AI Ecosystem Benchmark

Generalized framework for benchmarking AI tools that depend on database backends.

Python Package

This repository is also a Python package named ai-ecosystem-benchmark. The package is intentionally minimal for now while the benchmark framework takes shape.

Install and sync the development environment with uv:

uv sync

Run the local checks:

uv run pytest
uv run ruff check .
uv run ruff format --check .
uv run mypy
uv run deptry .

Install pre-commit hooks once per checkout:

uv run pre-commit install

Implementing a Workload

Benchmark workloads should subclass BaseBenchmarkWorkload and implement the required lifecycle hooks:

• setup(): prepare clients, data, or other state before benchmarks run.
• between_benchmarks(): reset or pause between benchmark executions.
• teardown(): clean up resources after benchmarks finish.

Connection strings are optional. If a connection string is omitted, that backend is assumed to be disabled for the benchmark run.

Backend-specific benchmark methods are discovered by name:

• Aerospike methods start with aerospike
• Postgres methods start with postgres
• Redis methods start with redis

from ai_ecosystem_benchmark import BaseBenchmarkWorkload


class MyWorkload(BaseBenchmarkWorkload):
    def __init__(
        self,
        aerospike_connection_string: str | None = None,
        postgres_connection_string: str | None = None,
        redis_connection_string: str | None = None,
    ) -> None:
        super().__init__(
            aerospike_connection_string=aerospike_connection_string,
            postgres_connection_string=postgres_connection_string,
            redis_connection_string=redis_connection_string,
        )

    def setup(self) -> None:
        return None

    def between_benchmarks(self) -> None:
        return None

    def teardown(self) -> None:
        return None

    def aerospike_test_insert(self) -> None:
        ...

    def redis_test_lookup(self) -> None:
        ...

Infrastructure

The first infrastructure target is a GCP Compute Terraform stack that can provision any combination of Redis, Postgres, and Aerospike benchmark clusters. The GitHub Actions workflow in .github/workflows/terraform.yml runs Terraform manually through workflow_dispatch and prints non-sensitive endpoint details after apply.

See infra/terraform/gcp-compute/README.md for setup details.

License

This project is licensed under the Apache License 2.0. See LICENSE for details.