This architecture seems very similar to existing offerings in the "in-memory data grid" category, like Apache Ignite and Hazelcast. I'm more familiar with Ignit

This architecture seems very similar to existing offerings in the "in-memory data grid" category, like Apache Ignite and Hazelcast. I'm more familiar with Ignite (I built a toy Notion backend with it over a few afternoons in 2020). The way Ignite works overall is similar. You make a cluster of JVM processes, your data partitioned and replicated across the cluster, and you upload some JARs of business logic to the cluster to do things. Your business logic can specify locality so it runs on the same nodes as the relevant data, which ideally makes things a lot faster compared to systems where you need to pull all your data across the wire from a DB. Like Rama, Ignite uses a Java API for everything, including serializing and storing plain 'ol java objects. Ignite's architecture isn't focused on "ETL" into "PStates". Instead it's more about distributed "caches" of data. It does have streaming for ingestion (https://ignite.apache.org/docs/latest/data-streaming https://ignite.apache.org/docs/latest/data-streaming), but you can transactionally update the datastore directly (https://ignite.apache.org/docs/latest/key-value-api/transactions https://ignite.apache.org/docs/latest/key-value-api/transact...). It also has a "continuous query" feature for those reactive queries to retrieve data (https://ignite.apache.org/docs/latest/key-value-api/continuous-queries https://ignite.apache.org/docs/latest/key-value-api/continuo...). Rama's data-structure oriented PState index seems easier to work with than building indexes yourself on top of Ignite's KV cache, but Ignite also offers an SQL language, so you can insert your data into the KV cache however, add some custom SQL functions, and then accept more flexible SQL querying of your data compared to the very purpose-built PCache things, but still be able to do lower-level or more performance-oriented logic with data locality. Anyways, if you like some of this stuff but want to use an existing, already battle-tested open source project, you can look for these "in-memory data grid", "distributed cache", kind of projects. There's a few more out there that have similar JVM cluster computing models.