Package cache implements expirable cache.
- Support LRC, LRU and TTL-based eviction.
- Package is thread-safe and doesn't spawn any goroutines.
- On every Set() call, cache deletes single oldest entry in case it's expired.
- In case MaxSize is set, cache deletes the oldest entry disregarding its expiration date to maintain the size, either using LRC or LRU eviction.
- In case of default TTL (10 years) and default MaxSize (0, unlimited) the cache will be truly unlimited and will never delete entries from itself automatically.
Important: only reliable way of not having expired entries stuck in a cache is to run cache.DeleteExpired periodically using time.Ticker, advisable period is 1/2 of TTL.
This cache is heavily inspired by hashicorp/golang-lru simplelru implementation. v3 implements simplelru.LRUCache interface, so if you use a subset of functions, so you can switch from github.com/hashicorp/golang-lru/v2/simplelru or github.com/hashicorp/golang-lru/v2/expirable without any changes in your code except for cache creation. Key differences are:
- Support LRC (Least Recently Created) in addition to LRU and TTL-based eviction
- Supports per-key TTL setting
- Doesn't spawn any goroutines, whereas
hashicorp/golang-lru/v2/expirablespawns goroutine which is never killed (as of now) - Provides stats about hits and misses, added and evicted entries
package main import ( "fmt" "time" "github.com/go-pkgz/expirable-cache/v3" ) func main() { // make cache with short TTL and 3 max keys c := cache.NewCache[string, string]().WithMaxKeys(3).WithTTL(time.Millisecond * 10) // set value under key1. // with 0 ttl (last parameter) will use cache-wide setting instead (10ms). c.Set("key1", "val1", 0) // get value under key1 r, ok := c.Get("key1") // check for OK value, because otherwise return would be nil and // type conversion will panic if ok { rstr := r.(string) // convert cached value from interface{} to real type fmt.Printf("value before expiration is found: %v, value: %v\n", ok, rstr) } time.Sleep(time.Millisecond * 11) // get value under key1 after key expiration r, ok = c.Get("key1") // don't convert to string as with ok == false value would be nil fmt.Printf("value after expiration is found: %v, value: %v\n", ok, r) // set value under key2, would evict old entry because it is already expired. // ttl (last parameter) overrides cache-wide ttl. c.Set("key2", "val2", time.Minute*5) fmt.Printf("%+v\n", c) // Output: // value before expiration is found: true, value: val1 // value after expiration is found: false, value: <nil> // Size: 1, Stats: {Hits:1 Misses:1 Added:2 Evicted:1} (50.0%) }For detailed benchmarks comparing different versions and cache implementations, see the benchmarks directory.
Based on all the benchmarks across four different caching libraries:
-
go-pkgz/expirable-cache remains the best overall option:
- Excellent performance across all operations
- Lowest memory usage and allocations
- Type safety through generics
- Clean API with method chaining
- Simple implementation
-
dgraph-io/ristretto is a strong contender for specific use cases:
- Great performance for read-heavy workloads
- Sophisticated memory management for very large caches
- Built-in metrics and statistics
- Designed for high-concurrency environments
-
patrickmn/go-cache is still fastest for pure raw performance but lacks modern features, and leaks goroutines
-
jellydator/ttlcache lags behind in performance compared to all other options.
v2 and v3 use Go generics and achieve significant performance improvements over v1:
- v2 is approximately 28-42% faster than v1 for basic operations
- v3 maintains the performance gains of v2 while being compatible with the Hashicorp
simplelruinterface - Recent optimizations have improved performance across all versions
| Operation | v1 | v2 | v3 | Improvement v1→v3 |
|---|---|---|---|---|
| Random LRU (no expire) | 188.3 ns/op | 127.5 ns/op | 132.3 ns/op | ~30% faster |
| Frequency LRU (no expire) | 180.3 ns/op | 127.4 ns/op | 128.1 ns/op | ~29% faster |
| Random LRU (with expire) | 191.9 ns/op | 129.7 ns/op | 130.7 ns/op | ~32% faster |
| Frequency LRU (with expire) | 181.3 ns/op | 126.7 ns/op | 131.2 ns/op | ~28% faster |
Recent benchmarks comparing expirable-cache with other popular Go caching libraries:
| Operation | go-pkgz/expirable-cache | patrickmn/go-cache | jellydator/ttlcache | dgraph-io/ristretto |
|---|---|---|---|---|
| Set | 69.14 ns/op | 82.67 ns/op | 448.8 ns/op | 820.0 ns/op |
| Get | 78.12 ns/op | 63.81 ns/op | 190.9 ns/op | 84.23 ns/op |
| Set+Get | 66.62 ns/op | 67.94 ns/op | 253.9 ns/op | 198.2 ns/op |
| Real-world scenario | 78.83 ns/op | 70.24 ns/op | 198.0 ns/op | 83.40 ns/op |
| Memory allocations | Lowest | Low | Medium | Highest |
v1 benchmark results
~/expirable-cache ❯ go test -bench=. goos: darwin goarch: arm64 pkg: github.com/go-pkgz/expirable-cache BenchmarkLRU_Rand_NoExpire-8 4494738 272.4 ns/op --- BENCH: BenchmarkLRU_Rand_NoExpire-8 cache_test.go:46: hit: 0 miss: 1 ratio: 0.000000 cache_test.go:46: hit: 1 miss: 99 ratio: 0.010000 cache_test.go:46: hit: 1352 miss: 8648 ratio: 0.135200 cache_test.go:46: hit: 248678 miss: 751322 ratio: 0.248678 cache_test.go:46: hit: 1121791 miss: 3372947 ratio: 0.249579 BenchmarkLRU_Freq_NoExpire-8 4612648 261.6 ns/op --- BENCH: BenchmarkLRU_Freq_NoExpire-8 cache_test.go:74: hit: 1 miss: 0 ratio: 1.000000 cache_test.go:74: hit: 100 miss: 0 ratio: 1.000000 cache_test.go:74: hit: 9825 miss: 175 ratio: 0.982500 cache_test.go:74: hit: 312345 miss: 687655 ratio: 0.312345 cache_test.go:74: hit: 1414620 miss: 3198028 ratio: 0.306683 BenchmarkLRU_Rand_WithExpire-8 4109704 286.5 ns/op --- BENCH: BenchmarkLRU_Rand_WithExpire-8 cache_test.go:99: hit: 0 miss: 1 ratio: 0.000000 cache_test.go:99: hit: 0 miss: 100 ratio: 0.000000 cache_test.go:99: hit: 1304 miss: 8696 ratio: 0.130400 cache_test.go:99: hit: 248310 miss: 751690 ratio: 0.248310 cache_test.go:99: hit: 1027317 miss: 3082387 ratio: 0.249973 BenchmarkLRU_Freq_WithExpire-8 4341217 279.6 ns/op --- BENCH: BenchmarkLRU_Freq_WithExpire-8 cache_test.go:127: hit: 1 miss: 0 ratio: 1.000000 cache_test.go:127: hit: 100 miss: 0 ratio: 1.000000 cache_test.go:127: hit: 9868 miss: 132 ratio: 0.986800 cache_test.go:127: hit: 38221 miss: 961779 ratio: 0.038221 cache_test.go:127: hit: 37296 miss: 4303921 ratio: 0.008591 PASS ok github.com/go-pkgz/expirable-cache 18.307s v3 benchmark results
~/Desktop/expirable-cache/v3 master !2 ❯ go test -bench=. goos: darwin goarch: arm64 pkg: github.com/go-pkgz/expirable-cache/v3 BenchmarkLRU_Rand_NoExpire-8 7556680 158.1 ns/op --- BENCH: BenchmarkLRU_Rand_NoExpire-8 cache_test.go:47: hit: 0 miss: 1 ratio: 0.000000 cache_test.go:47: hit: 0 miss: 100 ratio: 0.000000 cache_test.go:47: hit: 1409 miss: 8591 ratio: 0.140900 cache_test.go:47: hit: 249063 miss: 750937 ratio: 0.249063 cache_test.go:47: hit: 1887563 miss: 5669117 ratio: 0.249787 BenchmarkLRU_Freq_NoExpire-8 7876738 150.9 ns/op --- BENCH: BenchmarkLRU_Freq_NoExpire-8 cache_test.go:75: hit: 1 miss: 0 ratio: 1.000000 cache_test.go:75: hit: 100 miss: 0 ratio: 1.000000 cache_test.go:75: hit: 9850 miss: 150 ratio: 0.985000 cache_test.go:75: hit: 310888 miss: 689112 ratio: 0.310888 cache_test.go:75: hit: 2413312 miss: 5463426 ratio: 0.306385 BenchmarkLRU_Rand_WithExpire-8 6822362 175.3 ns/op --- BENCH: BenchmarkLRU_Rand_WithExpire-8 cache_test.go:100: hit: 0 miss: 1 ratio: 0.000000 cache_test.go:100: hit: 0 miss: 100 ratio: 0.000000 cache_test.go:100: hit: 1326 miss: 8674 ratio: 0.132600 cache_test.go:100: hit: 248508 miss: 751492 ratio: 0.248508 cache_test.go:100: hit: 1704172 miss: 5118190 ratio: 0.249792 BenchmarkLRU_Freq_WithExpire-8 7098261 168.1 ns/op --- BENCH: BenchmarkLRU_Freq_WithExpire-8 cache_test.go:128: hit: 1 miss: 0 ratio: 1.000000 cache_test.go:128: hit: 100 miss: 0 ratio: 1.000000 cache_test.go:128: hit: 9842 miss: 158 ratio: 0.984200 cache_test.go:128: hit: 90167 miss: 909833 ratio: 0.090167 cache_test.go:128: hit: 90421 miss: 7007840 ratio: 0.012738 PASS ok github.com/go-pkgz/expirable-cache/v3 24.315s Details
For detailed benchmarks and methodology, see the benchmarks directory.
