kafka::Producer Class Reference

#include <userver/kafka/producer.hpp>

Detailed Description

Apache Kafka Producer Client.

It exposes the API to send an arbitrary message to the Kafka Broker to certain topic by given key and optional partition.

Important implementation details

Message send tasks handling other messages' delivery reports, suspending their execution while no events exist. This makes message production parallel and leads to high Producer scalability.

Producer maintains per topic statistics including the broker connection errors.

Remarks

Destructor may wait for no more than a 2 x delivery_timeout to ensure all sent messages are properly delivered.

See also

https://docs.confluent.io/platform/current/clients/producer.html

Examples

samples/kafka_service/src/produce.cpp, samples/kafka_service/src/produce.hpp, and samples/kafka_service/src/producer_handler.hpp.

Definition at line 46 of file producer.hpp.

Public Member Functions
	Producer (const std::string &name, engine::TaskProcessor &producer_task_processor, const impl::ProducerConfiguration &configuration, const impl::Secret &secrets)
	Creates the Kafka Producer.
	~Producer ()
	Waits until all messages are sent for at most 2 x delivery_timeout and destroys the producer.
	Producer (const Producer &)=delete
	Producer (Producer &&)=delete
Producer &	operator= (const Producer &)=delete
Producer &	operator= (Producer &&)=delete
void	Send (const std::string &topic_name, std::string_view key, std::string_view message, std::optional< std::uint32_t > partition=std::nullopt) const
	Sends given message to topic topic_name by given key and partition (if passed) with payload contains the message data. Asynchronously waits until the message is delivered or the delivery error occurred.
engine::TaskWithResult< void >	SendAsync (std::string topic_name, std::string key, std::string message, std::optional< std::uint32_t > partition=std::nullopt) const
	Same as Producer::Send, but returns the task which can be used to wait the message delivery manually.
void	DumpMetric (utils::statistics::Writer &writer) const
	Dumps per topic messages produce statistics. No expected to be called manually.

Constructor & Destructor Documentation

Producer()

kafka::Producer::Producer	(	const std::string &	name,
		engine::TaskProcessor &	producer_task_processor,
		const impl::ProducerConfiguration &	configuration,
		const impl::Secret &	secrets )

Creates the Kafka Producer.

Parameters

producer_task_processor

is task processor where producer creates tasks for message delivery scheduling and waiting.

~Producer()

kafka::Producer::~Producer

(

)

Waits until all messages are sent for at most 2 x delivery_timeout and destroys the producer.

Remarks

In a basic producer use cases, the destructor returns immediately.

Member Function Documentation

DumpMetric()

void kafka::Producer::DumpMetric

(

utils::statistics::Writer &

writer

)

const

Dumps per topic messages produce statistics. No expected to be called manually.

See also

kafka/impl/stats.hpp

Send()

void kafka::Producer::Send	(	const std::string &	topic_name,
		std::string_view	key,
		std::string_view	message,
		std::optional< std::uint32_t >	partition = std::nullopt ) const

Sends given message to topic topic_name by given key and partition (if passed) with payload contains the message data. Asynchronously waits until the message is delivered or the delivery error occurred.

No payload data is copied. Method holds the data until message is delivered.

Thread-safe and can be called from any number of threads concurrently.

If partition not passed, partition is chosen by internal Kafka partitioner.

Warning

if enable_idempotence option is enabled, do not use both explicit partitions and Kafka-chosen ones.

Exceptions

SendException

and its descendants if message is not delivered and acked by Kafka Broker in configured timeout.

Note

Use SendException::IsRetryable method to understand whether there is a sense to retry the message sending.

    bool delivered{false};
    const auto deadline = engine::Deadline::FromDuration(producer_configuration.delivery_timeout);
    while (!delivered && !deadline.IsReached()) {
        try {
            make_send_request();
            delivered = true;
        } catch (const kafka::SendException& e) {
            if (e.IsRetryable()) {
                engine::InterruptibleSleepFor(std::chrono::milliseconds{10});
                continue;
            }
            break;
        }
    }

Examples

samples/kafka_service/src/produce.cpp.

SendAsync()

engine::TaskWithResult< void > kafka::Producer::SendAsync	(	std::string	topic_name,
		std::string	key,
		std::string	message,
		std::optional< std::uint32_t >	partition = std::nullopt ) const

Same as Producer::Send, but returns the task which can be used to wait the message delivery manually.

Warning

If user schedules a batch of send requests with Producer::SendAsync, some send requests may be retried by the library (for instance, in case of network blink). Though, the order messages are written to partition may differ from the order messages are initially sent

    std::vector<engine::TaskWithResult<void>> results;
    results.reserve(kSendCount);
    for (std::size_t send{0}; send < kSendCount; ++send) {
        results.emplace_back(
            producer.SendAsync(topic, fmt::format("test-key-{}", send), fmt::format("test-msg-{}", send))
        );
    }
 
    UEXPECT_NO_THROW(engine::WaitAllChecked(results));

---

The documentation for this class was generated from the following file:

• userver/kafka/producer.hpp