cluster.hpp

Go to the documentation of this file.

#pragma once
 
/// @file userver/storages/clickhouse/cluster.hpp
/// @brief @copybrief storages::clickhouse::Cluster
 
#include <atomic>
#include <string>
#include <string_view>
#include <vector>
 
#include <userver/clients/dns/resolver_fwd.hpp>
#include <userver/components/component_fwd.hpp>
 
#include <userver/storages/clickhouse/fwd.hpp>
#include <userver/storages/clickhouse/impl/insertion_request.hpp>
#include <userver/storages/clickhouse/impl/pool.hpp>
#include <userver/storages/clickhouse/options.hpp>
#include <userver/storages/clickhouse/query.hpp>
#include <userver/utils/statistics/writer.hpp>
 
USERVER_NAMESPACE_BEGIN
 
namespace storages::clickhouse {
 
class ExecutionResult;
 
namespace impl {
struct ClickhouseSettings;
}
 
/// @ingroup userver_clients
///
/// @brief Interface for executing queries on a cluster of ClickHouse servers.
///
/// Usually retrieved from components::ClickHouse component.
class Cluster final {
public:
    /// Cluster constructor
    /// @param resolver asynchronous DNS resolver
    /// @param settings struct with settings fields:
    /// endpoints - list of endpoints (host + port)
    /// auth_settings - authentication settings (user, password, database)
    /// @param config components::ClickHouse component config
    Cluster(
        clients::dns::Resolver& resolver,
        const impl::ClickhouseSettings& settings,
        const components::ComponentConfig& config
    );
    /// Cluster destructor
    ~Cluster();
 
    Cluster(const Cluster&) = delete;
 
    /// @brief Execute a statement at some host of the cluster
    /// with args as query parameters.
    ///
    /// It is convinient to keep SQL queries in separate files, see @ref scripts/docs/en/userver/sql_files.md
    /// for more info.
    template <typename... Args>
    ExecutionResult Execute(const Query& query, const Args&... args) const;
 
    /// @brief Execute a statement with specified command control settings
    /// at some host of the cluster with args as query parameters.
    ///
    /// It is convinient to keep SQL queries in separate files, see @ref scripts/docs/en/userver/sql_files.md
    /// for more info.
    template <typename... Args>
    ExecutionResult Execute(OptionalCommandControl, const Query& query, const Args&... args) const;
 
    /// @brief Insert data at some host of the cluster;
    /// `T` is expected to be a struct of vectors of same length.
    /// @param table_name table to insert into
    /// @param column_names names of columns of the table
    /// @param data data to insert
    /// See @ref clickhouse_io for better understanding of T's requirements.
    template <typename T>
    void Insert(const std::string& table_name, const std::vector<std::string_view>& column_names, const T& data) const;
 
    /// @brief Insert data with specified command control settings
    /// at some host of the cluster;
    /// `T` is expected to be a struct of vectors of same length.
    /// @param optional_cc optional request QOS overrides
    /// @param table_name table to insert into
    /// @param column_names names of columns of the table
    /// @param data data to insert
    /// See @ref clickhouse_io for better understanding of T's requirements.
    template <typename T>
    void Insert(
        OptionalCommandControl optional_cc,
        const std::string& table_name,
        const std::vector<std::string_view>& column_names,
        const T& data
    ) const;
 
    /// @brief Insert data at some host of the cluster;
    /// `Container` is expected to be an iterable of clickhouse-mapped type.
    /// @param table_name table to insert into
    /// @param column_names names of columns of the table
    /// @param data data to insert
    /// See @ref clickhouse_io for better understanding of
    /// `Container::value_type`'s requirements.
    /// @note This version of insert is less performant than `Insert` (it makes 2
    /// copies of data instead of just 1 copy) due to implementation details, so
    /// consider using less convenient but more performant analogue if performance
    /// is a concern.
    template <typename Container>
    void InsertRows(
        const std::string& table_name,
        const std::vector<std::string_view>& column_names,
        const Container& data
    ) const;
 
    /// @brief Insert data with specified command control settings
    /// at some host of the cluster;
    /// `Container` is expected to be an iterable of clickhouse-mapped type.
    /// @param optional_cc optional request QOS overrides
    /// @param table_name table to insert into
    /// @param column_names names of columns of the table
    /// @param data data to insert
    /// See @ref clickhouse_io for better understanding of
    /// `Container::value_type`'s requirements.
    /// @note This version of insert is less performant than `Insert` (it makes 2
    /// copies of data instead of just 1 copy) due to implementation details, so
    /// consider using less convenient but more performant analogue if performance
    /// is a concern.
    template <typename Container>
    void InsertRows(
        OptionalCommandControl optional_cc,
        const std::string& table_name,
        const std::vector<std::string_view>& column_names,
        const Container& data
    ) const;
 
    /// Write cluster statistics
    void WriteStatistics(USERVER_NAMESPACE::utils::statistics::Writer& writer) const;
 
    /// Exception that is thrown if all specified endpoints are unavailable
    class NoAvailablePoolError : public std::runtime_error {
        using std::runtime_error::runtime_error;
    };
 
private:
    void DoInsert(OptionalCommandControl, const impl::InsertionRequest& request) const;
 
    ExecutionResult DoExecute(OptionalCommandControl, const Query& query) const;
 
    const impl::Pool& GetPool() const;
 
    std::vector<impl::Pool> pools_;
    mutable std::atomic<std::size_t> current_pool_ind_{0};
};
 
template <typename T>
void Cluster::Insert(const std::string& table_name, const std::vector<std::string_view>& column_names, const T& data)
    const {
    Insert(OptionalCommandControl{}, table_name, column_names, data);
}
 
template <typename T>
void Cluster::Insert(
    OptionalCommandControl optional_cc,
    const std::string& table_name,
    const std::vector<std::string_view>& column_names,
    const T& data
) const {
    const auto request = impl::InsertionRequest::Create(table_name, column_names, data);
 
    DoInsert(optional_cc, request);
}
 
template <typename Container>
void Cluster::InsertRows(
    const std::string& table_name,
    const std::vector<std::string_view>& column_names,
    const Container& data
) const {
    InsertRows(OptionalCommandControl{}, table_name, column_names, data);
}
 
template <typename Container>
void Cluster::InsertRows(
    OptionalCommandControl optional_cc,
    const std::string& table_name,
    const std::vector<std::string_view>& column_names,
    const Container& data
) const {
    if (data.empty()) return;
 
    const auto request = impl::InsertionRequest::CreateFromRows(table_name, column_names, data);
 
    DoInsert(optional_cc, request);
}
 
template <typename... Args>
ExecutionResult Cluster::Execute(const Query& query, const Args&... args) const {
    return Execute(OptionalCommandControl{}, query, args...);
}
 
template <typename... Args>
ExecutionResult Cluster::Execute(OptionalCommandControl optional_cc, const Query& query, const Args&... args) const {
    const auto formatted_query = impl::WithArgs(query, args...);
    return DoExecute(optional_cc, formatted_query);
}
 
}  // namespace storages::clickhouse
 
USERVER_NAMESPACE_END