1. Split app into backend / frontend
2. Use cachepusher
3. ???
4. Profit

• git clone https://github.com/bnadland/cachepusher
• cd cachepusher
• vagrant up
• vagrant ssh -c "tail -f /var/log/cachepusher/*"

• go get github.com/bnadland/cachepusher/...
• cd $GOPATH/src/github.com/bnadland/cachepusher
• vagrant up
• vagrant ssh -c "sudo supervisorctl stop cp"
• go run syncer/main.go

There is a customer and a address table in postgresql. There is a python script fakedata.py that simulates a backend and inserts some customers and addresses into postgresql. There is database trigger on both tables, so whenever one is updated a event on the customer_updated channel is fired with the id_customer as payload. If a customer row gets deleted the trigger fires an event on customer_deleted also with the id_customer as a payload.

The cachepusher program (code in ./syncer/main.go) opens a connection to postgresql and listens on those two channels.

The denormalized data in redis uses the id_customer as a cachekey (i.e. customer:1 for id_customer=1) and contains a json datastructure with the customer data and embedded into that the corresponding addresses.

Whenever a customer_deleted event is fired, cachepusher deletes the key from redis.

When a customer_updated event is fired, cachepusher calls a stored procedure with the id_customer from the payload and stores the returned json directly into redis.

There is also a cache warmup functionality that gets called when the cachepusher program starts up: It starts with deleting all keys under it's control (i.e. customer:*), then sets up the listener to get updates and the calls a stored procedure that 'touches' all customer records by inserting a new timestamp into the touched_at column which then gets propagated to the redis instance via the normal listener functionality described above.

MIT License