teracache

README

Teracache

Teracache is a scalable, decentralized, highly-available in-memory cache for read-asymmetrical workflows.

Workflow

The workflow diagram depicts a system with multiple topics and nodes where each color represents the topic they're in.

What is Read-Asymmetrical?

In a read-asymmetrical workflow, reads are frequent while the writes are rare. For example a newsfeed will resize and optimize a thumbnail when a new article arrives and then serve the optimized thumbnail to thousands or millions of users.

In such cases, it is very profitable to use a cache that doesn't sacrifice performance to allow write operations. Teracache does exactly this trade-off.

How to use invalidations?

Cache entries are immutable and no entry is considered stale. Entries are evicted only to admit more popular entries.

Inspired By

• groupcache
• SWIM Membership Protocol
• Microsoft Orleans
• Amazon Dynamo
• Consistent Hashing and Random Trees

Documentation

Index

• Constants
• type Interface
  • func New(t Topic) (Interface, error)
• type Provider
• type Topic

Constants

const (
	RouterPort  = 20274
	ServicePort = 20275
)

Types

type Interface

type Interface interface {
	Get(context.Context, string) (payload.Payload, error)
}

func New

func New(t Topic) (Interface, error)

New creates a cache instance that participates in the given topic. Once a Get request to the cache fails, the cache uses the topic provider function to retrieve the missing element. The cache is bootstrapped using the given peer addresses.

type Provider

type Provider func(ctx context.Context, key string) (payload.Payload, error)

type Topic

type Topic struct {
	ID       string
	Capacity int
	Peers    []string
	Provider Provider
}