FEAT: Update Status Of Video Row On Media Convert Update

cms/internal/consumers/mediaconvert.go

@@ -0,0 +1,135 @@
+// Package consumers holds the queue-driven half of the service: work that
+// arrives on a queue rather than as an HTTP request. It is to SQS what
+// internal/handlers is to HTTP — it decodes a message, calls repositories, and
+// decides what the message's fate is — so the same seams apply, and no SQL
+// lives here.
+package consumers
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"log"
+	"thamanyah/cms/v2/internal/db/repositories"
+	"thamanyah/cms/v2/internal/models"
+	"thamanyah/cms/v2/internal/services"
+	"time"
+)
+
+// receiveBackoff is how long the loop waits after a failed receive, so a queue
+// that is unreachable produces a slow trickle of log lines rather than a spin.
+const receiveBackoff = 5 * time.Second
+
+// jobStateChange is the EventBridge event MediaConvert emits on every job
+// state transition. The subscription delivers it raw, so this is the whole
+// message body — there is no SNS envelope to unwrap. Only the members this
+// service acts on are modelled; AWS sends a good deal more.
+type jobStateChange struct {
+	Source     string `json:"source"`
+	DetailType string `json:"detail-type"`
+	Detail     struct {
+		JobID  string `json:"jobId"`
+		Status string `json:"status"`
+	} `json:"detail"`
+}
+
+// statusForJobState maps MediaConvert's job states onto the catalogue's. The
+// second return is false for states that are not an outcome — the job is still
+// running, and the record should stay where it is rather than being rewritten
+// with what it already says.
+func statusForJobState(state string) (models.VideoStatus, bool) {
+	switch state {
+	case "COMPLETE":
+		return models.VideoStatusReady, true
+	case "ERROR", "CANCELED":
+		return models.VideoStatusFailed, true
+	default:
+		return "", false
+	}
+}
+
+// RunMediaConvertEvents consumes job state changes until ctx is cancelled. It
+// is meant to be run in its own goroutine for the lifetime of the process.
+func RunMediaConvertEvents(ctx context.Context) {
+	log.Println("mediaconvert job events consumer started")
+
+	for {
+		if ctx.Err() != nil {
+			log.Println("mediaconvert job events consumer stopped")
+			return
+		}
+
+		messages, err := services.SQSClient.ReceiveMessages(ctx)
+		if err != nil {
+			if ctx.Err() != nil {
+				log.Println("mediaconvert job events consumer stopped")
+				return
+			}
+			log.Printf("Something Went Wrong Receiving Job Events: %s", err)
+			time.Sleep(receiveBackoff)
+			continue
+		}
+
+		for _, message := range messages {
+			if !handleJobEvent(ctx, message.Body) {
+				// Left on the queue on purpose: it becomes visible again when
+				// the visibility timeout expires, and reaches the dead-letter
+				// queue if it keeps failing.
+				continue
+			}
+
+			if err := services.SQSClient.DeleteMessage(ctx, message.ReceiptHandle); err != nil {
+				log.Printf("Something Went Wrong Acknowledging A Job Event: %s", err)
+			}
+		}
+	}
+}
+
+// handleJobEvent processes one message and reports whether it is finished
+// with — that is, whether it should be deleted from the queue. It returns
+// false only when a retry could plausibly succeed. Anything a retry cannot fix
+// (a body that will never parse, an event for a job nobody has) is finished
+// with, however little it accomplished: leaving it on the queue would only
+// stall the events behind it.
+func handleJobEvent(ctx context.Context, body string) bool {
+	var event jobStateChange
+	if err := json.Unmarshal([]byte(body), &event); err != nil {
+		log.Printf("Discarding A Job Event That Is Not JSON: %s", err)
+		return true
+	}
+
+	if event.Source != "aws.mediaconvert" || event.DetailType != "MediaConvert Job State Change" {
+		log.Printf("Discarding An Event That Is Not A MediaConvert Job State Change: source=%q detail-type=%q",
+			event.Source, event.DetailType)
+		return true
+	}
+
+	if event.Detail.JobID == "" {
+		log.Printf("Discarding A Job Event That Names No Job")
+		return true
+	}
+
+	status, isOutcome := statusForJobState(event.Detail.Status)
+	if !isOutcome {
+		// SUBMITTED, PROGRESSING and STATUS_UPDATE say the job is still
+		// running, which is what "processing" already records.
+		return true
+	}
+
+	err := repositories.VideoRepo.UpdateVideoStatusByJobID(ctx, event.Detail.JobID, status)
+	if errors.Is(err, repositories.ErrVideoNotFound) {
+		// Not a fault: the topic carries every job in the account, including
+		// ones this service never submitted.
+		log.Printf("Ignoring A Job Event For A Job No Video Has: job=%q status=%q",
+			event.Detail.JobID, event.Detail.Status)
+		return true
+	}
+	if err != nil {
+		log.Printf("Something Went Wrong Recording A Job Outcome: job=%q status=%q: %s",
+			event.Detail.JobID, event.Detail.Status, err)
+		return false
+	}
+
+	log.Printf("Recorded A Job Outcome: job=%q status=%q", event.Detail.JobID, status)
+	return true
+}