Indexer infrastructure becomes more complex from day to day, new components introduce changes to the architecture which can cause an impact on the whole system performance.
Benchmark framework and tools for modeling the workload profile can help us to plan our operations based on a clear understanding of different components performance.
It might be a good idea to build some benchmark subgraphs with a different set of features that we can test under synthetic load.
As discussed in discord, synthetic benchmarks could be super useful to improve indexer perfomance.
We would need some basic contracts to deploy on a Ganache environment, and 3-4 simple subgraphs that use the different contracts with different features (and maybe some with feature combinations).
We could then simulate different scenarios, like having the ganache instance on the same machine, different machine, different network, having different specs on the graph-node, horizontal scaling of the graph-nodes, etc.
It would also be super cool that once we have benchmarked infrastructures, to use them to benchmark and pinpoint specific subgraph bottlenecks, so we can derive a list of potential optimizations for them, but that’s a topic for another thread.