21 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
- Usage
- Persistence
- Eviction
- Expiration
- Server
- Running the server with Docker
- Performance
- FAQ
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). |
| WithForceNilInterfaceOnNilPointer | Configures whether values with a nil pointer passed to write functions should be forcefully set to nil. Defaults to true. |
| 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. |
| GetByKeys | 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. |
| GetAll | Gets all cache entries. |
| 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. |
For further documentation, please refer to Go Reference
Examples
Creating or updating an entry
cache.Set("key", "value")
cache.Set("key", 1)
cache.Set("key", struct{ Text string }{Test: "value"})
cache.SetWithTTL("key", []byte("value"), 24*time.Hour)
Getting an entry
value, exists := cache.Get("key")
You can also get multiple entries by using cache.GetByKeys([]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"
"time"
"github.com/TwinProduction/gocache"
)
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.GetByKeys([]string{"k1", "k2", "k3"}) {
fmt.Printf("[GetByKeys] 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
[GetByKeys] key=k2; value=v2
[GetByKeys] key=k3; value=v3
[GetByKeys] 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 in alpha.
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 100 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.
Expiration
There are two ways that the deletion of expired keys can take place:
- Active
- Passive
Active deletion of expired keys happens when an attempt is made to access the value of a cache entry that expired.
Get, GetByKeys and GetAll are the only functions that can trigger active deletion of expired keys.
Passive deletion of expired keys runs in the background and is managed by the janitor. If you do not start the janitor, there will be no passive deletion of expired keys.
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 limit its external dependencies to the strict
minimum (e.g. boltdb for persistence), however, rather than re-inventing the wheel, the server implementation uses
redcon, which is a very good 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).WithPort(6379)
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
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 |
// Normal map
BenchmarkMap_Get
BenchmarkMap_Get-8 46087372 26.7 ns/op
BenchmarkMap_Set
BenchmarkMap_Set/small_value-8 3841911 389 ns/op
BenchmarkMap_Set/medium_value-8 3887074 391 ns/op
BenchmarkMap_Set/large_value-8 3921956 393 ns/op
// Gocache
BenchmarkCache_Get
BenchmarkCache_Get/FirstInFirstOut-8 27273036 46.4 ns/op
BenchmarkCache_Get/LeastRecentlyUsed-8 26648248 46.3 ns/op
BenchmarkCache_Set
BenchmarkCache_Set/FirstInFirstOut_small_value-8 2919584 405 ns/op
BenchmarkCache_Set/FirstInFirstOut_medium_value-8 2990841 391 ns/op
BenchmarkCache_Set/FirstInFirstOut_large_value-8 2970513 391 ns/op
BenchmarkCache_Set/LeastRecentlyUsed_small_value-8 2962939 402 ns/op
BenchmarkCache_Set/LeastRecentlyUsed_medium_value-8 2962963 390 ns/op
BenchmarkCache_Set/LeastRecentlyUsed_large_value-8 2962928 394 ns/op
BenchmarkCache_SetUsingMaxMemoryUsage
BenchmarkCache_SetUsingMaxMemoryUsage/small_value-8 2683356 447 ns/op
BenchmarkCache_SetUsingMaxMemoryUsage/medium_value-8 2637578 441 ns/op
BenchmarkCache_SetUsingMaxMemoryUsage/large_value-8 2672434 443 ns/op
BenchmarkCache_SetWithMaxSize
BenchmarkCache_SetWithMaxSize/100_small_value-8 4782966 252 ns/op
BenchmarkCache_SetWithMaxSize/10000_small_value-8 4067967 296 ns/op
BenchmarkCache_SetWithMaxSize/100000_small_value-8 3762055 328 ns/op
BenchmarkCache_SetWithMaxSize/100_medium_value-8 4760479 252 ns/op
BenchmarkCache_SetWithMaxSize/10000_medium_value-8 4081050 295 ns/op
BenchmarkCache_SetWithMaxSize/100000_medium_value-8 3785050 330 ns/op
BenchmarkCache_SetWithMaxSize/100_large_value-8 4732909 254 ns/op
BenchmarkCache_SetWithMaxSize/10000_large_value-8 4079533 297 ns/op
BenchmarkCache_SetWithMaxSize/100000_large_value-8 3712820 331 ns/op
BenchmarkCache_SetWithMaxSizeAndLRU
BenchmarkCache_SetWithMaxSizeAndLRU/100_small_value-8 4761732 254 ns/op
BenchmarkCache_SetWithMaxSizeAndLRU/10000_small_value-8 4084474 296 ns/op
BenchmarkCache_SetWithMaxSizeAndLRU/100000_small_value-8 3761402 329 ns/op
BenchmarkCache_SetWithMaxSizeAndLRU/100_medium_value-8 4783075 254 ns/op
BenchmarkCache_SetWithMaxSizeAndLRU/10000_medium_value-8 4103980 296 ns/op
BenchmarkCache_SetWithMaxSizeAndLRU/100000_medium_value-8 3646023 331 ns/op
BenchmarkCache_SetWithMaxSizeAndLRU/100_large_value-8 4779025 254 ns/op
BenchmarkCache_SetWithMaxSizeAndLRU/10000_large_value-8 4096192 296 ns/op
BenchmarkCache_SetWithMaxSizeAndLRU/100000_large_value-8 3726823 331 ns/op
BenchmarkCache_GetSetMultipleConcurrent
BenchmarkCache_GetSetMultipleConcurrent-8 707142 1698 ns/op
BenchmarkCache_GetSetConcurrentWithFrequentEviction
BenchmarkCache_GetSetConcurrentWithFrequentEviction/FirstInFirstOut-8 3616256 334 ns/op
BenchmarkCache_GetSetConcurrentWithFrequentEviction/LeastRecentlyUsed-8 3636367 331 ns/op
BenchmarkCache_GetConcurrentWithLRU
BenchmarkCache_GetConcurrentWithLRU/FirstInFirstOut-8 4405557 268 ns/op
BenchmarkCache_GetConcurrentWithLRU/LeastRecentlyUsed-8 4445475 269 ns/op
BenchmarkCache_WithForceNilInterfaceOnNilPointer
BenchmarkCache_WithForceNilInterfaceOnNilPointer/true_with_nil_struct_pointer-8 6184591 191 ns/op
BenchmarkCache_WithForceNilInterfaceOnNilPointer/true-8 6090482 191 ns/op
BenchmarkCache_WithForceNilInterfaceOnNilPointer/false_with_nil_struct_pointer-8 6184629 187 ns/op
BenchmarkCache_WithForceNilInterfaceOnNilPointer/false-8 6281781 186 ns/op
(Trimmed "BenchmarkCache_" for readability)
WithForceNilInterfaceOnNilPointerWithConcurrency
WithForceNilInterfaceOnNilPointerWithConcurrency/true_with_nil_struct_pointer-8 4379564 268 ns/op
WithForceNilInterfaceOnNilPointerWithConcurrency/true-8 4379558 265 ns/op
WithForceNilInterfaceOnNilPointerWithConcurrency/false_with_nil_struct_pointer-8 4444456 261 ns/op
WithForceNilInterfaceOnNilPointerWithConcurrency/false-8 4493896 262 ns/op
FAQ
Why does the memory usage not go down?
Note
: As of Go 1.16, this will no longer apply. See golang/go#42330
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.