Rafael Cárdenas

Following our plan to move to microservices with an event-driven architecture, we should consider the use of [Kafka](https://kafka.apache.org/) to emit events that all other services could listen for.

Rosetta endpoints have their own logic in terms of structure, call flow, etc. They're mostly read-only endpoints with the exception of transaction broadcasts which should go directly to the new...

This would be handled only by nginx reverse proxy configs: https://github.com/hirosystems/devops/issues/983 Nginx will also handle any necessary logging for transaction tracing

Refactor the WebSockets/Socket.IO servers into a separate microservice

1

Since we want to expand the capabilities of our websocket events, it would be a better idea to spin them off into a separate service that can handle its own...

### Background Our Stacks API currently exposes [a few endpoints](https://docs.hiro.so/api#tag/Fungible-Tokens/operation/get_contract_ft_metadata) that attempt to serve FT metadata. This data is pulled by a background queue that listens to new contracts as...

add cache handlers for individual items (blocks, microblocks, principals)

2

We currently have two cache handlers that we use to create ETags for most of our endpoints: 1. **Chain tip**: Latest block/microblock hash 2. **Mempool**: Hash digest of all non-pruned...

Optimize event replay process

1

The current `import-events` process takes ~18 hours to complete in the current production deployments. It requires the importing of the complete node event TSV into our tables while calculating re-orgs,...

When a transaction is re-orgd because a new block came by which made it non-canonical, it is normally thrown back into the mempool so we show it again as `pending`....

switch `PgNotifier` from NOTIFY/LISTEN to pg realtime subscription

1

We've been having some replication problems recently which are reported by postgres as "deadlocks" on certain queries. In order for us to be able to try other replication modes, we...