Go Concurrency Patterns

Concurrency is hard, but Go makes it manageable

5 October 2021

Introduction

Concurrency is hard. It doesn't matter if you're a novice or an expert: even experts regularly get it wrong.

Kenton Varda - Protobuf/Cap'n Proto creator, Cloudflare Workers Tech Lead

• Race conditions do happen and there are multiple ways to avoid them

• In short: be humble, avoid locks

Channels

• Based on communicating sequential processes (CSP) - Tony Hoare, 1978

• Channels are thread-safe - sometimes that's not obvious - but they do block

• Unbuffered channel:

Channels (cont.)

• Buffered channel:

• Avoid blocking with the select statement, which is like switch/case, but for channels
• Timeouts can be handled with time.After() and cancellation with ctx.Done()

Synchronization Packages

sync

• Once
• Map
• Avoid WaitGroup if possible. It blocks and can cause deadlocks
  • A massive amount of goroutines can also be a performance bottleneck

sync/atomic

• Uses thread-safe lock-free CPU instructions under the hood
• Limited to integers and pointers
• Value

Conclusion

• Avoid double-initialization issues with sync.Once
• Try to use channels, avoid shared state
  • "Share memory by communicating; don't communicate by sharing memory"
• Avoid channel blocking with select
• Try higher-level options like sync.Map
• Try lock-free solutions with atomic.Value
• Only use locks/mutexes if, and only if, none of the above worked

References

1. Cloudflare Blog - Durable Objects: Easy, Fast, Correct - Choose three
2. Go by Example: Non-Blocking Channel Operations
3. Golang UK Conference 2017 | Arne Claus - Concurrency Patterns in Go
4. Stack Overflow - What are channels used for?

Thank you