ClickHouse/src/Interpreters/AsynchronousInsertQueue.cpp

#include <Interpreters/AsynchronousInsertQueue.h>

#include <Access/Common/AccessFlags.h>
#include <Access/EnabledQuota.h>
#include <Core/Settings.h>
#include <Formats/FormatFactory.h>
#include <IO/ConcatReadBuffer.h>
#include <IO/LimitReadBuffer.h>
#include <IO/ReadBufferFromMemory.h>
#include <IO/ReadBufferFromString.h>
#include <IO/copyData.h>
#include <Interpreters/AsynchronousInsertLog.h>
#include <Interpreters/Context.h>
#include <Interpreters/InterpreterInsertQuery.h>
#include <Interpreters/ProcessList.h>
#include <Interpreters/executeQuery.h>
#include <Parsers/ASTInsertQuery.h>
#include <Parsers/formatAST.h>
#include <Parsers/queryToString.h>
#include <Processors/Executors/CompletedPipelineExecutor.h>
#include <Processors/Executors/StreamingFormatExecutor.h>
#include <Processors/Sources/SourceFromSingleChunk.h>
#include <Processors/Transforms/AddingDefaultsTransform.h>
#include <Processors/Transforms/getSourceFromASTInsertQuery.h>
#include <QueryPipeline/BlockIO.h>
#include <QueryPipeline/Pipe.h>
#include <QueryPipeline/QueryPipeline.h>
#include <Storages/IStorage.h>
#include <Common/CurrentThread.h>
#include <Common/DateLUT.h>
#include <Common/FieldVisitorHash.h>
#include <Common/SensitiveDataMasker.h>
#include <Common/SipHash.h>
#include <Common/logger_useful.h>

namespace CurrentMetrics
{
    extern const Metric PendingAsyncInsert;
    extern const Metric AsynchronousInsertThreads;
    extern const Metric AsynchronousInsertThreadsActive;
    extern const Metric AsynchronousInsertThreadsScheduled;
    extern const Metric AsynchronousInsertQueueSize;
    extern const Metric AsynchronousInsertQueueBytes;
}

namespace ProfileEvents
{
    extern const Event AsyncInsertQuery;
    extern const Event AsyncInsertBytes;
    extern const Event AsyncInsertRows;
    extern const Event FailedAsyncInsertQuery;
}

namespace DB
{

namespace ErrorCodes
{
    extern const int TIMEOUT_EXCEEDED;
    extern const int UNKNOWN_EXCEPTION;
    extern const int UNKNOWN_FORMAT;
    extern const int BAD_ARGUMENTS;
    extern const int LOGICAL_ERROR;
    extern const int INVALID_SETTING_VALUE;
}

static const NameSet settings_to_skip
{
    /// We don't consider this setting because it is only for deduplication,
    /// which means we can put two inserts with different tokens in the same block safely.
    "insert_deduplication_token",
    "log_comment",
};

AsynchronousInsertQueue::InsertQuery::InsertQuery(
    const ASTPtr & query_,
    const std::optional<UUID> & user_id_,
    const std::vector<UUID> & current_roles_,
    const Settings & settings_,
    DataKind data_kind_)
    : query(query_->clone())
    , query_str(queryToString(query))
    , user_id(user_id_)
    , current_roles(current_roles_)
    , settings(settings_)
    , data_kind(data_kind_)
{
    SipHash siphash;

    siphash.update(data_kind);
    query->updateTreeHash(siphash, /*ignore_aliases=*/ true);

    if (user_id)
    {
        siphash.update(*user_id);
        for (const auto & current_role : current_roles)
            siphash.update(current_role);
    }

    for (const auto & setting : settings.allChanged())
    {
        if (settings_to_skip.contains(setting.getName()))
            continue;

        setting_changes.emplace_back(setting.getName(), setting.getValue());
        siphash.update(setting.getName());
        applyVisitor(FieldVisitorHash(siphash), setting.getValue());
    }

    hash = siphash.get128();
}

AsynchronousInsertQueue::InsertQuery &
AsynchronousInsertQueue::InsertQuery::operator=(const InsertQuery & other)
{
    if (this != &other)
    {
        query = other.query->clone();
        query_str = other.query_str;
        user_id = other.user_id;
        current_roles = other.current_roles;
        settings = other.settings;
        data_kind = other.data_kind;
        hash = other.hash;
        setting_changes = other.setting_changes;
    }

    return *this;
}

bool AsynchronousInsertQueue::InsertQuery::operator==(const InsertQuery & other) const
{
    return toTupleCmp() == other.toTupleCmp();
}

AsynchronousInsertQueue::InsertData::Entry::Entry(
    DataChunk && chunk_,
    String && query_id_,
    const String & async_dedup_token_,
    const String & format_,
    MemoryTracker * user_memory_tracker_)
    : chunk(std::move(chunk_))
    , query_id(std::move(query_id_))
    , async_dedup_token(async_dedup_token_)
    , format(format_)
    , user_memory_tracker(user_memory_tracker_)
    , create_time(std::chrono::system_clock::now())
{
}

void AsynchronousInsertQueue::InsertData::Entry::finish(std::exception_ptr exception_)
{
    if (finished.exchange(true))
        return;

    {
        // To avoid races on counter of user's MemoryTracker we should free memory at this moment.
        // Entries data must be destroyed in context of user who runs async insert.
        // Each entry in the list may correspond to a different user,
        // so we need to switch current thread's MemoryTracker.
        MemoryTrackerSwitcher switcher(user_memory_tracker);
        chunk = {};
    }

    if (exception_)
    {
        promise.set_exception(exception_);
        ProfileEvents::increment(ProfileEvents::FailedAsyncInsertQuery, 1);
    }
    else
    {
        promise.set_value();
    }
}

AsynchronousInsertQueue::QueueShardFlushTimeHistory::TimePoints
AsynchronousInsertQueue::QueueShardFlushTimeHistory::getRecentTimePoints() const
{
    std::shared_lock lock(mutex);
    return time_points;
}

void AsynchronousInsertQueue::QueueShardFlushTimeHistory::updateWithCurrentTime()
{
    std::unique_lock lock(mutex);
    time_points.first = time_points.second;
    time_points.second = std::chrono::steady_clock::now();
}

AsynchronousInsertQueue::AsynchronousInsertQueue(ContextPtr context_, size_t pool_size_, bool flush_on_shutdown_)
    : WithContext(context_)
    , pool_size(pool_size_)
    , flush_on_shutdown(flush_on_shutdown_)
    , queue_shards(pool_size)
    , flush_time_history_per_queue_shard(pool_size)
    , pool(
          CurrentMetrics::AsynchronousInsertThreads,
          CurrentMetrics::AsynchronousInsertThreadsActive,
          CurrentMetrics::AsynchronousInsertThreadsScheduled,
          pool_size)
{
    if (!pool_size)
        throw Exception(ErrorCodes::BAD_ARGUMENTS, "pool_size cannot be zero");

    const auto & settings = getContext()->getSettingsRef();

    for (size_t i = 0; i < pool_size; ++i)
        queue_shards[i].busy_timeout_ms
            = std::min(Milliseconds(settings.async_insert_busy_timeout_min_ms), Milliseconds(settings.async_insert_busy_timeout_max_ms));

    for (size_t i = 0; i < pool_size; ++i)
        dump_by_first_update_threads.emplace_back([this, i] { processBatchDeadlines(i); });
}

AsynchronousInsertQueue::~AsynchronousInsertQueue()
{
    try
    {
        LOG_TRACE(log, "Shutting down the asynchronous insertion queue");
        shutdown = true;

        for (size_t i = 0; i < pool_size; ++i)
        {
            auto & shard = queue_shards[i];

            shard.are_tasks_available.notify_one();
            assert(dump_by_first_update_threads[i].joinable());
            dump_by_first_update_threads[i].join();

            if (flush_on_shutdown)
            {
                for (auto & [_, elem] : shard.queue)
                    scheduleDataProcessingJob(elem.key, std::move(elem.data), getContext(), i);
            }
            else
            {
                for (auto & [_, elem] : shard.queue)
                    for (const auto & entry : elem.data->entries)
                        entry->finish(
                            std::make_exception_ptr(Exception(ErrorCodes::TIMEOUT_EXCEEDED, "Wait for async insert timeout exceeded)")));
            }
        }

        pool.wait();
        LOG_TRACE(log, "Asynchronous insertion queue finished");
    }
    catch (...)
    {
        tryLogCurrentException(log);
        pool.wait();
    }
}

void AsynchronousInsertQueue::scheduleDataProcessingJob(
    const InsertQuery & key, InsertDataPtr data, ContextPtr global_context, size_t shard_num)
{
    /// Intuitively it seems reasonable to process first inserted blocks first.
    /// We add new chunks in the end of entries list, so they are automatically ordered by creation time
    chassert(!data->entries.empty());
    const auto priority = Priority{data->entries.front()->create_time.time_since_epoch().count()};

    /// Wrap 'unique_ptr' with 'shared_ptr' to make this
    /// lambda copyable and allow to save it to the thread pool.
    pool.scheduleOrThrowOnError(
        [this, key, global_context, shard_num, my_data = std::make_shared<InsertDataPtr>(std::move(data))]() mutable
        { processData(key, std::move(*my_data), std::move(global_context), flush_time_history_per_queue_shard[shard_num]); },
        priority);
}

void AsynchronousInsertQueue::preprocessInsertQuery(const ASTPtr & query, const ContextPtr & query_context)
{
    auto & insert_query = query->as<ASTInsertQuery &>();
    insert_query.async_insert_flush = true;

    InterpreterInsertQuery interpreter(query, query_context, query_context->getSettingsRef().insert_allow_materialized_columns);
    auto table = interpreter.getTable(insert_query);
    auto sample_block = interpreter.getSampleBlock(insert_query, table, table->getInMemoryMetadataPtr(), query_context);

    if (!FormatFactory::instance().isInputFormat(insert_query.format))
        throw Exception(ErrorCodes::UNKNOWN_FORMAT, "Unknown input format {}", insert_query.format);

    /// For table functions we check access while executing
    /// InterpreterInsertQuery::getTable() -> ITableFunction::execute().
    if (insert_query.table_id)
        query_context->checkAccess(AccessType::INSERT, insert_query.table_id, sample_block.getNames());
}

AsynchronousInsertQueue::PushResult
AsynchronousInsertQueue::pushQueryWithInlinedData(ASTPtr query, ContextPtr query_context)
{
    query = query->clone();
    preprocessInsertQuery(query, query_context);

    String bytes;
    {
        /// Read at most 'async_insert_max_data_size' bytes of data.
        /// If limit is exceeded we will fallback to synchronous insert
        /// to avoid buffering of huge amount of data in memory.

        auto read_buf = getReadBufferFromASTInsertQuery(query);

        LimitReadBuffer limit_buf(
            *read_buf, query_context->getSettingsRef().async_insert_max_data_size,
            /*throw_exception=*/ false, /*exact_limit=*/ {});

        WriteBufferFromString write_buf(bytes);
        copyData(limit_buf, write_buf);

        if (!read_buf->eof())
        {
            write_buf.finalize();

            /// Concat read buffer with already extracted from insert
            /// query data and with the rest data from insert query.
            std::vector<std::unique_ptr<ReadBuffer>> buffers;
            buffers.emplace_back(std::make_unique<ReadBufferFromOwnString>(bytes));
            buffers.emplace_back(std::move(read_buf));

            return PushResult
            {
                .status = PushResult::TOO_MUCH_DATA,
                .future = {},
                .insert_data_buffer = std::make_unique<ConcatReadBuffer>(std::move(buffers)),
            };
        }
    }

    return pushDataChunk(std::move(query), std::move(bytes), std::move(query_context));
}

AsynchronousInsertQueue::PushResult
AsynchronousInsertQueue::pushQueryWithBlock(ASTPtr query, Block block, ContextPtr query_context)
{
    query = query->clone();
    preprocessInsertQuery(query, query_context);
    return pushDataChunk(std::move(query), std::move(block), std::move(query_context));
}

AsynchronousInsertQueue::PushResult
AsynchronousInsertQueue::pushDataChunk(ASTPtr query, DataChunk chunk, ContextPtr query_context)
{
    const auto & settings = query_context->getSettingsRef();
    validateSettings(settings, log);
    auto & insert_query = query->as<ASTInsertQuery &>();

    auto data_kind = chunk.getDataKind();
    auto entry = std::make_shared<InsertData::Entry>(
        std::move(chunk), query_context->getCurrentQueryId(),
        settings.insert_deduplication_token, insert_query.format,
        CurrentThread::getUserMemoryTracker());

    /// If data is parsed on client we don't care of format which is written
    /// in INSERT query. Replace it to put all such queries into one bucket in queue.
    if (data_kind == DataKind::Preprocessed)
        insert_query.format = "Native";

    InsertQuery key{query, query_context->getUserID(), query_context->getCurrentRoles(), settings, data_kind};
    InsertDataPtr data_to_process;
    std::future<void> insert_future;

    auto shard_num = key.hash % pool_size;
    auto & shard = queue_shards[shard_num];
    const auto flush_time_points = flush_time_history_per_queue_shard[shard_num].getRecentTimePoints();
    {
        std::lock_guard lock(shard.mutex);

        auto [it, inserted] = shard.iterators.try_emplace(key.hash);
        auto now = std::chrono::steady_clock::now();
        auto timeout_ms = getBusyWaitTimeoutMs(settings, shard, flush_time_points, now);
        if (timeout_ms != shard.busy_timeout_ms)
        {
            LOG_TRACE(
                log,
                "Asynchronous timeout {} from {} to {} for queue shard {}.",
                timeout_ms < shard.busy_timeout_ms ? "decreased" : "increased",
                shard.busy_timeout_ms.count(),
                timeout_ms.count(),
                size_t(shard_num));
        }

        if (inserted)
            it->second = shard.queue.emplace(now + timeout_ms, Container{key, std::make_unique<InsertData>(timeout_ms)}).first;

        auto queue_it = it->second;
        auto & data = queue_it->second.data;
        size_t entry_data_size = entry->chunk.byteSize();

        assert(data);
        auto size_in_bytes = data->size_in_bytes;
        data->size_in_bytes += entry_data_size;
        /// We rely on the fact that entries are being added to the list in order of creation time in `scheduleDataProcessingJob()`
        data->entries.emplace_back(entry);
        insert_future = entry->getFuture();

        LOG_TRACE(log, "Have {} pending inserts with total {} bytes of data for query '{}'",
            data->entries.size(), data->size_in_bytes, key.query_str);

        bool has_enough_bytes = data->size_in_bytes >= key.settings.async_insert_max_data_size;
        bool has_enough_queries = data->entries.size() >= key.settings.async_insert_max_query_number && key.settings.async_insert_deduplicate;

        auto max_busy_timeout_exceeded = [&shard, &settings, &now, &flush_time_points]() -> bool
        {
            if (!settings.async_insert_use_adaptive_busy_timeout || !shard.last_insert_time || !flush_time_points.first)
                return false;

            auto max_ms = Milliseconds(settings.async_insert_busy_timeout_max_ms);
            return *shard.last_insert_time + max_ms < now && *flush_time_points.first + max_ms < *flush_time_points.second;
        };

        /// Here we check whether we have hit the limit on the maximum data size in the buffer or
        /// if the elapsed time since the last insert exceeds the maximum busy wait timeout.
        /// We also use the limit settings from the query context.
        /// This works because queries with the same set of settings are already grouped together.
        if (!flush_stopped && (has_enough_bytes || has_enough_queries || max_busy_timeout_exceeded()))
        {
            data->timeout_ms = Milliseconds::zero();
            data_to_process = std::move(data);
            shard.iterators.erase(it);
            shard.queue.erase(queue_it);
        }

        shard.last_insert_time = now;
        shard.busy_timeout_ms = timeout_ms;

        CurrentMetrics::add(CurrentMetrics::PendingAsyncInsert);
        ProfileEvents::increment(ProfileEvents::AsyncInsertQuery);
        ProfileEvents::increment(ProfileEvents::AsyncInsertBytes, entry_data_size);

        if (data_to_process)
        {
            if (!inserted)
                CurrentMetrics::sub(CurrentMetrics::AsynchronousInsertQueueSize);
            CurrentMetrics::sub(CurrentMetrics::AsynchronousInsertQueueBytes, size_in_bytes);
        }
        else
        {
            if (inserted)
                CurrentMetrics::add(CurrentMetrics::AsynchronousInsertQueueSize);
            CurrentMetrics::add(CurrentMetrics::AsynchronousInsertQueueBytes, entry_data_size);
        }
    }

    if (data_to_process)
        scheduleDataProcessingJob(key, std::move(data_to_process), getContext(), shard_num);
    else
        shard.are_tasks_available.notify_one();

    return PushResult
    {
        .status = PushResult::OK,
        .future = std::move(insert_future),
        .insert_data_buffer = nullptr,
    };
}

AsynchronousInsertQueue::Milliseconds AsynchronousInsertQueue::getBusyWaitTimeoutMs(
    const Settings & settings,
    const QueueShard & shard,
    const QueueShardFlushTimeHistory::TimePoints & flush_time_points,
    std::chrono::steady_clock::time_point now) const
{
    if (!settings.async_insert_use_adaptive_busy_timeout)
        return settings.async_insert_busy_timeout_max_ms;

    const auto max_ms = Milliseconds(settings.async_insert_busy_timeout_max_ms);
    const auto min_ms = std::min(std::max(Milliseconds(settings.async_insert_busy_timeout_min_ms), Milliseconds(1)), max_ms);

    auto normalize = [&min_ms, &max_ms](const auto & t_ms) { return std::min(std::max(t_ms, min_ms), max_ms); };

    if (!shard.last_insert_time || !flush_time_points.first)
        return normalize(shard.busy_timeout_ms);

    const auto & last_insert_time = *shard.last_insert_time;
    const auto & [t1, t2] = std::tie(*flush_time_points.first, *flush_time_points.second);
    const double increase_rate = settings.async_insert_busy_timeout_increase_rate;
    const double decrease_rate = settings.async_insert_busy_timeout_decrease_rate;

    const auto decreased_timeout_ms = std::min(
        std::chrono::duration_cast<Milliseconds>(shard.busy_timeout_ms / (1.0 + decrease_rate)), shard.busy_timeout_ms - Milliseconds(1));

    /// Increase the timeout for frequent inserts.
    if (last_insert_time + min_ms > now)
    {
        auto timeout_ms = std::max(
            std::chrono::duration_cast<Milliseconds>(shard.busy_timeout_ms * (1.0 + increase_rate)),
            shard.busy_timeout_ms + Milliseconds(1));

        return normalize(timeout_ms);
    }
    /// Decrease the timeout if inserts are not frequent,
    /// that is, if the time since the last insert and the difference between the last two queue flushes were both
    /// long enough (exceeding the adjusted timeout).
    /// This ensures the timeout value converges to the minimum over time for non-frequent inserts.
    else if (last_insert_time + decreased_timeout_ms < now && t1 + decreased_timeout_ms < t2)
        return normalize(decreased_timeout_ms);

    return normalize(shard.busy_timeout_ms);
}

void AsynchronousInsertQueue::validateSettings(const Settings & settings, LoggerPtr log)
{
    const auto max_ms = std::chrono::milliseconds(settings.async_insert_busy_timeout_max_ms);

    if (max_ms == std::chrono::milliseconds::zero())
        throw Exception(ErrorCodes::INVALID_SETTING_VALUE, "Setting 'async_insert_busy_timeout_max_ms' can't be zero");

    if (!settings.async_insert_use_adaptive_busy_timeout)
        return;

    /// Adaptive timeout settings.
    const auto min_ms = std::chrono::milliseconds(settings.async_insert_busy_timeout_min_ms);

    if (min_ms > max_ms)
        if (log)
            LOG_WARNING(
                log,
                "Setting 'async_insert_busy_timeout_min_ms'={} is greater than 'async_insert_busy_timeout_max_ms'={}. Ignoring "
                "'async_insert_busy_timeout_min_ms'",
                min_ms.count(),
                max_ms.count());

    if (settings.async_insert_busy_timeout_increase_rate <= 0)
        throw Exception(ErrorCodes::INVALID_SETTING_VALUE, "Setting 'async_insert_busy_timeout_increase_rate' must be greater than zero");

    if (settings.async_insert_busy_timeout_decrease_rate <= 0)
        throw Exception(ErrorCodes::INVALID_SETTING_VALUE, "Setting 'async_insert_busy_timeout_decrease_rate' must be greater than zero");
}

void AsynchronousInsertQueue::flushAll()
{
    std::lock_guard flush_lock(flush_mutex);

    LOG_DEBUG(log, "Requested to flush asynchronous insert queue");

    /// Disable background flushes to avoid adding new elements to the queue.
    flush_stopped = true;
    std::vector<Queue> queues_to_flush(pool_size);

    for (size_t i = 0; i < pool_size; ++i)
    {
        std::lock_guard lock(queue_shards[i].mutex);
        queues_to_flush[i] = std::move(queue_shards[i].queue);
        queue_shards[i].iterators.clear();
    }

    size_t total_queries = 0;
    size_t total_bytes = 0;
    size_t total_entries = 0;

    for (size_t i = 0; i < pool_size; ++i)
    {
        auto & queue = queues_to_flush[i];
        total_queries += queue.size();
        for (auto & [_, entry] : queue)
        {
            total_bytes += entry.data->size_in_bytes;
            total_entries += entry.data->entries.size();
            scheduleDataProcessingJob(entry.key, std::move(entry.data), getContext(), i);
        }
    }

    /// Note that jobs scheduled before the call of 'flushAll' are not counted here.
    LOG_DEBUG(log,
        "Will wait for finishing of {} flushing jobs (about {} inserts, {} bytes, {} distinct queries)",
        pool.active(), total_entries, total_bytes, total_queries);

    /// Wait until all jobs are finished. That includes also jobs
    /// that were scheduled before the call of 'flushAll'.
    pool.wait();

    LOG_DEBUG(log, "Finished flushing of asynchronous insert queue");
    flush_stopped = false;
}

void AsynchronousInsertQueue::processBatchDeadlines(size_t shard_num)
{
    auto & shard = queue_shards[shard_num];

    while (!shutdown)
    {
        std::vector<Container> entries_to_flush;
        {
            std::unique_lock lock(shard.mutex);

            const auto rel_time
                = std::min(shard.busy_timeout_ms, Milliseconds(getContext()->getSettingsRef().async_insert_poll_timeout_ms));
            shard.are_tasks_available.wait_for(
                lock,
                rel_time,
                [&shard, this]
                {
                    if (shutdown)
                        return true;

                    if (!shard.queue.empty() && shard.queue.begin()->first < std::chrono::steady_clock::now())
                        return true;

                    return false;
                });

            if (shutdown)
                return;

            if (flush_stopped)
                continue;

            const auto now = std::chrono::steady_clock::now();

            size_t size_in_bytes = 0;
            while (true)
            {
                if (shard.queue.empty() || shard.queue.begin()->first > now)
                    break;

                auto it = shard.queue.begin();
                size_in_bytes += it->second.data->size_in_bytes;

                shard.iterators.erase(it->second.key.hash);

                entries_to_flush.emplace_back(std::move(it->second));
                shard.queue.erase(it);
            }

            if (!entries_to_flush.empty())
            {
                CurrentMetrics::sub(CurrentMetrics::AsynchronousInsertQueueSize, entries_to_flush.size());
                CurrentMetrics::sub(CurrentMetrics::AsynchronousInsertQueueBytes, size_in_bytes);
            }
        }

        for (auto & entry : entries_to_flush)
            scheduleDataProcessingJob(entry.key, std::move(entry.data), getContext(), shard_num);
    }
}

namespace
{

using TimePoint = std::chrono::time_point<std::chrono::system_clock>;

void appendElementsToLogSafe(
    AsynchronousInsertLog & log,
    std::vector<AsynchronousInsertLogElement> elements,
    TimePoint flush_time,
    const String & flush_query_id,
    const String & flush_exception)
try
{
    using Status = AsynchronousInsertLogElement::Status;

    for (auto & elem : elements)
    {
        elem.flush_time = timeInSeconds(flush_time);
        elem.flush_time_microseconds = timeInMicroseconds(flush_time);
        elem.flush_query_id = flush_query_id;
        elem.exception = flush_exception;
        elem.status = flush_exception.empty() ? Status::Ok : Status::FlushError;
        log.add(std::move(elem));
    }
}
catch (...)
{
    tryLogCurrentException("AsynchronousInsertQueue", "Failed to add elements to AsynchronousInsertLog");
}

String serializeQuery(const IAST & query, size_t max_length)
{
    return query.hasSecretParts()
        ? query.formatForLogging(max_length)
        : wipeSensitiveDataAndCutToLength(serializeAST(query), max_length);
}

}

// static
void AsynchronousInsertQueue::processData(
    InsertQuery key, InsertDataPtr data, ContextPtr global_context, QueueShardFlushTimeHistory & queue_shard_flush_time_history)
try
{
    if (!data)
        return;

    SCOPE_EXIT(CurrentMetrics::sub(CurrentMetrics::PendingAsyncInsert, data->entries.size()));

    const auto log = getLogger("AsynchronousInsertQueue");
    const auto & insert_query = assert_cast<const ASTInsertQuery &>(*key.query);

    auto insert_context = Context::createCopy(global_context);
    bool internal = false; // To enable logging this query
    bool async_insert = true;

    /// Disabled query spans. Could be activated by initializing this to a SpanHolder
    std::shared_ptr<OpenTelemetry::SpanHolder> query_span{nullptr};

    /// 'resetParser' doesn't work for parallel parsing.
    key.settings.set("input_format_parallel_parsing", false);
    /// It maybe insert into distributed table.
    /// It doesn't make sense to make insert into destination tables asynchronous.
    key.settings.set("async_insert", false);

    insert_context->makeQueryContext();

    /// Access rights must be checked for the user who executed the initial INSERT query.
    if (key.user_id)
        insert_context->setUser(*key.user_id, key.current_roles);

    insert_context->setSettings(key.settings);

    /// Set initial_query_id, because it's used in InterpreterInsertQuery for table lock.
    insert_context->setCurrentQueryId("");

    auto insert_query_id = insert_context->getCurrentQueryId();
    auto query_start_time = std::chrono::system_clock::now();

    Stopwatch start_watch{CLOCK_MONOTONIC};
    insert_context->setQueryKind(ClientInfo::QueryKind::INITIAL_QUERY);
    insert_context->setInitialQueryStartTime(query_start_time);
    insert_context->setCurrentQueryId(insert_query_id);
    insert_context->setInitialQueryId(insert_query_id);

    DB::CurrentThread::QueryScope query_scope_holder(insert_context);

    auto query_for_logging = serializeQuery(*key.query, insert_context->getSettingsRef().log_queries_cut_to_length);

    /// We add it to the process list so
    /// a) it appears in system.processes
    /// b) can be cancelled if we want to
    /// c) has an associated process list element where runtime metrics are stored
    auto process_list_entry = insert_context->getProcessList().insert(
        query_for_logging,
        key.query.get(),
        insert_context,
        start_watch.getStart());

    auto query_status = process_list_entry->getQueryStatus();
    insert_context->setProcessListElement(std::move(query_status));

    String query_database;
    String query_table;

    if (insert_query.table_id)
    {
        query_database = insert_query.table_id.getDatabaseName();
        query_table = insert_query.table_id.getTableName();
        insert_context->setInsertionTable(insert_query.table_id);
    }

    std::unique_ptr<DB::IInterpreter> interpreter;
    QueryPipeline pipeline;
    QueryLogElement query_log_elem;

    auto async_insert_log = global_context->getAsynchronousInsertLog();
    std::vector<AsynchronousInsertLogElement> log_elements;
    if (async_insert_log)
        log_elements.reserve(data->entries.size());

    try
    {
        interpreter = std::make_unique<InterpreterInsertQuery>(
            key.query, insert_context, key.settings.insert_allow_materialized_columns, false, false, true);

        pipeline = interpreter->execute().pipeline;
        chassert(pipeline.pushing());

        query_log_elem = logQueryStart(
            query_start_time,
            insert_context,
            query_for_logging,
            key.query,
            pipeline,
            interpreter,
            internal,
            query_database,
            query_table,
            async_insert);
    }
    catch (...)
    {
        logExceptionBeforeStart(query_for_logging, insert_context, key.query, query_span, start_watch.elapsedMilliseconds());
        throw;
    }

    auto add_entry_to_log = [&](const auto & entry,
                                const auto & entry_query_for_logging,
                                const auto & exception,
                                size_t num_rows,
                                size_t num_bytes,
                                Milliseconds timeout_ms)
    {
        if (!async_insert_log)
            return;

        AsynchronousInsertLogElement elem;
        elem.event_time = timeInSeconds(entry->create_time);
        elem.event_time_microseconds = timeInMicroseconds(entry->create_time);
        elem.query_for_logging = entry_query_for_logging;
        elem.database = query_database;
        elem.table = query_table;
        elem.format = entry->format;
        elem.query_id = entry->query_id;
        elem.bytes = num_bytes;
        elem.rows = num_rows;
        elem.exception = exception;
        elem.data_kind = entry->chunk.getDataKind();
        elem.timeout_milliseconds = timeout_ms.count();

        /// If there was a parsing error,
        /// the entry won't be flushed anyway,
        /// so add the log element immediately.
        if (!elem.exception.empty())
        {
            elem.status = AsynchronousInsertLogElement::ParsingError;
            async_insert_log->add(std::move(elem));
        }
        else
        {
            log_elements.push_back(elem);
        }
    };

    auto finish_entries = [&]
    {
        for (const auto & entry : data->entries)
        {
            if (!entry->isFinished())
                entry->finish();
        }

        if (!log_elements.empty())
        {
            auto flush_time = std::chrono::system_clock::now();
            appendElementsToLogSafe(*async_insert_log, std::move(log_elements), flush_time, insert_query_id, "");
        }
    };

    Chunk chunk;
    auto header = pipeline.getHeader();

    if (key.data_kind == DataKind::Parsed)
        chunk = processEntriesWithParsing(key, data, header, insert_context, log, add_entry_to_log);
    else
        chunk = processPreprocessedEntries(key, data, header, insert_context, add_entry_to_log);

    ProfileEvents::increment(ProfileEvents::AsyncInsertRows, chunk.getNumRows());

    if (chunk.getNumRows() == 0)
    {
        finish_entries();
        return;
    }

    try
    {
        size_t num_rows = chunk.getNumRows();
        size_t num_bytes = chunk.bytes();

        auto source = std::make_shared<SourceFromSingleChunk>(header, std::move(chunk));
        pipeline.complete(Pipe(std::move(source)));

        CompletedPipelineExecutor completed_executor(pipeline);
        completed_executor.execute();

        LOG_INFO(log, "Flushed {} rows, {} bytes for query '{}'", num_rows, num_bytes, key.query_str);

        queue_shard_flush_time_history.updateWithCurrentTime();

        bool pulling_pipeline = false;
        logQueryFinish(query_log_elem, insert_context, key.query, pipeline, pulling_pipeline, query_span, QueryCache::Usage::None, internal);
    }
    catch (...)
    {
        bool log_error = true;
        logQueryException(query_log_elem, insert_context, start_watch, key.query, query_span, internal, log_error);
        if (!log_elements.empty())
        {
            auto exception = getCurrentExceptionMessage(false);
            auto flush_time = std::chrono::system_clock::now();
            appendElementsToLogSafe(*async_insert_log, std::move(log_elements), flush_time, insert_query_id, exception);
        }
        throw;
    }

    finish_entries();
}
catch (const Exception & e)
{
    finishWithException(key.query, data->entries, e);
}
catch (const Poco::Exception & e)
{
    finishWithException(key.query, data->entries, e);
}
catch (const std::exception & e)
{
    finishWithException(key.query, data->entries, e);
}
catch (...)
{
    finishWithException(key.query, data->entries, Exception(ErrorCodes::UNKNOWN_EXCEPTION, "Unknown exception"));
}

template <typename LogFunc>
Chunk AsynchronousInsertQueue::processEntriesWithParsing(
    const InsertQuery & key,
    const InsertDataPtr & data,
    const Block & header,
    const ContextPtr & insert_context,
    const LoggerPtr logger,
    LogFunc && add_to_async_insert_log)
{
    size_t total_rows = 0;
    InsertData::EntryPtr current_entry;
    String current_exception;

    const auto & insert_query = assert_cast<const ASTInsertQuery &>(*key.query);
    auto format = getInputFormatFromASTInsertQuery(key.query, false, header, insert_context, nullptr);
    std::shared_ptr<ISimpleTransform> adding_defaults_transform;

    if (insert_context->getSettingsRef().input_format_defaults_for_omitted_fields && insert_query.table_id)
    {
        StoragePtr storage = DatabaseCatalog::instance().getTable(insert_query.table_id, insert_context);
        auto metadata_snapshot = storage->getInMemoryMetadataPtr();
        const auto & columns = metadata_snapshot->getColumns();
        if (columns.hasDefaults())
            adding_defaults_transform = std::make_shared<AddingDefaultsTransform>(header, columns, *format, insert_context);
    }

    auto on_error = [&](const MutableColumns & result_columns, Exception & e)
    {
        current_exception = e.displayText();
        LOG_ERROR(logger, "Failed parsing for query '{}' with query id {}. {}",
            key.query_str, current_entry->query_id, current_exception);

        for (const auto & column : result_columns)
            if (column->size() > total_rows)
                column->popBack(column->size() - total_rows);

        current_entry->finish(std::current_exception());
        return 0;
    };

    StreamingFormatExecutor executor(header, format, std::move(on_error), std::move(adding_defaults_transform));
    auto chunk_info = std::make_shared<AsyncInsertInfo>();
    auto query_for_logging = serializeQuery(*key.query, insert_context->getSettingsRef().log_queries_cut_to_length);

    for (const auto & entry : data->entries)
    {
        current_entry = entry;

        const auto * bytes = entry->chunk.asString();
        if (!bytes)
            throw Exception(ErrorCodes::LOGICAL_ERROR,
                "Expected entry with data kind Parsed. Got: {}", entry->chunk.getDataKind());

        auto buffer = std::make_unique<ReadBufferFromString>(*bytes);
        size_t num_bytes = bytes->size();
        size_t num_rows = executor.execute(*buffer);
        total_rows += num_rows;
        chunk_info->offsets.push_back(total_rows);
        chunk_info->tokens.push_back(entry->async_dedup_token);

        add_to_async_insert_log(entry, query_for_logging, current_exception, num_rows, num_bytes, data->timeout_ms);
        current_exception.clear();
    }

    Chunk chunk(executor.getResultColumns(), total_rows);
    chunk.setChunkInfo(std::move(chunk_info));
    return chunk;
}

template <typename LogFunc>
Chunk AsynchronousInsertQueue::processPreprocessedEntries(
    const InsertQuery & key,
    const InsertDataPtr & data,
    const Block & header,
    const ContextPtr & insert_context,
    LogFunc && add_to_async_insert_log)
{
    size_t total_rows = 0;
    auto chunk_info = std::make_shared<AsyncInsertInfo>();
    auto result_columns = header.cloneEmptyColumns();

    std::unordered_map<String, String> format_to_query;

    auto get_query_by_format = [&](const String & format) -> const String &
    {
        auto [it, inserted] = format_to_query.try_emplace(format);
        if (!inserted)
            return it->second;

        auto query = key.query->clone();
        assert_cast<ASTInsertQuery &>(*query).format = format;
        it->second = serializeQuery(*query, insert_context->getSettingsRef().log_queries_cut_to_length);
        return it->second;
    };

    for (const auto & entry : data->entries)
    {
        const auto * block = entry->chunk.asBlock();
        if (!block)
            throw Exception(ErrorCodes::LOGICAL_ERROR,
                "Expected entry with data kind Preprocessed. Got: {}", entry->chunk.getDataKind());

        auto columns = block->getColumns();
        for (size_t i = 0, s = columns.size(); i < s; ++i)
            result_columns[i]->insertRangeFrom(*columns[i], 0, columns[i]->size());

        total_rows += block->rows();
        chunk_info->offsets.push_back(total_rows);
        chunk_info->tokens.push_back(entry->async_dedup_token);

        const auto & query_for_logging = get_query_by_format(entry->format);
        add_to_async_insert_log(entry, query_for_logging, "", block->rows(), block->bytes(), data->timeout_ms);
    }

    Chunk chunk(std::move(result_columns), total_rows);
    chunk.setChunkInfo(std::move(chunk_info));
    return chunk;
}

template <typename E>
void AsynchronousInsertQueue::finishWithException(
    const ASTPtr & query, const std::list<InsertData::EntryPtr> & entries, const E & exception)
{
    tryLogCurrentException("AsynchronousInsertQueue", fmt::format("Failed insertion for query '{}'", queryToString(query)));

    for (const auto & entry : entries)
    {
        if (!entry->isFinished())
        {
            /// Make a copy of exception to avoid concurrent usage of
            /// one exception object from several threads.
            entry->finish(std::make_exception_ptr(exception));
        }
    }
}

}