This is a great initiative. Benchmarking managed services is notoriously tricky due to varying configurations and 'black-box' optimizations. For anyone looking to run these benchmarks on their own or wanting to dive deeper into the why behind the numbers, pgmetrics (https://pgmetrics.io) is a fantastic open-source tool. It collects a massive amount of internal PG stats into a structured JSON format, making it easy to see exactly what's happening during a load test. For a more persistent and visual way to track these metrics across different providers, pgDash (https://pgdash.io) is built on top of that same data.
Great write-up—the distinction between healthy OS caching and actual memory pressure is often misunderstood. To get a granular view of where your memory is actually going (shared buffers, cache hits, etc.) without the overhead of heavy agents, pgmetrics (https://pgmetrics.io) is very effective. If you need to track these metrics over time to catch when 'good' caching turns into 'bad' pressure, pgDash (https://pgdash.io) builds directly on top of it to provide a specialized monitoring dashboard with zero-config setup.
This looks like a great addition to the Postgres ecosystem. When adding specialized extensions like this, it's always worth keeping an eye on how they impact overall system performance, especially memory usage and lock contention as the dataset grows. For anyone testing this out, I'd recommend using an open-source tool like *pgmetrics* (https://pgmetrics.io) to get a baseline and then monitor how the new indexes and search workloads affect your underlying metrics. It’s zero-dependency and gives you a very deep look into the internals without much overhead.