'Instead of debugging the one environment, I'll just reimplement the entire service!'
Regardless, I still find the potential lock-in costs and issues incurred with a proprietary datastore like FireBase to be a total and complete deal breaker.
For truly production-grade systems, relinquishing control over your data stores does not seem like a viable path to better reliability or control for any business with significantly large swaths of data.
If there were a compatible open-source equivalent with easy import functionality for a FireBase dump, I'd be singing a different tune.
Am I missing or overlooking something here? I get the use case for a small project or app, but for mission-critical production, it doesn't seem like a good idea.
I still think it's a cool project.
Before we (https://www.tesla.im) switched to Firebase we had (still have) our own realtime system built with NodeJS, Websockets, Amazon SQS (as backup to ensure messages don't get lost) and a RIAK cluster.
2 NodeJS app servers
1 Redis Master + 1 Slave
3 RIAK cluster (recommended 5)
You might want to checkout https://www.meteor.com/ as well.
Edit: I see what you mean: Pouchdb-server  is a couchdb-compatible server that speaks the same sync protocol. If you don't want to install it, you can work with any hosted couchdb-speaking cloud instance, such as iriscouch  or cloudant 
I wonder why real-time features aren't more popular? Real-time order status, live-updating maps, the list goes on!