Accumulate incoming messages in a buffered channel

Instead of using a deque, store incoming messages in a native
buffered channel, if buffering is enabled.

In addition, modify the batching algorithm so the enforced delay
between consecutive `addMessages` invocations is applied after
all pending messages are processed. This acts as a "cooldown", rather
than a "warmup". This avoids the need for more complex timing logic to
dispatch batches, removes latency in adding messages when received
infrequently, and natively blocking the goroutine until messages are
received.

Because the message processing loop always performs a blocking read
first, it is appropriate for low-throughput environments just as much as
high-throughput ones.

The default value of batchSize has been changed to 10, with a zero
cooldown. This means that when messages are arriving faster than they
can be inserted into sqlite, they will automatically become batched in
groups of up to 10.
This commit is contained in:
Nick Farrell 2022-12-12 18:31:44 +11:00
parent 6f170b1ad7
commit 09e8fb81b5
No known key found for this signature in database
GPG key ID: 740D3A86CF435835
7 changed files with 140 additions and 168 deletions

View file

@ -1,86 +0,0 @@
package util
import (
"sync"
"time"
)
// BatchingQueue is a queue that creates batches of the enqueued elements based on a
// max batch size and a batch timeout.
//
// Example:
//
// q := NewBatchingQueue[int](2, 500 * time.Millisecond)
// go func() {
// for batch := range q.Dequeue() {
// fmt.Println(batch)
// }
// }()
// q.Enqueue(1)
// q.Enqueue(2)
// q.Enqueue(3)
// time.Sleep(time.Second)
//
// This example will emit batch [1, 2] immediately (because the batch size is 2), and
// a batch [3] after 500ms.
type BatchingQueue[T any] struct {
batchSize int
timeout time.Duration
in []T
out chan []T
mu sync.Mutex
}
// NewBatchingQueue creates a new BatchingQueue
func NewBatchingQueue[T any](batchSize int, timeout time.Duration) *BatchingQueue[T] {
q := &BatchingQueue[T]{
batchSize: batchSize,
timeout: timeout,
in: make([]T, 0),
out: make(chan []T),
}
go q.timeoutTicker()
return q
}
// Enqueue enqueues an element to the queue. If the configured batch size is reached,
// the batch will be emitted immediately.
func (q *BatchingQueue[T]) Enqueue(element T) {
q.mu.Lock()
q.in = append(q.in, element)
var elements []T
if len(q.in) == q.batchSize {
elements = q.dequeueAll()
}
q.mu.Unlock()
if len(elements) > 0 {
q.out <- elements
}
}
// Dequeue returns a channel emitting batches of elements
func (q *BatchingQueue[T]) Dequeue() <-chan []T {
return q.out
}
func (q *BatchingQueue[T]) dequeueAll() []T {
elements := make([]T, len(q.in))
copy(elements, q.in)
q.in = q.in[:0]
return elements
}
func (q *BatchingQueue[T]) timeoutTicker() {
if q.timeout == 0 {
return
}
ticker := time.NewTicker(q.timeout)
for range ticker.C {
q.mu.Lock()
elements := q.dequeueAll()
q.mu.Unlock()
if len(elements) > 0 {
q.out <- elements
}
}
}

View file

@ -1,58 +0,0 @@
package util_test
import (
"github.com/stretchr/testify/require"
"heckel.io/ntfy/util"
"math/rand"
"sync"
"testing"
"time"
)
func TestBatchingQueue_InfTimeout(t *testing.T) {
q := util.NewBatchingQueue[int](25, 1*time.Hour)
batches, total := make([][]int, 0), 0
var mu sync.Mutex
go func() {
for batch := range q.Dequeue() {
mu.Lock()
batches = append(batches, batch)
total += len(batch)
mu.Unlock()
}
}()
for i := 0; i < 101; i++ {
go q.Enqueue(i)
}
time.Sleep(time.Second)
mu.Lock()
require.Equal(t, 100, total) // One is missing, stuck in the last batch!
require.Equal(t, 4, len(batches))
mu.Unlock()
}
func TestBatchingQueue_WithTimeout(t *testing.T) {
q := util.NewBatchingQueue[int](25, 100*time.Millisecond)
batches, total := make([][]int, 0), 0
var mu sync.Mutex
go func() {
for batch := range q.Dequeue() {
mu.Lock()
batches = append(batches, batch)
total += len(batch)
mu.Unlock()
}
}()
for i := 0; i < 101; i++ {
go func(i int) {
time.Sleep(time.Duration(rand.Intn(700)) * time.Millisecond)
q.Enqueue(i)
}(i)
}
time.Sleep(time.Second)
mu.Lock()
require.Equal(t, 101, total)
require.True(t, len(batches) > 4) // 101/25
require.True(t, len(batches) < 21)
mu.Unlock()
}