ClickHouse/src/Storages/MessageQueueSink.h

62 lines
1.9 KiB
C++
Raw Normal View History

#pragma once
#include <Processors/Sinks/SinkToStorage.h>
#include <Storages/IMessageProducer.h>
#include <Interpreters/Context.h>
namespace DB
{
class IOutputFormat;
class IRowOutputFormat;
using IOutputFormatPtr = std::shared_ptr<IOutputFormat>;
/// Storage sink for streaming engines like Kafka/RabbitMQ/NATS.
/// It implements formatting input data into messages.
/// For row-based formats like TSV, CSV, JSONEachRow, etc, each message
/// contains from 1 to max_rows_per_message rows.
/// For block-based formats like Native, Arrow, Parquet, the whole block is formatted into one message.
/// Each message is created independently, so it contains all format
/// prefixes/suffixes and can fully parsed back by corresponding input format.
/// After formatting, created message is propagated to IMessageProducer::produce() method.
/// To use MessageQueueSink for specific streaming engine, you should implement
/// IMessageProducer for it.
2022-12-15 19:47:10 +00:00
class MessageQueueSink : public SinkToStorage, WithContext
{
public:
MessageQueueSink(
const Block & header,
const String & format_name_,
size_t max_rows_per_message_,
std::unique_ptr<IMessageProducer> producer_,
const String & storage_name_,
const ContextPtr & context_);
String getName() const override { return storage_name + "Sink"; }
void consume(Chunk chunk) override;
void onStart() override;
void onFinish() override;
void onCancel() override { onFinish(); }
void onException() override { onFinish(); }
protected:
/// Do some specific initialization before consuming data.
virtual void initialize() {}
private:
const String format_name;
size_t max_rows_per_message;
std::unique_ptr<WriteBufferFromOwnString> buffer;
IOutputFormatPtr format;
IRowOutputFormat * row_format;
std::unique_ptr<IMessageProducer> producer;
const String storage_name;
};
}