The way we currently transition from e.g. fast sync to consensus is very messy (the reactors enable/disable message processing and call methods directly each other). This will become even more messy when we implement state sync (#828) and with fallback to fast sync for lagging nodes (#129).

I suggest we instead have a concept of node phases (e.g. WAL replay, block replay, state sync, fast sync, consensus, and so on), and an overall manager which enables/disables various reactors and processes based on heuristics (e.g. switch to fast sync if consensus is lagging). This requires support for announcing changes in P2P channels (#4394).

This is somewhat related to the idea of modes (#2237).

For Admin Use

• Not duplicate issue
• Appropriate labels applied
• Appropriate contributors tagged
• Contributor assigned/self-assigned