13.3K ·GitHub
What is this?
groupcache is a distributed caching and cache-filling library for Go, built as a replacement for a pool of memcached nodes in many cases.
groupcache is a distributed caching and cache-filling library for Go, built as a replacement for a pool of memcached nodes in many cases.Is this a standalone server like memcached, or something I embed in my code?It's a Go library, not a separate server. It's a client library as well as a server that connects to its own peers, forming a distributed cache directly inside your application processes.
It's a Go library, not a separate server. It's a client library as well as a server that connects to its own peers, forming a distributed cache directly inside your application processes.How do I add it to a project?Import it as a Go package and configure a group of peers; check the API docs and examples for wiring it into your app.
Import it as a Go package and configure a group of peers; check the API docs and examples for wiring it into your app.Do I need Go, or can I use it from another language?Go only — the library is currently only available for Go, and the author says it's very unlikely to be ported to another language.
Go only — the library is currently only available for Go, and the author says it's very unlikely to be ported to another language.Can I update or expire a cached value once it's set?No. groupcache does not support versioned values, cache expiration, or explicit evictions, so there's no CAS or Increment/Decrement either — once key 'foo' is 'bar', it must always be 'bar'.
No. groupcache does not support versioned values, cache expiration, or explicit evictions, so there's no CAS or Increment/Decrement either — once key 'foo' is 'bar', it must always be 'bar'.Who actually relies on this in production?Google uses it internally: dl.google.com (its original user), parts of Blogger, Google Code, Google Fiber, and Google's production monitoring systems.
Google uses it internally: dl.google.com (its original user), parts of Blogger, Google Code, Google Fiber, and Google's production monitoring systems.How does groupcache avoid the thundering herd problem memcached has on cache misses?It coordinates cache fills so that only one load, in one process across the whole replicated set, populates the cache for a given key, then multiplexes the loaded value to every other caller waiting on it.
It coordinates cache fills so that only one load, in one process across the whole replicated set, populates the cache for a given key, then multiplexes the loaded value to every other caller waiting on it.What happens if the peer that owns a key is unreachable?If the RPC to the owning peer fails, the requesting node just loads the value locally instead, still deduplicating concurrent local loads.
If the RPC to the owning peer fails, the requesting node just loads the value locally instead, still deduplicating concurrent local loads.How does groupcache pick which node owns a given key?It shards by key using consistent hashing across the set of peers, similar to how memcached shards, so each key maps deterministically to one owning node.
It shards by key using consistent hashing across the set of peers, similar to how memcached shards, so each key maps deterministically to one owning node.Does groupcache help with 'hot' keys that get hammered by traffic?Yes — it automatically mirrors super-hot items to multiple processes, preventing the hot-spotting issue where one memcached machine's CPU or NIC gets overloaded by a popular key.
Yes — it automatically mirrors super-hot items to multiple processes, preventing the hot-spotting issue where one memcached machine's CPU or NIC gets overloaded by a popular key.What are similar projects?go-redis, dragonfly, twemproxy, phpredis, 7days-golang, valkey, garnet, memcached
go-redis, dragonfly, twemproxy, phpredis, 7days-golang, valkey, garnet, memcachedWhich topics is it in?cache, caching, google, golang, distributed-systems, backend, database, networking, performance, go, api, server
cache, caching, google, golang, distributed-systems, backend, database, networking, performance, go, api, server