GoSudaReducing Response Latency in Go HTTP File Uploads: Implementing an Asynchronous Worker Queue Based on Channels
Overview
When first learning Go, developers often build backend servers as a learning exercise. Let us consider an example of creating an implementation that receives a file stream from a RestAPI or similar interface and uploads it to a server. Because the Go net/http server natively handles multiple requests concurrently, concurrent uploads themselves do not present a problem. However, if all operations following stream reception are processed synchronously, the response of the endpoint will be delayed. Let us examine techniques to prevent such situations.
Cause
Receiving a stream generally requires a significant amount of time, and for large files, a single request may take several minutes to process. In such cases, it is important to process operations subsequent to reception as swiftly as possible. This example scenario involves receiving a stream, saving it as a temporary file, and pushing it to a container. If the step of pushing the temporary file to the container is handled by a worker pool, response latency can be reduced.
1package file_upload
2
3import (
4 "fmt"
5 "io"
6 "log"
7 "net/http"
8 "os"
9 "path/filepath"
10 "time"
11)
12
13const uploadTempDir = "/tmp/incus_uploads" // Host temporary directory
14
15// UploadTask holds data for asynchronous file push.
16type UploadTask struct {
17 HostTempFilePath string
18 ContainerName string
19 HostFilename string
20 ContainerDestinationPath string
21}
22
23// UploadHandler processes file uploads. Saves to temp file, then queues for Incus push.
24func UploadHandler(wr http.ResponseWriter, req *http.Request) {
25 if req.Method != http.MethodPost {
26 http.Error(wr, "POST method required.", http.StatusMethodNotAllowed)
27 return
28 }
29 originalFilePath := req.Header.Get("X-File-Path")
30 originalFilename := filepath.Base(req.Header.Get("X-Host-Path"))
31 containerName := req.Header.Get("X-Container-Name")
32 if originalFilePath == "" || containerName == "" {
33 http.Error(wr, "Missing X-File-Path or X-Container-Name header.", http.StatusBadRequest)
34 return
35 }
36
37 cleanContainerDestPath := filepath.Clean(originalFilePath)
38 if !filepath.IsAbs(cleanContainerDestPath) {
39 http.Error(wr, "File path must be absolute.", http.StatusBadRequest)
40 return
41 }
42
43 // Create unique temporary file path on host
44 tempFileName := fmt.Sprintf("%d-%s", time.Now().UnixNano(), filepath.Base(originalFilePath))
45 hostTempFilePath := filepath.Join(uploadTempDir, tempFileName)
46
47 if err := os.MkdirAll(uploadTempDir, 0755); err != nil {
48 log.Printf("ERROR: Failed to create temp upload directory: %v", err)
49 http.Error(wr, "Server error.", http.StatusInternalServerError)
50 return
51 }
52
53 // Create and copy request body to temporary file (synchronous)
54 outFile, err := os.Create(hostTempFilePath)
55 if err != nil {
56 log.Printf("ERROR: Failed to create temporary file: %v", err)
57 http.Error(wr, "Server error.", http.StatusInternalServerError)
58 return
59 }
60 defer outFile.Close()
61
62 bytesWritten, err := io.Copy(outFile, req.Body)
63 if err != nil {
64 outFile.Close()
65 os.Remove(hostTempFilePath)
66 log.Printf("ERROR: Failed to copy request body to temp file: %v", err)
67 http.Error(wr, "File transfer failed.", http.StatusInternalServerError)
68 return
69 }
70 log.Printf("Upload Info: Received %d bytes, saved to %s.", bytesWritten, hostTempFilePath)
71
72 // Enqueue task for asynchronous Incus push
73 task := UploadTask{
74 HostTempFilePath: hostTempFilePath,
75 ContainerName: containerName,
76 HostFilename : originalFilename,
77 ContainerDestinationPath: cleanContainerDestPath,
78 }
79 EnqueueTask(task) //THIS PART
80 log.Printf("Upload Info: Task enqueued for %s to %s.", originalFilePath, containerName)
81
82 // Send immediate 202 Accepted response
83 wr.WriteHeader(http.StatusAccepted)
84 fmt.Fprintf(wr, "File '%s' queued for processing on container '%s'.\n", originalFilename, containerName)
85}
Examining the section marked as THIS PART, you will have noticed that tasks are inserted into a queue.
Now, let us observe how the work queue operates.
1package file_upload
2
3import (
4 "log"
5 "sync"
6)
7
8var taskQueue chan UploadTask
9var once sync.Once
10
11// InitWorkQueue initializes the in-memory task queue.
12func InitWorkQueue() {
13 once.Do(func() {
14 taskQueue = make(chan UploadTask, 100)
15 log.Println("Upload Info: Work queue initialized.")
16 })
17}
18
19// EnqueueTask adds an UploadTask to the queue.
20func EnqueueTask(task UploadTask) {
21 if taskQueue == nil {
22 log.Fatal("ERROR: Task queue not initialized.")
23 }
24 taskQueue <- task
25 log.Printf("Upload Info: Queue: Task enqueued. Size: %d", len(taskQueue))
26}
27
28// DequeueTask retrieves an UploadTask from the queue, blocking if empty.
29func DequeueTask() UploadTask {
30 if taskQueue == nil {
31 log.Fatal("ERROR: Task queue not initialized.")
32 }
33 task <-taskQueue
34 log.Printf("Upload Info: Queue: Task dequeued. Size: %d", len(taskQueue))
35 return task
36}
37
38// GetQueueLength returns current queue size.
39func GetQueueLength() int {
40 if taskQueue == nil {
41 return 0
42 }
43 return len(taskQueue)
44}
The work queue provided in the example is implemented simply. It possesses a straightforward structure that extracts tasks residing in the queue from the channel.
Below is the worker method for pushing to the container after file uploading. To achieve good responsiveness and straightforward implementation, the current method utilizes an infinite loop, though you may add algorithms depending on your specific use case.
1func StartFilePushWorker() {
2 for {
3 task := DequeueTask()
4 log.Printf("Upload Info: Worker processing task for %s from %s.", task.ContainerName, task.HostFilename)
5
6 // Defer cleanup of the temporary file
7 defer func(filePath string) {
8 if err := os.Remove(filePath); err != nil {
9 log.Printf("ERROR: Worker: Failed to remove temp file '%s': %v", filePath, err)
10 } else {
11 log.Printf("Upload Info: Worker: Cleaned up temp file: %s", filePath)
12 }
13 }(task.HostTempFilePath)
14
15 // Process task with retries for transient Incus errors
16 for i := 0; i <= MaxRetries; i++ {
17 err := processUploadTask(task) //separate upload task
18 if err == nil {
19 log.Printf("SUCCESS: Worker: Task completed for %s.", task.ContainerName)
20 break
21 }
22
23 isTransient := true
24 if err.Error() == "incus: container not found" { // Example permanent error
25 isTransient = false
26 }
27
28 if isTransient && i < MaxRetries {
29 log.Printf("WARNING: Worker: Task failed for %s (attempt %d/%d): %v. Retrying.",
30 task.ContainerName, i+1, MaxRetries, err)
31 time.Sleep(RetryDelay)
32 } else {
33 log.Printf("ERROR: Worker: Task permanently failed for %s after %d attempts: %v.",
34 task.ContainerName, i+1, err)
35 break
36 }
37 }
38 }
39}
First, this function continuously loops in an attempt to retrieve tasks from the queue. Subsequently, within the retry limits, it attempts the upload phase from the temporary file to the container rather than from the stream to the temporary file.
Benefits
The advantages of this processing approach are that, provided the stream upload is successful, latency for subsequent operations can be reduced, and resource exhaustion caused by simultaneous container operations can be prevented. As demonstrated in the current code, the number of concurrently executable container operations is limited by the channel capacity. We have thus explored an example where Go's parallel processing can be applied practically. If you wish to view additional examples, please visit the links below.Module containing examples Project utilizing examples Since the project itself contains many auxiliary components, your study of the worker itself only requires a brief look at how the worker initialization function is called in main.go, after which you may move on. Please note that the module also includes other forms of workers.