18 KiB
gocache
gocache is an easy-to-use, high-performance, lightweight and thread-safe (goroutine-safe) in-memory key-value cache with support for LRU and FIFO eviction policies as well as expiration, bulk operations and even persistence to file.
Table of Contents
Features
gocache supports the following cache eviction policies:
- First in first out (FIFO)
- Least recently used (LRU)
It also supports cache entry TTL, which is both active and passive. Active expiration means that if you attempt to retrieve a cache key that has already expired, it will delete it on the spot and the behavior will be as if the cache key didn't exist. As for passive expiration, there's a background task that will take care of deleting expired keys.
It also includes what you'd expect from a cache, like bulk operations, persistence and patterns.
While meant to be used as a library, there's a Redis-compatible cache server included. See the Server section. It may also serve as a good reference to use in order to implement gocache in your own applications.
Usage
go get -u github.com/TwinProduction/gocache
Initializing the cache
cache := gocache.NewCache().WithMaxSize(1000).WithEvictionPolicy(gocache.LeastRecentlyUsed)
If you're planning on using expiration (SetWithTTL or Expire) and you want expired entries to be automatically deleted
in the background, make sure to start the janitor when you instantiate the cache:
cache.StartJanitor()
Functions
| Function | Description |
|---|---|
| WithMaxSize | Sets the max size of the cache. gocache.NoMaxSize means there is no limit. If not set, the default max size is gocache.DefaultMaxSize. |
| WithMaxMemoryUsage | Sets the max memory usage of the cache. gocache.NoMaxMemoryUsage means there is no limit. The default behavior is to not evict based on memory usage. |
| WithEvictionPolicy | Sets the eviction algorithm to be used when the cache reaches the max size. If not set, the default eviction policy is gocache.FirstInFirstOut (FIFO). |
| StartJanitor | Starts the janitor, which is in charge of deleting expired cache entries in the background. |
| StopJanitor | Stops the janitor. |
| Set | Same as SetWithTTL, but with no expiration (gocache.NoExpiration) |
| SetAll | Same as Set, but in bulk |
| SetWithTTL | Creates or updates a cache entry with the given key, value and expiration time. If the max size after the aforementioned operation is above the configured max size, the tail will be evicted. Depending on the eviction policy, the tail is defined as the oldest |
| Get | Gets a cache entry by its key. |
| GetAll | Gets a map of entries by their keys. The resulting map will contain all keys, even if some of the keys in the slice passed as parameter were not present in the cache. |
| GetKeysByPattern | Retrieves a slice of keys that matches a given pattern. |
| Delete | Removes a key from the cache. |
| DeleteAll | Removes multiple keys from the cache. |
| Count | Gets the size of the cache. This includes cache keys which may have already expired, but have not been removed yet. |
| Clear | Wipes the cache. |
| TTL | Gets the time until a cache key expires. |
| Expire | Sets the expiration time of an existing cache key. |
| SaveToFile | Stores the content of the cache to a file so that it can be read using ReadFromFile. See persistence. |
| ReadFromFile | Populates the cache using a file created using SaveToFile. See persistence. |
Examples
Creating or updating an entry
cache.Set("key", "value")
cache.Set("key", 1)
cache.Set("key", struct{ Text string }{Test: "value"})
Getting an entry
value, ok := cache.Get("key")
You can also get multiple entries by using cache.GetAll([]string{"key1", "key2"})
Deleting an entry
cache.Delete("key")
You can also delete multiple entries by using cache.DeleteAll([]string{"key1", "key2"})
Complex example
package main
import (
"fmt"
"github.com/TwinProduction/gocache"
"time"
)
func main() {
cache := gocache.NewCache().WithEvictionPolicy(gocache.LeastRecentlyUsed).WithMaxSize(10000)
cache.StartJanitor() // Passively manages expired entries
cache.Set("key", "value")
cache.SetWithTTL("key-with-ttl", "value", 60*time.Minute)
cache.SetAll(map[string]interface{}{"k1": "v1", "k2": "v2", "k3": "v3"})
value, exists := cache.Get("key")
fmt.Printf("[Get] key=key; value=%s; exists=%v\n", value, exists)
for key, value := range cache.GetAll([]string{"k1", "k2", "k3"}) {
fmt.Printf("[GetAll] key=%s; value=%s\n", key, value)
}
for _, key := range cache.GetKeysByPattern("key*", 0) {
fmt.Printf("[GetKeysByPattern] key=%s\n", key)
}
fmt.Println("Cache size before persisting cache to file:", cache.Count())
err := cache.SaveToFile("cache.bak")
if err != nil {
panic(fmt.Sprintf("failed to persist cache to file: %s", err.Error()))
}
cache.Expire("key", time.Hour)
time.Sleep(500*time.Millisecond)
timeUntilExpiration, _ := cache.TTL("key")
fmt.Println("Number of minutes before 'key' expires:", int(timeUntilExpiration.Seconds()))
cache.Delete("key")
cache.DeleteAll([]string{"k1", "k2", "k3"})
fmt.Println("Cache size before restoring cache from file:", cache.Count())
_, err = cache.ReadFromFile("cache.bak")
if err != nil {
panic(fmt.Sprintf("failed to restore cache from file: %s", err.Error()))
}
fmt.Println("Cache size after restoring cache from file:", cache.Count())
cache.Clear()
fmt.Println("Cache size after clearing the cache:", cache.Count())
}
Output
[Get] key=key; value=value; exists=true
[GetAll] key=k2; value=v2
[GetAll] key=k3; value=v3
[GetAll] key=k1; value=v1
[GetKeysByPattern] key=key
[GetKeysByPattern] key=key-with-ttl
Cache size before persisting cache to file: 5
Number of minutes before 'key' expires: 3599
Cache size before restoring cache from file: 1
Cache size after restoring cache from file: 5
Cache size after clearing the cache: 0
Persistence
While gocache is an in-memory cache, you can still save the content of the cache in a file and vice versa.
To save the content of the cache to a file:
err := cache.SaveToFile(TestCacheFile)
To retrieve the content of the cache from a file:
numberOfEntriesEvicted, err := newCache.ReadFromFile(TestCacheFile)
The numberOfEntriesEvicted will be non-zero only if the number of entries
in the file is higher than the cache's configured MaxSize.
Limitations
While you can cache structs in memory out of the box, persisting structs to a file requires you to
register the custom interfaces that your application uses with the gob package.
type YourCustomStruct struct {
A string
B int
}
// ...
cache.Set("key", YourCustomStruct{A: "test", B: 123})
To persist your custom struct properly:
gob.Register(YourCustomStruct{})
cache.SaveToFile("gocache.bak")
The same applies for restoring the cache from a file:
cache := NewCache()
gob.Register(YourCustomStruct{})
cache.ReadFromFile(TestCacheFile)
value, _ := cache.Get("key")
fmt.Println(value.(YourCustomStruct))
You only need to persist the struct once, so adding the following function in a file would suffice:
func init() {
gob.Register(YourCustomStruct{})
}
Failure to register your custom structs will prevent gocache from persisting and/or parsing the value of each keys that use said custom structs.
That being said, assuming that you're using gocache as a cache, this shouldn't create any bugs on your end, because
every key that cannot be parsed are not populated into the cache by ReadFromFile.
In other words, if you're falling back to a database or something similar when the cache doesn't have the key requested, you'll be fine.
Eviction
MaxSize
Eviction by MaxSize is the default behavior, and is also the most efficient.
The code below will create a cache that has a maximum size of 1000:
cache := gocache.NewCache().WithMaxSize(1000)
This means that whenever an operation causes the total size of the cache to go above 1000, the tail will be evicted.
MaxMemoryUsage
Eviction by MaxMemoryUsage is disabled by default, and is still a work in progress.
The code below will create a cache that has a maximum memory usage of 50MB:
cache := gocache.NewCache().WithMaxSize(0).WithMaxMemoryUsage(50*gocache.Megabyte)
This means that whenever an operation causes the total memory usage of the cache to go above 50MB, one or more tails will be evicted.
Unlike evictions caused by reaching the MaxSize, evictions triggered by MaxMemoryUsage may lead to multiple entries being evicted in a row. The reason for this is that if, for instance, you had 500 entries of 0.1MB each and you suddenly added a single entry of 10MB, 100 entries would need to be evicted to make enough space for that new big entry.
It's very important to keep in mind that eviction by MaxMemoryUsage is approximate.
The only memory taken into consideration is the size of the cache, not the size of the entire application. If you pass along 100MB worth of data in a matter of seconds, even though the cache's memory usage will remain under 50MB (or whatever you configure the MaxMemoryUsage to), the memory footprint generated by that 100MB will still exist until the next GC cycle.
As previously mentioned, this is a work in progress, and here's a list of the things you should keep in mind:
- The memory usage of structs are a gross estimation and may not reflect the actual memory usage.
- Native types (string, int, bool, []byte, etc.) are the most accurate for calculating the memory usage.
- Adding an entry bigger than the configured MaxMemoryUsage will work, but it will evict all other entries.
Server
For the sake of convenience, a ready-to-go cache server is available
through the gocacheserver package.
The reason why the server is in a different package is because gocache does not use
any external dependencies, but rather than re-inventing the wheel, the server
implementation uses redcon, which is a Redis server framework for Go.
That way, those who desire to use gocache without the server will not add any extra dependencies
as long as they don't import the gocacheserver package.
package main
import (
"github.com/TwinProduction/gocache"
"github.com/TwinProduction/gocache/gocacheserver"
)
func main() {
cache := gocache.NewCache().WithEvictionPolicy(gocache.LeastRecentlyUsed).WithMaxSize(100000)
server := gocacheserver.NewServer(cache)
server.Start()
}
Any Redis client should be able to interact with the server, though only the following instructions are supported:
- GET
- SET
- DEL
- PING
- QUIT
- INFO
- EXPIRE
- SETEX
- TTL
- FLUSHDB
- EXISTS
- ECHO
- MGET
- MSET
- SCAN (kind of - cursor is not currently supported)
- KEYS
Running the server with Docker
To build it locally, refer to the Makefile's docker-build and docker-run steps.
Note that the server version of gocache is still under development.
docker run --name gocache-server -p 6379:6379 twinproduction/gocache-server:v0.1.0
Performance
Summary
- Set: Both map and gocache have the same performance.
- Get: Map is faster than gocache.
This is because gocache keeps track of the head and the tail for eviction and expiration/TTL.
Ultimately, the difference is negligible.
We could add a way to disable eviction or disable expiration altogether just to match the map's performance, but if you're looking into using a library like gocache, odds are, you want more than just a map.
Results
| key | value |
|---|---|
| goos | windows |
| goarch | amd64 |
| cpu | i7-9700K |
| mem | 32G DDR4 |
BenchmarkMap_Get-8 47943618 26.6 ns/op
BenchmarkMap_SetSmallValue-8 3800810 394 ns/op
BenchmarkMap_SetMediumValue-8 3904794 400 ns/op
BenchmarkMap_SetLargeValue-8 3934033 383 ns/op
BenchmarkCache_Get-8 27254640 45.0 ns/op
BenchmarkCache_SetSmallValue-8 2991620 401 ns/op
BenchmarkCache_SetMediumValue-8 3051128 381 ns/op
BenchmarkCache_SetLargeValue-8 2995904 382 ns/op
BenchmarkCache_SetSmallValueWhenUsingMaxMemoryUsage-8 2752288 428 ns/op
BenchmarkCache_SetMediumValueWhenUsingMaxMemoryUsage-8 2744899 436 ns/op
BenchmarkCache_SetLargeValueWhenUsingMaxMemoryUsage-8 2756816 430 ns/op
BenchmarkCache_SetSmallValueWithMaxSize10-8 5308886 226 ns/op
BenchmarkCache_SetMediumValueWithMaxSize10-8 5304098 226 ns/op
BenchmarkCache_SetLargeValueWithMaxSize10-8 5277986 227 ns/op
BenchmarkCache_SetSmallValueWithMaxSize1000-8 5130580 236 ns/op
BenchmarkCache_SetMediumValueWithMaxSize1000-8 5102404 237 ns/op
BenchmarkCache_SetLargeValueWithMaxSize1000-8 5084695 237 ns/op
BenchmarkCache_SetSmallValueWithMaxSize100000-8 3858066 315 ns/op
BenchmarkCache_SetMediumValueWithMaxSize100000-8 3909277 315 ns/op
BenchmarkCache_SetLargeValueWithMaxSize100000-8 3870913 315 ns/op
BenchmarkCache_SetSmallValueWithMaxSize100000AndLRU-8 3856012 316 ns/op
BenchmarkCache_SetMediumValueWithMaxSize100000AndLRU-8 3809518 316 ns/op
BenchmarkCache_SetLargeValueWithMaxSize100000AndLRU-8 3834754 318 ns/op
BenchmarkCache_GetAndSetConcurrently-8 1779258 672 ns/op
BenchmarkCache_GetAndSetConcurrentlyWithRandomKeysAndLRU-8 2569590 487 ns/op
BenchmarkCache_GetAndSetConcurrentlyWithRandomKeysAndFIFO-8 2608369 474 ns/op
BenchmarkCache_GetAndSetConcurrentlyWithRandomKeysAndNoEvictionAndLRU-8 2185795 582 ns/op
BenchmarkCache_GetAndSetConcurrentlyWithRandomKeysAndNoEvictionAndFIFO-8 2238811 568 ns/op
BenchmarkCache_GetAndSetConcurrentlyWithFrequentEvictionsAndLRU-8 3726714 320 ns/op
BenchmarkCache_GetAndSetConcurrentlyWithFrequentEvictionsAndFIFO-8 3682808 325 ns/op
BenchmarkCache_GetConcurrentlyWithLRU-8 1536589 739 ns/op
BenchmarkCache_GetConcurrentlyWithFIFO-8 1558513 737 ns/op
BenchmarkCache_GetKeysThatDoNotExistConcurrently-8 10173138 119 ns/op
FAQ
Why does the memory usage not go down?
By default, Go uses MADV_FREE if the kernel supports it to release memory, which is significantly more efficient
than using MADV_DONTNEED. Unfortunately, this means that RSS doesn't go down unless the OS actually needs the
memory.
Technically, the memory is available to the kernel, even if it shows a high memory usage, but the OS will only use that memory if it needs to. In the case that the OS does need the freed memory, the RSS will go down and you'll notice the memory usage lowering.
You can reproduce this by following the steps below:
- Start gocacheserver
- Note the memory usage
- Create 500k keys
- Note the memory usage
- Flush the cache
- Note that the memory usage has not decreased, despite the cache being empty.
Substituting gocache for a normal map will yield the same result.
If the released memory still appearing as used is a problem for you,
you can set the environment variable GODEBUG to madvdontneed=1.