feat: batch kafka message persistence

This commit is contained in:
richjohnrambo
2026-09-24 22:45:07 +08:00
parent 7f2df1dd4f
commit f898a7171c
18 changed files with 682 additions and 98 deletions
+134 -44
View File
@@ -5,6 +5,7 @@ import (
"encoding/json"
"errors"
"fmt"
"strings"
"time"
"github.com/segmentio/kafka-go"
@@ -24,8 +25,12 @@ type kafkaMessagePersistPayload struct {
var publishKafkaMessagePersist = publishKafkaEnvelope
func startKafkaMessagePersistConsumer() {
startKafkaConsumer("message_persist", handleKafkaMessagePersist)
startKafkaConsumer("group_message_persist", handleKafkaGroupMessagePersist)
startKafkaBatchConsumer("message_persist", func(ctx context.Context, messages []kafka.Message) error {
return handleKafkaMessagePersistBatch(ctx, messages, "message_persist")
})
startKafkaBatchConsumer("group_message_persist", func(ctx context.Context, messages []kafka.Message) error {
return handleKafkaMessagePersistBatch(ctx, messages, "group_message_persist")
})
}
func saveMessageMaybeKafka(ctx context.Context, tenantID types.TenantID, msg *types.Message, attachmentURLs []string, readBySender bool) (*types.MessageSaveResult, error) {
@@ -132,70 +137,155 @@ func enqueueKafkaMessagePersist(ctx context.Context, tenantID types.TenantID, ms
return publishKafkaMessagePersist(ctx, kafkaMessagePersistTopicName(msg.Topic), msg.Topic, envelope)
}
func handleKafkaMessagePersist(ctx context.Context, envelope kafkaEnvelope, _ kafka.Message) error {
return handleKafkaMessagePersistFromTopic(ctx, envelope, "message_persist")
type kafkaMessagePersistItem struct {
envelope kafkaEnvelope
tenantID types.TenantID
payload kafkaMessagePersistPayload
}
func handleKafkaGroupMessagePersist(ctx context.Context, envelope kafkaEnvelope, _ kafka.Message) error {
return handleKafkaMessagePersistFromTopic(ctx, envelope, "group_message_persist")
func handleKafkaMessagePersistBatch(ctx context.Context, messages []kafka.Message, sourceTopicName string) error {
groups := make(map[string][]*kafkaMessagePersistItem)
for _, message := range messages {
item, terminal, err := prepareKafkaMessagePersistItem(ctx, message, sourceTopicName)
if err != nil {
return err
}
if terminal {
continue
}
key := fmt.Sprintf("%d", item.tenantID)
if item.payload.MessageKind == "group" {
key += ":" + item.payload.Message.ShardMonth
}
groups[key] = append(groups[key], item)
}
for _, items := range groups {
batch := make([]store.PreparedMessageBatchItem, 0, len(items))
for _, item := range items {
batch = append(batch, store.PreparedMessageBatchItem{Message: &item.payload.Message, ReadBySender: item.payload.ReadBySender})
}
if err := store.PersistPreparedMessageBatch(items[0].tenantID, batch); err == nil {
for _, item := range items {
if err = finishKafkaMessagePersist(ctx, item); err != nil {
if err = persistKafkaMessageIndividually(ctx, item); err != nil {
return err
}
}
}
continue
} else {
logs.Warn.Printf("kafka message batch persist failed, degrading to singles: tenant=%d kind=%s shard=%s size=%d err=%v",
items[0].tenantID, items[0].payload.MessageKind, items[0].payload.Message.ShardMonth, len(items), err)
}
for _, item := range items {
if err := persistKafkaMessageIndividually(ctx, item); err != nil {
return err
}
}
}
return nil
}
func handleKafkaMessagePersistFromTopic(ctx context.Context, envelope kafkaEnvelope, sourceTopicName string) error {
func prepareKafkaMessagePersistItem(ctx context.Context, message kafka.Message, sourceTopicName string) (*kafkaMessagePersistItem, bool, error) {
var envelope kafkaEnvelope
if err := json.Unmarshal(message.Value, &envelope); err != nil {
logs.Err.Printf("kafka message malformed envelope skipped: partition=%d offset=%d err=%v", message.Partition, message.Offset, err)
return nil, true, nil
}
tenantID := envelope.TenantID
var err error
if tenantID.IsZero() {
tenantID, err = kafkaTenantID(envelope.TenantCode)
if err != nil {
return err
return nil, false, err
}
}
consume, claimed, err := beginKafkaConsumeForTopic(tenantID, envelope, sourceTopicName)
if err != nil || !claimed {
return err
if err != nil {
return nil, false, err
}
var payload kafkaMessagePersistPayload
if err = json.Unmarshal(envelope.Payload, &payload); err != nil {
return markKafkaDeadLetter(ctx, tenantID, envelope, consume, "MALFORMED_PAYLOAD", err.Error())
if !claimed {
return nil, true, nil
}
if payload.Message.TenantID == types.ZeroTenantID {
payload.Message.TenantID = tenantID
item := &kafkaMessagePersistItem{envelope: envelope, tenantID: tenantID}
if err = json.Unmarshal(envelope.Payload, &item.payload); err != nil {
return nil, true, markKafkaDeadLetter(ctx, tenantID, envelope, consume, "MALFORMED_PAYLOAD", err.Error())
}
if payload.Message.ShardMonth == "" {
payload.Message.ShardMonth = payload.ShardMonth
if item.payload.Message.TenantID == types.ZeroTenantID {
item.payload.Message.TenantID = tenantID
}
if payload.Message.TenantID != tenantID {
return markKafkaDeadLetter(ctx, tenantID, envelope, consume, "TENANT_MISMATCH", "message tenant does not match envelope tenant")
if item.payload.Message.ShardMonth == "" {
item.payload.Message.ShardMonth = item.payload.ShardMonth
}
if err = validateKafkaMessageKind(payload); err != nil {
return markKafkaDeadLetter(ctx, tenantID, envelope, consume, "MALFORMED_PAYLOAD", err.Error())
if item.payload.Message.TenantID != tenantID {
return nil, true, markKafkaDeadLetter(ctx, tenantID, envelope, consume, "TENANT_MISMATCH", "message tenant does not match envelope tenant")
}
if _, err = store.PersistPreparedMessage(tenantID, &payload.Message, nil, payload.ReadBySender); err != nil && !errors.Is(err, types.ErrDuplicate) {
failed, markErr := store.HTTPInfrastructure.MarkKafkaConsumeFailed(tenantID, envelope.EventID, kafkaBus.config.ConsumerGroup,
kafkaCommandErrorReason(err), err.Error(), time.Now().UTC().Round(time.Millisecond))
if markErr != nil {
return markErr
if err = validateKafkaMessageKind(item.payload); err != nil {
return nil, true, markKafkaDeadLetter(ctx, tenantID, envelope, consume, "MALFORMED_PAYLOAD", err.Error())
}
return item, false, nil
}
func persistKafkaMessageIndividually(ctx context.Context, item *kafkaMessagePersistItem) error {
operation := func() error {
err := store.PersistPreparedMessageBatch(item.tenantID, []store.PreparedMessageBatchItem{{
Message: &item.payload.Message, ReadBySender: item.payload.ReadBySender,
}})
if errors.Is(err, types.ErrUnsupported) {
_, err = store.PersistPreparedMessage(item.tenantID, &item.payload.Message, nil, item.payload.ReadBySender)
}
if failed != nil && failed.RetryCount >= kafkaBus.config.MaxRetryCount {
return markKafkaDeadLetter(ctx, tenantID, envelope, failed, kafkaCommandErrorReason(err), err.Error())
}
sleepKafkaRetryBackoff(failed.RetryCount)
return err
}
if len(payload.AttachmentURLs) > 0 {
if err = enqueueKafkaFileAttachmentsBind(ctx, tenantID, &payload.Message, payload.AttachmentURLs); err != nil {
failed, markErr := store.HTTPInfrastructure.MarkKafkaConsumeFailed(tenantID, envelope.EventID, kafkaBus.config.ConsumerGroup,
"KAFKA_FILE_LIFECYCLE_FAILED", err.Error(), time.Now().UTC().Round(time.Millisecond))
if markErr != nil {
return markErr
}
if failed != nil && failed.RetryCount >= kafkaBus.config.MaxRetryCount {
return markKafkaDeadLetter(ctx, tenantID, envelope, failed, "KAFKA_FILE_LIFECYCLE_FAILED", err.Error())
}
sleepKafkaRetryBackoff(failed.RetryCount)
if err != nil && !errors.Is(err, types.ErrDuplicate) {
return err
}
return finishKafkaMessagePersist(ctx, item)
}
return store.HTTPInfrastructure.MarkKafkaConsumeSucceeded(tenantID, envelope.EventID, kafkaBus.config.ConsumerGroup, time.Now().UTC().Round(time.Millisecond))
markFailed := func(err error) (*types.KafkaConsumeRecord, error) {
return store.HTTPInfrastructure.MarkKafkaConsumeFailed(item.tenantID, item.envelope.EventID, kafkaBus.config.ConsumerGroup,
kafkaMessagePersistFailureReason(err), err.Error(), time.Now().UTC().Round(time.Millisecond))
}
markDead := func(failed *types.KafkaConsumeRecord, err error) error {
return markKafkaDeadLetter(ctx, item.tenantID, item.envelope, failed, kafkaMessagePersistFailureReason(err), err.Error())
}
return runKafkaMessageSingleRetries(ctx, kafkaBus.config.MessageRetries, operation, markFailed, markDead, waitKafkaRetry)
}
func runKafkaMessageSingleRetries(ctx context.Context, retryCount int, operation func() error,
markFailed func(error) (*types.KafkaConsumeRecord, error), markDead func(*types.KafkaConsumeRecord, error) error,
wait kafkaRetryWaiter) error {
var lastErr error
var failed *types.KafkaConsumeRecord
for attempt := 1; attempt <= retryCount+1; attempt++ {
if lastErr = operation(); lastErr == nil {
return nil
}
var err error
if failed, err = markFailed(lastErr); err != nil {
return err
}
if attempt <= retryCount {
if err = wait(ctx, kafkaRetryBackoff(attempt)); err != nil {
return err
}
}
}
return markDead(failed, lastErr)
}
func finishKafkaMessagePersist(ctx context.Context, item *kafkaMessagePersistItem) error {
if len(item.payload.AttachmentURLs) > 0 {
if err := enqueueKafkaFileAttachmentsBind(ctx, item.tenantID, &item.payload.Message, item.payload.AttachmentURLs); err != nil {
return fmt.Errorf("file lifecycle enqueue failed: %w", err)
}
}
return store.HTTPInfrastructure.MarkKafkaConsumeSucceeded(item.tenantID, item.envelope.EventID,
kafkaBus.config.ConsumerGroup, time.Now().UTC().Round(time.Millisecond))
}
func kafkaMessagePersistFailureReason(err error) string {
if err != nil && strings.Contains(err.Error(), "file lifecycle") {
return "KAFKA_FILE_LIFECYCLE_FAILED"
}
return kafkaCommandErrorReason(err)
}
func kafkaMessagePersistTopicName(topic string) string {