Comment by lucisferre | Hacker News Reader

Comment by lucisferre | Hacker News Reader

Parent Full thread

I know Greg personally, and I have heard him say things like "you don't need a library" in the past. However despite the fact that the basics are simple there are a lot of repeatable concepts that make sense to centralize around and even standardize. There is nothing wrong with a library, just as it is possible to do it without.

One thing that absolutely needs library/drivers is persistence of the event store. I say this having built my own as well as contributing to open source ones in the past. It is still a hard and not well solved problem to do this well. Again it can be simple, but in the real world it isn't usually that simple.

Tools are useful, tools are important. If CQRS/ES wants to make it into mainstream it needs to think about tooling. So far I have not been impressed with the people in the community hand waving about this and not having much to show for it.

> If CQRS/ES wants to make it into mainstream it needs to think about tooling.

I agree with this entirely. CQRS/ES applications really help from tailored tooling. Tracking chains of events using correlation and causation identifiers. Auditing commands. Projecting events out to datastores like Elasticsearch. Then using Kibana[1] to create dashboards to drill down into the data. I'd like to go into this in more detail in the future. Revisit this case study after 6-12 months and describe the pitfalls, and countermeasures. Identify tools that allow you to monitor the system proactively.

[1] https://www.elastic.co/products/kibana

> One thing that absolutely needs library/drivers is persistence of the event store. I say this having built my own as well as contributing to open source ones in the past. It is still a hard and not well solved problem to do this well. Again it can be simple, but in the real world it isn't usually that simple.

Would it also be such a hard thing to do if you can delegate the actual persistence to something like a rdbms? What are the typical pitfalls?

Having built a few RDBMS-based event stores - it's pretty easy to do. There are two parts that I can think of that are not completely trivial - appending events to the stream while respecting "expected version" modifiers with optimal concurrency and allowing for fast and light subscriptions to new events (e.g. LISTEN/NOTIFY in Postgres, ringbuffer on the client).

The good thing is that you can adapt the event store to your performance requirements and do the simplest thing possible in a huge amount of cases.

Reply on news.ycombinator.com