- Matthew Oden Potter
I quickly threw together a method for zipping/reducing across collections1, but also warned them about breaking the Golden Rule of Node:
Don't do work in node. Describe work in node.
Doing work is data manipulation
Node servers are, for the most part, single-threaded applications. They depend heavily on non-blocking I/O for performance. Every time you rework data as part of your server's response, you're hold up the queue of incoming requests.
Of course, it's possible to spin up child processes and delegate this work to another thread. But sending data back and forth over IPC is slow, and requires serializing/unserializing on each process, which blocks both threads during that time. It would seem that this kind of delegation just creates more work.2
Describing work is message-passing
Node shines as a message broker, transforming small requests into streaming replies. Think of your node app as a middle-manager, gathering requirements, handling bad requests, and delegating responsibility.
Practices and patterns for describing work:
Avoid formatting the incoming request. Clients should pass in data exactly as you need it.
Get your data from the database exactly as your client expects it. If you need the data sorted, make that part of the query. If you need it aliased, and grouped? Do it in the query. Your database is always going to be faster at organizing data than your server.
Leave database responses as they are. If you've asked for exactly what you need, there's no point in mapping, filtering, or sorting the results.
Use static, parameterized queries whenever possible. Custom string-building/ query-building is just more logic to keep track of, and more work to be done on every call.
If you don't have any say over the format of the data (maybe you're calling another API), then minimize the work you have to do. Use streams and sequences to manipulate the data as it arrives, rather than all at once.
In short, a valid request from the client should map to a static, parameterized query. The database's response should be attach onto a parameterized reply. Outside of basic validation, your server shouldn't ever care about the data passing through it.
1 Regardless of whether or not something's a good idea, sometimes you just need to get things done.2 This is part of why scaling out a Node service is typically done by spinning up more servers to run in parallel, rather than a using a single server that manages multiple threads.