I have been taking a look at different chain indexers (or APIs to indexers) to see what's available today and how they differ - see resource list below.
I still wanted to raise the question, what would be a good option for an event-based backend (not necessarily written in Haskell) that shall process on-chain events as they happen. In particular, we want to observe certain addresses and aggregate on-chain data using an CQRS / event sourcing architecture.
My current take
I wouldn't want to poll data via REST which makes Kupo and Blockfrost not an ideal candidate.
Oura looks like a great candidate but from what I see, it's not really indexing and much rather filtering on event types (like blocks, txs, assets etc.) and would still require some more processing on the backend to actually get the data I am interested in. What I like about it is that it support Node-to-Client (N2C) communication.
Scrolls as of now looks like a great candidate as it offers opinionated read-optimized views. I would imagine a Redis centered event store architecture combined with Redis Streams. The only thing I'd love support for is N2C which looks like is not yet implemented with version 0.4.2 mainly for improved performance.
Ogmios looks also very attractive as it would easily allow for a Node.js backend that could connect via websockets to directly read from the node. It's not an indexer I believe, but more an interface to the node. Hence, it requires additional filtering/ indexing in the backend similar to Oura.
Carp by dcSpark looks interesting too. It indexes and stores everything to a Postgres db, allows for N2C. Might very well be one of the strongest candidates and most likely complete in terms of development.
Lastly, Marconi looks like the best alternative to Scrolls & Carp unless I am forgetting about some other available resource. It's Haskell based and unfortunately, as of now has no event streams implemented. It will offer it in the near future.
In short, it looks like the available options are still very early in development but very promising. Nevertheless, I would appreciate feedback, things I've missed or other advice from the community!