Navigation Menu

Skip to content

morganstanley/modern-cpp-kafka

About the Modern C++ Kafka API

Lifecycle Active

The modern-cpp-kafka API is a layer of C++ wrapper based on librdkafka (the C part only), with high quality, but more friendly to users.

KAFKA is a registered trademark of The Apache Software Foundation and
has been licensed for use by modern-cpp-kafka. modern-cpp-kafka has no
affiliation with and is not endorsed by The Apache Software Foundation.

Why it's here

The librdkafka is a robust high performance C/C++ library, widely used and well maintained.

Unfortunately, to maintain C++98 compatibility, the C++ interface of librdkafka is not quite object-oriented or user-friendly.

Since C++ is evolving quickly, we want to take advantage of new C++ features, thus making life easier for developers. And this led us to create a new C++ API for Kafka clients.

Eventually, we worked out the modern-cpp-kafka, -- a header-only library that uses idiomatic C++ features to provide a safe, efficient and easy to use way of producing and consuming Kafka messages.

Features

  • Header-only

    • Easy to deploy, and no extra library required to link
  • Ease of Use

    • Interface/Naming matches the Java API

    • Object-oriented

    • RAII is used for lifetime management

    • librdkafka's polling and queue management is now hidden

  • Robust

    • Verified with kinds of test cases, which cover many abnormal scenarios (edge cases)

      • Stability test with unstable brokers

      • Memory leak check for failed client with in-flight messages

      • Client failure and taking over, etc.

  • Efficient

    • No extra performance cost (No deep copy introduced internally)

    • Much better (2~4 times throughput) performance result than those native language (Java/Scala) implementation, in most commonly used cases (message size: 256 B ~ 2 KB)

Installation / Requirements

  • Just include the include/kafka directory for your project

  • The compiler should support C++17

    • Or, C++14, but with pre-requirements

      • Need boost headers (for boost::optional)

      • For GCC compiler, it needs optimization options (e.g. -O2)

  • Dependencies

User Manual

Properties

kafka::Properties Class Reference

  • It is a map which contains all configuration info needed to initialize a Kafka client, and it's the only parameter needed for a constructor.

  • The configuration items are key-value pairs, -- the type of key is always std::string, while the type for a value could be one of the followings

    • std::string

      • Most items are identical with librdkafka configuration

      • But with exceptions

        • Default value changes

          Key String Default Description
          log_level 5 Default was 6 from librdkafka
          client.id random string No default from librdkafka
          group.id random string (for KafkaConsumer only) No default from librdkafka
        • Additional options

          Key String Default Description
          enable.manual.events.poll false To poll the (offset-commit/message-delivery callback) events manually
          max.poll.records 500 (for KafkaConsumer only) The maximum number of records that a single call to poll() would return
        • Ignored options

          Key String Explanation
          enable.auto.offset.store modern-cpp-kafka will save the offsets in its own way
          auto.commit.interval.ms modern-cpp-kafka will only commit the offsets within each poll() operation
    • std::function<...>

      • For kinds of callbacks
      Key String Value Type
      log_cb LogCallback (std::function<void(int, const char*, int, const char* msg)>)
      error_cb ErrorCallback (std::function<void(const Error&)>)
      stats_cb StatsCallback (std::function<void(const std::string&)>)
      oauthbearer_token_refresh_cb OauthbearerTokenRefreshCallback (std::function<SaslOauthbearerToken(const std::string&)>)
    • Interceptors

      • To intercept thread start/exit events, etc.
      Key String Value Type
      interceptors Interceptors

Examples

  1. std::string brokers = "192.168.0.1:9092,192.168.0.2:9092,192.168.0.3:9092";
    
    kafka::Properties props ({
        {"bootstrap.servers",  {brokers}},
        {"enable.idempotence", {"true"}},
    });
    
  2. kafka::Properties props;
    props.put("bootstrap.servers", brokers);
    props.put("enable.idempotence", "true");
    
  • Note: bootstrap.servers is the only mandatory property for a Kafka client

KafkaProducer

kafka::clients::producer::KafkaProducer Class Reference

A Simple Example

Here's a very simple example to see how to send a message with a KafkaProducer.

#include <kafka/KafkaProducer.h>

#include <cstdlib>
#include <iostream>
#include <string>


int main()
{
    using namespace kafka;
    using namespace kafka::clients::producer;

    // E.g. KAFKA_BROKER_LIST: "192.168.0.1:9092,192.168.0.2:9092,192.168.0.3:9092"
    const std::string brokers = getenv("KAFKA_BROKER_LIST"); // NOLINT
    const Topic topic = getenv("TOPIC_FOR_TEST");            // NOLINT

    // Prepare the configuration
    const Properties props({{"bootstrap.servers", brokers}});

    // Create a producer
    KafkaProducer producer(props);

    // Prepare a message
    std::cout << "Type message value and hit enter to produce message..." << std::endl;
    std::string line;
    std::getline(std::cin, line);

    ProducerRecord record(topic, NullKey, Value(line.c_str(), line.size()));

    // Prepare delivery callback
    auto deliveryCb = [](const RecordMetadata& metadata, const Error& error) {
        if (!error) {
            std::cout << "Message delivered: " << metadata.toString() << std::endl;
        } else {
            std::cerr << "Message failed to be delivered: " << error.message() << std::endl;
        }
    };

    // Send a message
    producer.send(record, deliveryCb);

    // Close the producer explicitly(or not, since RAII will take care of it)
    producer.close();
}

Notes

  • The send() is an unblocked operation unless the message buffering queue is full.

  • Make sure the memory block for ProducerRecord's key is valid until the send is called.

  • Make sure the memory block for ProducerRecord's value is valid until the message delivery callback is called (unless the send is with option KafkaProducer::SendOption::ToCopyRecordValue).

  • It's guaranteed that the message delivery callback would be triggered anyway after send, -- a producer would even be waiting for it before close.

  • At the end, we could close Kafka client (i.e. KafkaProducer or KafkaConsumer) explicitly, or just leave it to the destructor.

The Lifecycle of the Message

The message for the KafkaProducer is called ProducerRecord, it contains Topic, Partition (optional), Key and Value. Both Key & Value are const_buffer, and since there's no deep-copy for the Value, the user should make sure the memory block for the Value be valid, until the delivery callback has been executed.

In the previous example, we don't need to worry about the lifecycle of Value, since the content of the line keeps to be available before closing the producer, and all message delivery callbacks would be triggered before finishing closing the producer.

Example for shared_ptr

A trick is capturing the shared pointer (for the memory block of Value) in the message delivery callback.

    std::cout << "Type message value and hit enter to produce message... (empty line to quit)" << std::endl;

    // Get input lines and forward them to Kafka
    for (auto line = std::make_shared<std::string>();
         std::getline(std::cin, *line);
         line = std::make_shared<std::string>()) {

        // Empty line to quit
        if (line->empty()) break;

        // Prepare a message
        ProducerRecord record(topic, NullKey, Value(line->c_str(), line->size()));

        // Prepare delivery callback
        // Note: Here we capture the shared pointer of `line`, which holds the content for `record.value()`
        auto deliveryCb = [line](const RecordMetadata& metadata, const Error& error) {
            if (!error) {
                std::cout << "Message delivered: " << metadata.toString() << std::endl;
            } else {
                std::cerr << "Message failed to be delivered: " << error.message() << std::endl;
            }
        };

        // Send the message
        producer.send(record, deliveryCb);
    }

Example for deep-copy

The option KafkaProducer::SendOption::ToCopyRecordValue could be used for producer.send(...), thus the memory block of record.value() would be copied into the internal sending buffer.

    std::cout << "Type message value and hit enter to produce message... (empty line to quit)" << std::endl;

    // Get input lines and forward them to Kafka
    for (std::string line; std::getline(std::cin, line); ) {

        // Empty line to quit
        if (line.empty()) break;

        // Prepare a message
        ProducerRecord record(topic, NullKey, Value(line.c_str(), line.size()));

        // Prepare delivery callback
        auto deliveryCb = [](const RecordMetadata& metadata, const Error& error) {
            if (!error) {
                std::cout << "Message delivered: " << metadata.toString() << std::endl;
            } else {
                std::cerr << "Message failed to be delivered: " << error.message() << std::endl;
            }
        };

        // Send the message (deep-copy the payload)
        producer.send(record, deliveryCb, KafkaProducer::SendOption::ToCopyRecordValue);
    }

Embed More Info in a ProducerRecord

Besides the payload (i.e. value()), a ProducerRecord could also put extra info in its key() & headers().

Headers is a vector of Header which contains kafka::Header::Key (i.e. std::string) and kafka::Header::Value (i.e. const_buffer).

Example

    const kafka::Topic     topic     = "someTopic";
    const kafka::Partition partition = 0;

    const std::string key       = "some key";
    const std::string value     = "some payload";

    const std::string category  = "categoryA";
    const std::size_t sessionId = 1;

    {
        kafka::clients::producer::ProducerRecord record(topic,
                                                        partition,
                                                        kafka::Key{key.c_str(), key.size()},
                                                        kafka::Value{value.c_str(), value.size()});

        record.headers() = {{
            kafka::Header{kafka::Header::Key{"Category"},  kafka::Header::Value{category.c_str(), category.size()}},
            kafka::Header{kafka::Header::Key{"SessionId"}, kafka::Header::Value{&sessionId, sizeof(sessionId)}}
        }};

        std::cout << "ProducerRecord: " << record.toString() << std::endl;
    }

About enable.manual.events.poll

By default, KafkaProducer would be constructed with enable.manual.events.poll=false configuration. That means, a background thread would be created, which keeps polling the events (thus calls the message delivery callbacks)

Here we have another choice, -- using enable.manual.events.poll=true, thus the MessageDelivery callbacks would be called within member function pollEvents().

  • Note: in this case, the send() will be an unblocked operation even if the message buffering queue is full, -- it would throw an exception (or return an error code with the input reference parameter), instead of blocking there.

Example

    // Prepare the configuration (with "enable.manual.events.poll=true")
    const Properties props({{"bootstrap.servers",         {brokers}},
                            {"enable.manual.events.poll", {"true" }}});

    // Create a producer
    KafkaProducer producer(props);

    std::cout << "Type message value and hit enter to produce message... (empty line to finish)" << std::endl;

    // Get all input lines
    std::list<std::shared_ptr<std::string>> messages;
    for (auto line = std::make_shared<std::string>(); std::getline(std::cin, *line) && !line->empty();) {
        messages.emplace_back(line);
    }

    while (!messages.empty()) {
        // Pop out a message to be sent
        auto payload = messages.front();
        messages.pop_front();

        // Prepare the message
        ProducerRecord record(topic, NullKey, Value(payload->c_str(), payload->size()));

        // Prepare the delivery callback
        // Note: if fails, the message will be pushed back to the sending queue, and then retries later
        auto deliveryCb = [payload, &messages](const RecordMetadata& metadata, const Error& error) {
            if (!error) {
                std::cout << "Message delivered: " << metadata.toString() << std::endl;
            } else {
                std::cerr << "Message failed to be delivered: " << error.message() << ", will be retried later" << std::endl;
                messages.emplace_back(payload);
            }
        };

        // Send the message
        producer.send(record, deliveryCb);

        // Poll events (e.g. message delivery callback)
        producer.pollEvents(std::chrono::milliseconds(0));
    }

Error Handling

kafka::Error might occur at different places while sending a message,

  • A kafka::KafkaException would be triggered if KafkaProducer fails to call the send operation.

  • Delivery kafka::Error could be fetched via the delivery-callback.

  • The kafka::Error::value() for failures

    • Local errors

      • RD_KAFKA_RESP_ERR__UNKNOWN_TOPIC -- The topic doesn't exist

      • RD_KAFKA_RESP_ERR__UNKNOWN_PARTITION -- The partition doesn't exist

      • RD_KAFKA_RESP_ERR__INVALID_ARG -- Invalid topic (topic is null or the length is too long (>512))

      • RD_KAFKA_RESP_ERR__MSG_TIMED_OUT -- No ack received within the time limit

      • RD_KAFKA_RESP_ERR_INVALID_MSG_SIZE -- The message size conflicts with local configuration message.max.bytes

    • Broker errors

      • Error Codes

      • Typical errors are

        • Invalid message: RD_KAFKA_RESP_ERR_CORRUPT_MESSAGE, RD_KAFKA_RESP_ERR_MSG_SIZE_TOO_LARGE, RD_KAFKA_RESP_ERR_INVALID_REQUIRED_ACKS, RD_KAFKA_RESP_ERR_UNSUPPORTED_FOR_MESSAGE_FORMAT, RD_KAFKA_RESP_ERR_RECORD_LIST_TOO_LARGE.

        • Topic/Partition not exist: RD_KAFKA_RESP_ERR_UNKNOWN_TOPIC_OR_PART, -- automatic topic creation is disabled on the broker or the application is specifying a partition that does not exist.

        • Authorization failure: RD_KAFKA_RESP_ERR_TOPIC_AUTHORIZATION_FAILED, RD_KAFKA_RESP_ERR_CLUSTER_AUTHORIZATION_FAILED

Idempotent Producer

The enable.idempotence=true configuration is highly RECOMMENDED.

Example

        kafka::Properties props;
        props.put("bootstrap.servers", brokers);
        props.put("enable.idempotence", "true");

        // Create an idempotent producer
        kafka::clients::producer::KafkaProducer producer(props);

Kafka Consumer

kafka::clients::consumer::KafkaConsumer Class Reference

A Simple Example

#include <kafka/KafkaConsumer.h>

#include <cstdlib>
#include <iostream>
#include <signal.h>
#include <string>

std::atomic_bool running = {true};

void stopRunning(int sig) {
    if (sig != SIGINT) return;

    if (running) {
        running = false;
    } else {
        // Restore the signal handler, -- to avoid stuck with this handler
        signal(SIGINT, SIG_IGN); // NOLINT
    }
}

int main()
{
    using namespace kafka;
    using namespace kafka::clients::consumer;

    // Use Ctrl-C to terminate the program
    signal(SIGINT, stopRunning);    // NOLINT

    // E.g. KAFKA_BROKER_LIST: "192.168.0.1:9092,192.168.0.2:9092,192.168.0.3:9092"
    const std::string brokers = getenv("KAFKA_BROKER_LIST"); // NOLINT
    const Topic topic = getenv("TOPIC_FOR_TEST");            // NOLINT

    // Prepare the configuration
    const Properties props({{"bootstrap.servers", {brokers}}});

    // Create a consumer instance
    KafkaConsumer consumer(props);

    // Subscribe to topics
    consumer.subscribe({topic});

    while (running) {
        // Poll messages from Kafka brokers
        auto records = consumer.poll(std::chrono::milliseconds(100));

        for (const auto& record: records) {
            if (!record.error()) {
                std::cout << "Got a new message..." << std::endl;
                std::cout << "    Topic    : " << record.topic() << std::endl;
                std::cout << "    Partition: " << record.partition() << std::endl;
                std::cout << "    Offset   : " << record.offset() << std::endl;
                std::cout << "    Timestamp: " << record.timestamp().toString() << std::endl;
                std::cout << "    Headers  : " << toString(record.headers()) << std::endl;
                std::cout << "    Key   [" << record.key().toString() << "]" << std::endl;
                std::cout << "    Value [" << record.value().toString() << "]" << std::endl;
            } else {
                std::cerr << record.toString() << std::endl;
            }
        }
    }

    // No explicit close is needed, RAII will take care of it
    consumer.close();
}
  • By default, the KafkaConsumer is constructed with property enable.auto.commit=true

    • It means it will automatically commit previously polled offsets on each poll (and the final close) operations.

      • Note: the internal offset commit is asynchronous, which is not guaranteed to succeed. Since the operation is supposed to be triggered (again) at a later time (within each poll), thus the occasional failure doesn't matter.
  • subscribe could take a topic list. It's a block operation, and would wait for the consumer to get partitions assigned.

  • poll must be called periodically, thus to trigger kinds of callback handling internally. In practice, it could be put in a while loop.

Rebalance events

The KafkaConsumer could specify the RebalanceCallback while it subscribes the topics, and the callback will be triggered while partitions are assigned or revoked.

Example

    // The consumer would read all messages from the topic and then quit.

    // Prepare the configuration
    const Properties props({{"bootstrap.servers",    {brokers}},
                            // Emit RD_KAFKA_RESP_ERR__PARTITION_EOF event
                            // whenever the consumer reaches the end of a partition.
                            {"enable.partition.eof", {"true"}},
                            // Action to take when there is no initial offset in offset store
                            // it means the consumer would read from the very beginning
                            {"auto.offset.reset",    {"earliest"}}});

    // Create a consumer instance
    KafkaConsumer consumer(props);

    // Prepare the rebalance callbacks
    std::atomic<std::size_t> assignedPartitions{};
    auto rebalanceCb = [&assignedPartitions](kafka::clients::consumer::RebalanceEventType et, const kafka::TopicPartitions& tps) {
                           if (et == kafka::clients::consumer::RebalanceEventType::PartitionsAssigned) {
                               assignedPartitions += tps.size();
                               std::cout << "Assigned partitions: " << kafka::toString(tps) << std::endl;
                           } else {
                               assignedPartitions -= tps.size();
                               std::cout << "Revoked partitions: " << kafka::toString(tps) << std::endl;
                           }
                       };

    // Subscribe to topics with rebalance callback
    consumer.subscribe({topic}, rebalanceCb);

    TopicPartitions finishedPartitions;
    while (finishedPartitions.size() != assignedPartitions.load()) {
        // Poll messages from Kafka brokers
        auto records = consumer.poll(std::chrono::milliseconds(100));

        for (const auto& record: records) {
            if (!record.error()) {
                std::cerr << record.toString() << std::endl;
            } else {
                if (record.error().value() == RD_KAFKA_RESP_ERR__PARTITION_EOF) {
                    // Record the partition which has been reached the end
                    finishedPartitions.emplace(record.topic(), record.partition());
                } else {
                    std::cerr << record.toString() << std::endl;
                }
            }
        }
    }

To Commit Offset Manually

Once the KafkaConsumer is configured with enable.auto.commit=false, the user has to find out the right places to call commitSync(...)/commitAsync(...).

Example

    // Prepare the configuration
    Properties props({{"bootstrap.servers", {brokers}}});
    props.put("enable.auto.commit", "false");

    // Create a consumer instance
    KafkaConsumer consumer(props);

    // Subscribe to topics
    consumer.subscribe({topic});

    while (running) {
        auto records = consumer.poll(std::chrono::milliseconds(100));

        for (const auto& record: records) {
            std::cout << record.toString() << std::endl;
        }

        if (!records.empty()) {
            consumer.commitAsync();
        }
    }

    consumer.commitSync();

    // No explicit close is needed, RAII will take care of it
    // consumer.close();

Error Handling

Callbacks for KafkaClient

We're free to set callbacks in Properties with a kafka::clients::ErrorCallback, kafka::clients::LogCallback, or kafka::clients::StatsCallback.

Example

    // Prepare the configuration
    Properties props({{"bootstrap.servers", {brokers}}});

    // To print out the error
    props.put("error_cb", [](const kafka::Error& error) {
                              // https://en.wikipedia.org/wiki/ANSI_escape_code
                              std::cerr << "\033[1;31m" << "[" << kafka::utility::getCurrentTime() << "] ==> Met Error: " << "\033[0m";
                              std::cerr << "\033[4;35m" << error.toString() << "\033[0m" << std::endl;
                          });

    // To enable the debug-level log
    props.put("log_level", "7");
    props.put("debug", "all");
    props.put("log_cb", [](int /*level*/, const char* /*filename*/, int /*lineno*/, const char* msg) {
                            std::cout << "[" << kafka::utility::getCurrentTime() << "]" << msg << std::endl;
                        });

    // To enable the statistics dumping
    props.put("statistics.interval.ms", "1000");
    props.put("stats_cb", [](const std::string& jsonString) {
                              std::cout << "Statistics: " << jsonString << std::endl;
                          });

Thread Model

  • Number of Background Threads within a Kafka Client

    • N threads for the message transmission (towards N brokers).

    • 2 (for KafkaProducer) / 3 (for KafkaConsumer) threads to handle internal operations, timers, consumer group operations, etc.

    • 1 thread for (message-delivery/offset-commit) callback events polling, -- the thread only exists while the client is configured with enable.manual.events.poll=false (the default config)

  • Which Thread Handles the Callbacks

    • consumer::RebalanceCallback: the thread which calls consumer.poll(...)

    • consumer::OffsetCommitCallback

      • While enable.manual.events.poll=false: the thread which calls consumer.pollEvents(...)

      • While enable.manual.events.poll=true: the background (events polling) thread

    • producer::Callback

      • While enable.manual.events.poll=false: the thread which calls producer.pollEvents(...)

      • While enable.manual.events.poll=true: the background (events polling) thread

For Developers

Build (for tests/tools/examples)

  • Specify library locations with environment variables

    Environment Variable Description
    LIBRDKAFKA_INCLUDE_DIR librdkafka headers
    LIBRDKAFKA_LIBRARY_DIR librdkafka libraries
    GTEST_ROOT googletest headers and libraries
    BOOST_ROOT boost headers and libraries
    SASL_LIBRARYDIR/SASL_LIBRARY [optional] for SASL connection support
    RAPIDJSON_INCLUDE_DIRS addons/KafkaMetrics.h requires rapidjson headers
  • Build commands

    • cd empty-folder-for-build

    • cmake path-to-project-root (following options could be used with -D)

      Build Option Description
      BUILD_OPTION_USE_TSAN=ON Use Thread Sanitizer
      BUILD_OPTION_USE_ASAN=ON Use Address Sanitizer
      BUILD_OPTION_USE_UBSAN=ON Use Undefined Behavior Sanitizer
      BUILD_OPTION_CLANG_TIDY=ON Enable clang-tidy checking
      BUILD_OPTION_GEN_DOC=ON Generate documentation as well
      BUILD_OPTION_DOC_ONLY=ON Only generate documentation
      BUILD_OPTION_GEN_COVERAGE=ON Generate test coverage, only support by clang currently
    • make

    • make install (to install tools)

Run Tests

  • Kafka cluster setup

  • To run the binary, the test runner requires following environment variables

    Environment Variable Descrioption Example
    KAFKA_BROKER_LIST The broker list for the Kafka cluster export KAFKA_BROKER_LIST=127.0.0.1:29091,127.0.0.1:29092,127.0.0.1:29093
    KAFKA_BROKER_PIDS The broker PIDs for test runner to manipulate export KAFKA_BROKER_PIDS=61567,61569,61571
    KAFKA_CLIENT_ADDITIONAL_SETTINGS Could be used for addtional configuration for Kafka clients export KAFKA_CLIENT_ADDITIONAL_SETTINGS="security.protocol=SASL_PLAINTEXT;sasl.kerberos.service.name=...;sasl.kerberos.keytab=...;sasl.kerberos.principal=..."
    • The environment variable KAFKA_BROKER_LIST is mandatory for integration/robustness test, which requires the Kafka cluster.

    • The environment variable KAFKA_BROKER_PIDS is mandatory for robustness test, which requires the Kafka cluster and the privilege to stop/resume the brokers.

    Test Type KAFKA_BROKER_LIST KAFKA_BROKER_PIDS
    tests/unit - -
    tests/integration Required -
    tests/robustness Required Required