Let's GO
Scale Backend Service Development in Go
Agenda
- Current State of FB backend services
- Scalability Challenges
- Where Dynamic Languages Fall Short
- Go: A New Kind of Static Language
- How Go helps us scale
- Where to GO from here
Agenda
- Current State of FB backend services
- Scalability Challenges
- Dynamic vs Static
- How Go helps us scale
- Where to GO from here
Agenda
- Current State of FB backend services
- Scalability Challenges
- Dynamic vs Static
- How Go helps us scale
- Where to GO from here
Scalability Challenges
Performance
Development
- Larger Code Base
- Order of Magnitude More Devs
Benefits of Dynamic Languages
- More Products Faster Initially
- Expressive , Convenient , "Fun"
Dynamic Language
Where Do They Fall Short
- Much harder to reason about code
- Dynamic type = hard to optimize
- Interpreted = overhead
- JIT not always applicable
Dynamic Language
The techniques you use to build a hut is not the same technique you use to build the skyscraper
- Joshua Bloch
GO Static?
public static void main(String[] argv) protected <T extends CWBaseModel, S extends GridPresenter.Display<T>, Q extends GridPresenter<T, S>> void doBindGridPresenter(final Q p) {
...
}GO Static
Fewer Ceremonies
- Type inferencing
- Small set of features
GO Static
Batteries Included
- Large standard library
- net/http
- logging
- crypto
- Beloved everyday data structures
- map
- slice
- Built-in concurrency primitives
- channel
- goroutine
- Garbage-collected
GO Static
Interfaces
type Duck interface {
Quack()
}
type Duckula struct {}
func (d Duckula) Quack() {...}type Donald struct {}
func (d Donald) Quack() {...}
func QuackTogether(ducks []Duck) {
for _, duck := range ducks {
duck.Quack()
}
}
GO Static
Don't Inherit, Embed
GO Static
Errors Ain't Exceptions
- Put more thoughts into dealing with errors
- panic/recover for true exceptions
Scaling Backend Dev
Removes bikeshedding
go fmt
Scaling Backend Dev
Type-Checking and Code Reasoning
- Explicit Contracts
- Reveal bugs early
- Maintainability for large code base & dev team
Scaling Backend Dev
Simple Deployment
-
RVM/Rake?
- Virtualenv, setuptools, pip, distutils?
- Move binary to the server, use process mgmt
- systemd, upstart, supervisord, etc
Scale Backend Dev
Near Metal Performance
- Compiles to platform-specific MC
- Interpreter process
- Shared library loading
- Virtual machine
- [chart]
We're Not Alone
- YouTube
- Docker
- Heroku
- SoundCloud
- Disqus
- Cloudflare
Where To GO From Here?
let's go
By Kevin Jing Qiu
