[NIT-2868] Allow pubsub interface to return an error

pubsub/common.go

@@ -11,6 +11,7 @@ import (
 )
 
 func ResultKeyFor(streamName, id string) string { return fmt.Sprintf("%s.%s", streamName, id) }
+func ErrorKeyFor(streamName, id string) string  { return fmt.Sprintf("%s.%s.error", streamName, id) }
 
 // CreateStream tries to create stream with given name, if it already exists
 // does not return an error.

pubsub/consumer.go

@@ -36,6 +36,8 @@ var TestConsumerConfig = ConsumerConfig{
 	IdletimeToAutoclaim:  30 * time.Millisecond,
 }
 
+var AlreadySetError = errors.New("redis key already set")
+
 func ConsumerConfigAddOptions(prefix string, f *pflag.FlagSet) {
 	f.Duration(prefix+".response-entry-timeout", DefaultConsumerConfig.ResponseEntryTimeout, "timeout for response entry")
 	f.Duration(prefix+".idletime-to-autoclaim", DefaultConsumerConfig.IdletimeToAutoclaim, "After a message spends this amount of time in PEL (Pending Entries List i.e claimed by another consumer but not Acknowledged) it will be allowed to be autoclaimed by other consumers")
@@ -221,7 +223,10 @@ func (c *Consumer[Request, Response]) SetResult(ctx context.Context, messageID s
 	resultKey := ResultKeyFor(c.StreamName(), messageID)
 	log.Debug("consumer: setting result", "cid", c.id, "msgIdInStream", messageID, "resultKeyInRedis", resultKey)
 	acquired, err := c.client.SetNX(ctx, resultKey, resp, c.cfg.ResponseEntryTimeout).Result()
-	if err != nil || !acquired {
+	if !acquired && err == nil {
+		err = AlreadySetError
+	}
+	if err != nil {
 		return fmt.Errorf("setting result for message with message-id in stream: %v, error: %w", messageID, err)
 	}
 	log.Debug("consumer: xack", "cid", c.id, "messageId", messageID)
@@ -233,3 +238,27 @@ func (c *Consumer[Request, Response]) SetResult(ctx context.Context, messageID s
 	}
 	return nil
 }
+
+func (c *Consumer[Request, Response]) SetError(ctx context.Context, messageID string, error string) error {
+	resp, err := json.Marshal(error)
+	if err != nil {
+		return fmt.Errorf("marshaling result: %w", err)
+	}
+	errorKey := ErrorKeyFor(c.StreamName(), messageID)
+	log.Debug("consumer: setting error", "cid", c.id, "msgIdInStream", messageID, "errorKeyInRedis", errorKey)
+	acquired, err := c.client.SetNX(ctx, errorKey, resp, c.cfg.ResponseEntryTimeout).Result()
+	if !acquired && err == nil {
+		err = AlreadySetError
+	}
+	if err != nil {
+		return fmt.Errorf("setting error for message with message-id in stream: %v, error: %w", messageID, err)
+	}
+	log.Debug("consumer: xack", "cid", c.id, "messageId", messageID)
+	if _, err := c.client.XAck(ctx, c.redisStream, c.redisGroup, messageID).Result(); err != nil {
+		return fmt.Errorf("acking message: %v, error: %w", messageID, err)
+	}
+	if _, err := c.client.XDel(ctx, c.redisStream, messageID).Result(); err != nil {
+		return fmt.Errorf("deleting message: %v, error: %w", messageID, err)
+	}
+	return nil
+}

pubsub/producer.go

@@ -142,6 +142,24 @@ func (p *Producer[Request, Response]) checkResponses(ctx context.Context) time.D
 			return 0
 		}
 		checked++
+		// First check if there is an error for this promise
+		errorKey := ErrorKeyFor(p.redisStream, id)
+		errorResponse, err := p.client.Get(ctx, errorKey).Result()
+		if err != nil && !errors.Is(err, redis.Nil) {
+			// If we get an error that is not redis.Nil, then log it and continue.
+			log.Error("Error reading error in redis", "key", errorKey, "error", err)
+			continue
+		}
+		if err == nil {
+			// If we found the error key, then delete it and return the error to the promise and continue.
+			p.client.Del(ctx, errorKey)
+			promise.ProduceError(errors.New(errorResponse))
+			log.Error("error getting response", "error", errorResponse)
+			errored++
+			delete(p.promises, id)
+			continue
+		}
+		// If we do not find the error key, then check for the result key.
 		resultKey := ResultKeyFor(p.redisStream, id)
 		res, err := p.client.Get(ctx, resultKey).Result()
 		if err != nil {

validator/valnode/redis/consumer.go

@@ -163,7 +163,11 @@ func (s *ValidationServer) Start(ctx_in context.Context) {
 				res, err := valRun.Await(ctx)
 				if err != nil {
 					log.Error("Error validating", "request value", work.req.Value, "error", err)
+					err := s.consumers[work.moduleRoot].SetError(ctx, work.req.ID, err.Error())
 					work.req.Ack()
+					if err != nil {
+						log.Error("Error setting error for request", "id", work.req.ID, "error", err)
+					}
 				} else {
 					log.Debug("done work", "thread", i, "workid", work.req.ID)
 					err := s.consumers[work.moduleRoot].SetResult(ctx, work.req.ID, res)