Flow control

This document provides information about using flow control with messages published to a topic.

About flow control

A publisher client may attempt to publish messages faster than that client is capable of sending data to the Pub/Sub service. Clients are limited by many factors, including:

• Machine CPU, RAM, and network capacity
• Network settings, such as the number of outstanding requests and available bandwidth
• The latency of each publish request, largely determined by the network connections between the Pub/Sub service, the client, and Google Cloud

If the publish request rate exceeds these limits, requests accumulate in memory until they fail with a DEADLINE_EXCEEDED error. This is especially likely when tens of thousands of messages are published in a loop, generating thousands of requests in milliseconds.

You can diagnose this issue by checking the server side metrics in Monitoring. You won't be able to see the requests that have failed with DEADLINE_EXCEEDED, only the successful requests. The rate of successful requests tells you the throughput capacity of your client machines, providing a baseline for configuring flow control.

Go to the Monitoring page

To mitigate flow rate issues, configure your publisher client with flow control to limit the rate of publish requests. You can configure the maximum number of bytes allocated for outstanding requests, and the maximum number of outstanding messages permitted. Set these limits according to the throughput capacity of your client machines.

Note: Use flow control if messages are periodically published at a higher rate than they can be sent. If this is a persistent problem rather than a transient spike in messages, consider increasing the number of publisher client instances.

Before you begin

Before configuring the publish workflow, ensure you have completed the following tasks:

• Learn about topics and the publishing workflow.
• Create a topic.

Required roles

To get the permissions that you need to use flow control, ask your administrator to grant you the Pub/Sub Publisher (roles/pubsub.publisher) IAM role on your topic. For more information about granting roles, see Manage access to projects, folders, and organizations.

You might also be able to get the required permissions through custom roles or other predefined roles.

You need additional permissions to create or update topics and subscriptions.

Use flow control with messages

Publisher flow control is available using the Pub/Sub client libraries in the following languages:

C++

Before trying this sample, follow the C++ setup instructions in Quickstart: Using Client Libraries. For more information, see the Pub/Sub C++ API reference documentation.

namespace pubsub = ::google::cloud::pubsub;
using ::google::cloud::future;
using ::google::cloud::Options;
using ::google::cloud::StatusOr;
[](std::string project_id, std::string topic_id) {
  auto topic = pubsub::Topic(std::move(project_id), std::move(topic_id));
  // Configure the publisher to block if either (1) 100 or more messages, or
  // (2) messages with 100MiB worth of data have not been acknowledged by the
  // service. By default the publisher never blocks, and its capacity is only
  // limited by the system's memory.
  auto publisher = pubsub::Publisher(pubsub::MakePublisherConnection(
      std::move(topic),
      Options{}
          .set<pubsub::MaxPendingMessagesOption>(100)
          .set<pubsub::MaxPendingBytesOption>(100 * 1024 * 1024L)
          .set<pubsub::FullPublisherActionOption>(
              pubsub::FullPublisherAction::kBlocks)));

  std::vector<future<void>> ids;
  for (char const* data : {"a", "b", "c"}) {
    ids.push_back(
        publisher.Publish(pubsub::MessageBuilder().SetData(data).Build())
            .then([data](future<StatusOr<std::string>> f) {
              auto s = f.get();
              if (!s) return;
              std::cout << "Sent '" << data << "' (" << *s << ")\n";
            }));
  }
  publisher.Flush();
  // Block until they are actually sent.
  for (auto& id : ids) id.get();
}

Go

Before trying this sample, follow the Go setup instructions in Quickstart: Using Client Libraries. For more information, see the Pub/Sub Go API reference documentation.

import (
	"context"
	"fmt"
	"io"
	"strconv"
	"sync"
	"sync/atomic"

	"cloud.google.com/go/pubsub"
)

func publishWithFlowControlSettings(w io.Writer, projectID, topicID string) error {
	// projectID := "my-project-id"
	// topicID := "my-topic"
	ctx := context.Background()
	client, err := pubsub.NewClient(ctx, projectID)
	if err != nil {
		return fmt.Errorf("pubsub.NewClient: %w", err)
	}
	defer client.Close()

	t := client.Topic(topicID)
	t.PublishSettings.FlowControlSettings = pubsub.FlowControlSettings{
		MaxOutstandingMessages: 100,                     // default 1000
		MaxOutstandingBytes:    10 * 1024 * 1024,        // default 0 (unlimited)
		LimitExceededBehavior:  pubsub.FlowControlBlock, // default Ignore, other options: Block and SignalError
	}

	var wg sync.WaitGroup
	var totalErrors uint64

	numMsgs := 1000
	// Rapidly publishing 1000 messages in a loop may be constrained by flow control.
	for i := 0; i < numMsgs; i++ {
		wg.Add(1)
		result := t.Publish(ctx, &pubsub.Message{
			Data: []byte("message #" + strconv.Itoa(i)),
		})
		go func(i int, res *pubsub.PublishResult) {
			fmt.Fprintf(w, "Publishing message %d\n", i)
			defer wg.Done()
			// The Get method blocks until a server-generated ID or
			// an error is returned for the published message.
			_, err := res.Get(ctx)
			if err != nil {
				// Error handling code can be added here.
				fmt.Fprintf(w, "Failed to publish: %v", err)
				atomic.AddUint64(&totalErrors, 1)
				return
			}
		}(i, result)
	}

	wg.Wait()

	if totalErrors > 0 {
		return fmt.Errorf("%d of %d messages did not publish successfully", totalErrors, numMsgs)
	}
	return nil
}

Java

Before trying this sample, follow the Java setup instructions in Quickstart: Using Client Libraries. For more information, see the Pub/Sub Java API reference documentation.

import com.google.api.core.ApiFuture;
import com.google.api.core.ApiFutures;
import com.google.api.gax.batching.BatchingSettings;
import com.google.api.gax.batching.FlowControlSettings;
import com.google.api.gax.batching.FlowController.LimitExceededBehavior;
import com.google.cloud.pubsub.v1.Publisher;
import com.google.protobuf.ByteString;
import com.google.pubsub.v1.PubsubMessage;
import com.google.pubsub.v1.TopicName;
import java.io.IOException;
import java.util.ArrayList;
import java.util.List;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.TimeUnit;

public class PublishWithFlowControlExample {
  public static void main(String... args) throws Exception {
    // TODO(developer): Replace these variables before running the sample.
    String projectId = "your-project-id";
    String topicId = "your-topic-id";

    publishWithFlowControlExample(projectId, topicId);
  }

  public static void publishWithFlowControlExample(String projectId, String topicId)
      throws IOException, ExecutionException, InterruptedException {
    TopicName topicName = TopicName.of(projectId, topicId);
    Publisher publisher = null;
    List<ApiFuture<String>> messageIdFutures = new ArrayList<>();

    try {
      // Configure how many messages the publisher client can hold in memory
      // and what to do when messages exceed the limit.
      FlowControlSettings flowControlSettings =
          FlowControlSettings.newBuilder()
              // Block more messages from being published when the limit is reached. The other
              // options are Ignore (or continue publishing) and ThrowException (or error out).
              .setLimitExceededBehavior(LimitExceededBehavior.Block)
              .setMaxOutstandingRequestBytes(10 * 1024 * 1024L) // 10 MiB
              .setMaxOutstandingElementCount(100L) // 100 messages
              .build();

      // By default, messages are not batched.
      BatchingSettings batchingSettings =
          BatchingSettings.newBuilder().setFlowControlSettings(flowControlSettings).build();

      publisher = Publisher.newBuilder(topicName).setBatchingSettings(batchingSettings).build();

      // Publish 1000 messages in quick succession may be constrained by publisher flow control.
      for (int i = 0; i < 1000; i++) {
        String message = "message " + i;
        ByteString data = ByteString.copyFromUtf8(message);
        PubsubMessage pubsubMessage = PubsubMessage.newBuilder().setData(data).build();

        // Once published, returns a server-assigned message id (unique within the topic)
        ApiFuture<String> messageIdFuture = publisher.publish(pubsubMessage);
        messageIdFutures.add(messageIdFuture);
      }
    } finally {
      // Wait on any pending publish requests.
      List<String> messageIds = ApiFutures.allAsList(messageIdFutures).get();

      System.out.println(
          "Published " + messageIds.size() + " messages with flow control settings.");

      if (publisher != null) {
        // When finished with the publisher, shut down to free up resources.
        publisher.shutdown();
        publisher.awaitTermination(1, TimeUnit.MINUTES);
      }
    }
  }
}

Node.js

Before trying this sample, follow the Node.js setup instructions in Quickstart: Using Client Libraries. For more information, see the Pub/Sub Node.js API reference documentation.

/**
 * TODO(developer): Uncomment this variable before running the sample.
 */
// const topicNameOrId = 'YOUR_TOPIC_NAME_OR_ID';

// Imports the Google Cloud client library
const {PubSub} = require('@google-cloud/pubsub');

// Creates a client; cache this for further use
const pubSubClient = new PubSub();

async function publishWithFlowControl(topicNameOrId) {
  // Create publisher options
  const options = {
    flowControlOptions: {
      maxOutstandingMessages: 50,
      maxOutstandingBytes: 10 * 1024 * 1024, // 10 MB
    },
  };

  // Get a publisher. Cache topic objects (publishers) and reuse them.
  const topic = pubSubClient.topic(topicNameOrId, options);

  // For flow controlled publishing, we'll use a publisher flow controller
  // instead of `topic.publish()`.
  const flow = topic.flowControlled();

  // Publish messages in a fast loop.
  const testMessage = {data: Buffer.from('test!')};
  for (let i = 0; i < 1000; i++) {
    // You can also just `await` on `publish()` unconditionally, but if
    // you want to avoid pausing to the event loop on each iteration,
    // you can manually check the return value before doing so.
    const wait = flow.publish(testMessage);
    if (wait) {
      await wait;
    }
  }

  // Wait on any pending publish requests. Note that you can call `all()`
  // earlier if you like, and it will return a Promise for all messages
  // that have been sent to `flowController.publish()` so far.
  const messageIds = await flow.all();
  console.log(`Published ${messageIds.length} with flow control settings.`);
}

Node.js

Before trying this sample, follow the Node.js setup instructions in Quickstart: Using Client Libraries. For more information, see the Pub/Sub Node.js API reference documentation.

/**
 * TODO(developer): Uncomment this variable before running the sample.
 */
// const topicNameOrId = 'YOUR_TOPIC_NAME_OR_ID';

// Imports the Google Cloud client library
import {PubSub, PublishOptions} from '@google-cloud/pubsub';

// Creates a client; cache this for further use
const pubSubClient = new PubSub();

async function publishWithFlowControl(topicNameOrId: string) {
  // Create publisher options
  const options: PublishOptions = {
    flowControlOptions: {
      maxOutstandingMessages: 50,
      maxOutstandingBytes: 10 * 1024 * 1024, // 10 MB
    },
  };

  // Get a publisher. Cache topic objects (publishers) and reuse them.
  const topic = pubSubClient.topic(topicNameOrId, options);

  // For flow controlled publishing, we'll use a publisher flow controller
  // instead of `topic.publish()`.
  const flow = topic.flowControlled();

  // Publish messages in a fast loop.
  const testMessage = {data: Buffer.from('test!')};
  for (let i = 0; i < 1000; i++) {
    // You can also just `await` on `publish()` unconditionally, but if
    // you want to avoid pausing to the event loop on each iteration,
    // you can manually check the return value before doing so.
    const wait = flow.publish(testMessage);
    if (wait) {
      await wait;
    }
  }

  // Wait on any pending publish requests. Note that you can call `all()`
  // earlier if you like, and it will return a Promise for all messages
  // that have been sent to `flowController.publish()` so far.
  const messageIds = await flow.all();
  console.log(`Published ${messageIds.length} with flow control settings.`);
}

Python

Before trying this sample, follow the Python setup instructions in Quickstart: Using Client Libraries. For more information, see the Pub/Sub Python API reference documentation.

from concurrent import futures
from google.cloud import pubsub_v1
from google.cloud.pubsub_v1.types import (
    LimitExceededBehavior,
    PublisherOptions,
    PublishFlowControl,
)

# TODO(developer)
# project_id = "your-project-id"
# topic_id = "your-topic-id"

# Configure how many messages the publisher client can hold in memory
# and what to do when messages exceed the limit.
flow_control_settings = PublishFlowControl(
    message_limit=100,  # 100 messages
    byte_limit=10 * 1024 * 1024,  # 10 MiB
    limit_exceeded_behavior=LimitExceededBehavior.BLOCK,
)
publisher = pubsub_v1.PublisherClient(
    publisher_options=PublisherOptions(flow_control=flow_control_settings)
)
topic_path = publisher.topic_path(project_id, topic_id)
publish_futures = []

# Resolve the publish future in a separate thread.
def callback(publish_future: pubsub_v1.publisher.futures.Future) -> None:
    message_id = publish_future.result()
    print(message_id)

# Publish 1000 messages in quick succession may be constrained by
# publisher flow control.
for n in range(1, 1000):
    data_str = f"Message number {n}"
    # Data must be a bytestring
    data = data_str.encode("utf-8")
    publish_future = publisher.publish(topic_path, data)
    # Non-blocking. Allow the publisher client to batch messages.
    publish_future.add_done_callback(callback)
    publish_futures.append(publish_future)

futures.wait(publish_futures, return_when=futures.ALL_COMPLETED)

print(f"Published messages with flow control settings to {topic_path}.")

Ruby

Before trying this sample, follow the Ruby setup instructions in Quickstart: Using Client Libraries. For more information, see the Pub/Sub Ruby API reference documentation.

# topic_id = "your-topic-id"

pubsub = Google::Cloud::Pubsub.new

topic = pubsub.topic topic_id, async: {
  # Configure how many messages the publisher client can hold in memory
  # and what to do when messages exceed the limit.
  flow_control: {
    message_limit: 100,
    byte_limit: 10 * 1024 * 1024, # 10 MiB
    # Block more messages from being published when the limit is reached. The
    # other options are :ignore and :error.
    limit_exceeded_behavior: :block
  }
}
# Rapidly publishing 1000 messages in a loop may be constrained by flow control.
1000.times do |i|
  topic.publish_async "message #{i}" do |result|
    raise "Failed to publish the message." unless result.succeeded?
  end
end

# Stop the async_publisher to send all queued messages immediately.
topic.async_publisher.stop.wait!
puts "Published messages with flow control settings to #{topic_id}."

What's next

• Compress messages

• Retry requests

• Concurrency control

• Batch messaging