Awesome

oncecache: on-demand in-memory object cache

oncecache is a strongly-typed, concurrency-safe, context-aware, dependency-free, in-memory, on-demand Golang object cache, focused on write-once, read-often ergonomics.

The package also provides an event mechanism useful for logging, metrics, or propagating cache entries between overlapping composite caches.

oncecache is targeted at write-once, read-often situations, where a value corresponding to a key is expensive to compute or fetch, the value is likely to be read multiple times, and is not expected to change. The cache is not intended for general purpose caching where values are frequently updated.

Install

Add to your go.mod via go get:

go get github.com/neilotoole/oncecache

Usage

The basic theory of operation is that a oncecache.Cache is created with a function that returns the value corresponding to a key. When a key is requested from the cache, the cache checks if the value is already present. If not, the cache calls the provided function to compute the value, stores the value, and returns it. Subsequent requests for the same key return the cached value.

Here's a trivial example that caches computed fibonacci numbers:

func ExampleFibonacci() {
	// Ignore error handling for brevity.
	ctx := context.Background()
	c := oncecache.New[int, int](calcFibonacci)

	key := 6
	val, _ := c.Get(ctx, key) // Cache MISS - calcFibonacci(6) is invoked
	fmt.Println(key, val)
	val, _ = c.Get(ctx, key) // Cache HIT
	fmt.Println(key, val)

	key = 9
	val, _ = c.Get(ctx, key) // Cache MISS - calcFibonacci(9) is invoked
	fmt.Println(key, val)

	// Output:
	// 6 8
	// 6 8
	// 9 34
}

func calcFibonacci(ctx context.Context, n int) (val int, err error) {
	a, b, temp := 0, 1, 0 //nolint:wastedassign
	for i := 0; i < n && ctx.Err() == nil; i++ {
		temp = a
		a = b
		b = temp + a
	}

	if ctx.Err() != nil {
		return 0, ctx.Err()
	}

	return a, nil
}

oncecache.Cache provides typical operations to interact with the cache, such as Delete, Has, Keys, etc.

func ExampleOperations() {
	// Ignore error handling for brevity.
	ctx := context.Background()
	c := oncecache.New[int, int](calcFibonacci)

	for key := 4; key < 7; key++ {
		val, _ := c.Get(ctx, key) // Prime the cache for keys 4, 5, 6
		fmt.Println(key, val)
	}

	keys := c.Keys() // Keys returns indeterminate order
	slices.Sort(keys)
	fmt.Println("Keys in cache:", keys)
	fmt.Println("Num entries:", c.Len())
	fmt.Println("Has key 2?", c.Has(2))

	c.Delete(ctx, 5)
	keys = c.Keys()
	slices.Sort(keys)
	fmt.Println("Keys in cache after Delete(5):", keys)

	// MaybeSet sets the value if the key is not already in the cache.
	didSet := c.MaybeSet(ctx, 4, 3, nil) // No-op: 4 already in cache
	fmt.Println("Did set 4?", didSet)
	didSet = c.MaybeSet(ctx, 7, 13, nil) // Cache write: 7 not in cache
	fmt.Println("Did set 7?", didSet)

	c.Clear(ctx) // Clear empties c, but it's still usable
	fmt.Println("Keys after cache clear:", c.Keys())

	// Close clears c and releases resources. Afterwards, c is unusable,
	// and operations on it may return an error.
	_ = c.Close()

	// Output:
	// 4 3
	// 5 5
	// 6 8
	// Keys in cache: [4 5 6]
	// Num entries: 3
	// Has key 2? false
	// Keys in cache after Delete(5): [4 6]
	// Did set 4? false
	// Did set 7? true
	// Keys after cache clear: []
}

Callbacks

When constructing a cache, you can provide callback functions that are invoked when cache events occur. Callbacks are useful for logging, metrics, or propagating cache entries between overlapping composite caches.

Here's an example that logs cache events:

func main() {
	log := slog.Default()
	c := oncecache.New[int, int](
		calcFibonacci,
		oncecache.Name("fibs"), // Name the cache for logging
		oncecache.Log(log, slog.LevelInfo, oncecache.OpFill, oncecache.OpEvict),
		oncecache.Log(log, slog.LevelDebug, oncecache.OpHit, oncecache.OpMiss),
	)

	_, _ = c.Get(ctx, 10) // Cache miss, and fill
	_, _ = c.Get(ctx, 10) // Cache hit
}

This would produce log output similar to:

level=DEBUG msg="Cache event" ev.cache=fibs ev.op=miss ev.k=10
level=INFO msg="Cache event" ev.cache=fibs ev.op=fill ev.k=10 ev.v=55
level=DEBUG msg="Cache event" ev.cache=fibs ev.op=hit ev.k=10 ev.v=55

Note that oncecache.Log is a pre-canned functional opt that writes events to a slog.Logger. For custom callbacks, you can use one of the (synchronous) OnHit, OnMiss, OnFill or OnEvict handlers, or the more generic OnEvent handler, which receives cache events on a channel.

See TestCallbacks or TestOnEventChan in oncecache_test.go for more details, or the take a look at the hrsystem example.

Cache composition & propagation

Consider a trivial HR system:

---
title: HR System
---
erDiagram
    Org ||--|{ Department : contains
    Org {
        string name
    }
    Department ||--|{  Employee : contains
    Department {
				string name
		}
		Employee {
        int id
				string name
				string role
		}

In Go, we might model this system as:

type Org struct {
	Name        string        `json:"name"`
	Departments []*Department `json:"departments"`
}

type Department struct {
	Name  string      `json:"name"`
	Staff []*Employee `json:"staff"`
}

type Employee struct {
	Name string `json:"name"`
	Role string `json:"role"`
	ID   int    `json:"id"`
}

type HRSystem interface {
	GetOrg(ctx context.Context, org string) (*Org, error)
	GetDepartment(ctx context.Context, dept string) (*Department, error)
	GetEmployee(ctx context.Context, ID int) (*Employee, error)
}

Now, consider when HRSystem.GetOrg is called: it returns the entire constructed tree containing all Departments, which in turn contain all Employees. We could use a oncecache.Cache[string, *Org] to cache the Org objects.

But, later, we might want to retrieve a single Employee via HRSystem.GetEmployee(ctx, 1234). Typically, the HRSystem impl would fetch that Employee from the database, but note that the Employee is already present in the Org cache, as a child of a Department object.

We can use oncecache to propagate cache entries across composite caches.

// NewHRCache wraps db with a caching layer.
func NewHRCache(log *slog.Logger, db HRSystem) *HRCache {
	c := &HRCache{
		log:   log,
		db:    db,
	}

	c.orgs = oncecache.New[string, *Org](
		db.GetOrg,
		oncecache.OnFill(c.onFillOrg),
	)

	c.depts = oncecache.New[string, *Department](
		db.GetDepartment,
		oncecache.OnFill(c.onFillDept),
	)

	c.employees = oncecache.New[int, *Employee](
		db.GetEmployee,
	)

	return c
}

In the code above, the NewHRCache constructor adds OnFill event handlers to the Org and Department caches. Thus, a Get call to the Org cache will trigger onFillOrg:

// onFillOrg is invoked by HRCache.orgs when that cache fills an [Org] value from
// the DB. This handler propagates values from the returned [Org] to the
// HRCache.depts cache.
func (c *HRCache) onFillOrg(ctx context.Context, _ string, org *Org, err error) {
	if err != nil {
		return
	}

	for _, dept := range org.Departments {
		// Filling a dept entry should in turn propagate to the employees cache.
		_ = c.depts.MaybeSet(ctx, dept.Name, dept, nil)
	}
}

When onFillOrg is invoked, it iterates over the Departments in the Org and calls MaybeSet on the Department cache. This in turn triggers the onFillDept, which invokes MaybeSet on the Employee cache:

// onFillDept is invoked by HRCache.depts when that cache fills a [Department]
// value from the DB. This handler propagates [Employee] values from the
// returned [Department] to the HRCache.employees cache.
func (c *HRCache) onFillDept(ctx context.Context, _ string, dept *Department, err error) {
	if err != nil {
		return
	}

	for _, emp := range dept.Staff {
		_ = c.employees.MaybeSet(ctx, emp.ID, emp, nil)
	}
}

TBD

oncecache currently lacks a TTL or reaper mechanism.

oncecache was developed for use by sq, which uses oncecache to cache database metadata, e.g. the entire metadata of a database schema, or individual tables withing that schema. See it in use here.