[andreclimaco]

• Added module reconnecting.py with client ReconnectingClient:

  • Automatic reconnection on memcached network/server failures.

  • Optional monkey-patch replacing memcache.Client.

  • Configuration via environment variables:

    • MCDUTILS_DISABLE_RECONNECT (disables).
    • MCDUTILS_BACKOFF_MIN_MS and MCDUTILS_BACKOFF_MAX_MS (backoff intervals).
    • MCDUTILS_LOG / MCDUTILS_LOG_LEVEL (structured logs).
    • MCDUTILS_METRICS (enables internal metrics).

  • Prometheus metrics export via function export_prometheus_textfile.

• Implemented automatic retry in tpc_vote of MemCacheMapping:

  • New function _tpc_vote_with_retry encapsulates MemCacheError failures.

  • Configuration via environment variables:

    • MCDUTILS_DISABLE_TPC_RETRY (disables retry).
    • MCDUTILS_TPC_RETRY_ATTEMPTS (number of extra attempts).
    • MCDUTILS_TPC_RETRY_BACKOFF_MS (time between attempts).

  • Additional metrics registration: tpc_retry_attempts_total,
    tpc_retry_success_total, tpc_retry_fail_total and accumulated backoff times.

[andreclimaco]

I modernized the project structure by adding a Makefile for task automation, updating the test workflow (.github/workflows/tests.yml), and updating the documentation.

[dataflake]

@andreclimaco I really appreciate the contribution, but I have a problem when you mix the functional changes and tests with package configuration and style changes.

What you did here was basically throw out all the tooling and configuration files that I use to maintain this package and impose your own choices and styles. With all due respect for this work, I find that inacceptable. I am (and will continue to be) the maintainer of this package so I am asking everyone else to respect the choices I have made for package management, coding style and the toolchain I use.

I am happy to review and incorporate the FUNCTIONAL changes once you have removed all changes to package management-related files and removed all changes that are about style.