ClickHouse/dbms/programs/server/TCPHandler.cpp

1255 lines
41 KiB
C++
Raw Normal View History

2012-03-19 12:57:56 +00:00
#include <iomanip>
#include <ext/scope_guard.h>
2013-01-13 22:13:54 +00:00
#include <Poco/Net/NetException.h>
#include <Common/ClickHouseRevision.h>
#include <Common/CurrentThread.h>
#include <Common/Stopwatch.h>
#include <Common/NetException.h>
2018-08-31 00:59:48 +00:00
#include <Common/setThreadName.h>
#include <Common/config_version.h>
#include <IO/Progress.h>
2018-12-28 18:15:26 +00:00
#include <Compression/CompressedReadBuffer.h>
#include <Compression/CompressedWriteBuffer.h>
#include <IO/ReadBufferFromPocoSocket.h>
#include <IO/WriteBufferFromPocoSocket.h>
2018-09-20 20:51:21 +00:00
#include <IO/ReadHelpers.h>
#include <IO/WriteHelpers.h>
#include <IO/copyData.h>
#include <DataStreams/AsynchronousBlockInputStream.h>
#include <DataStreams/NativeBlockInputStream.h>
#include <DataStreams/NativeBlockOutputStream.h>
#include <Interpreters/executeQuery.h>
#include <Interpreters/TablesStatus.h>
#include <Interpreters/InternalTextLogsQueue.h>
#include <Storages/StorageMemory.h>
#include <Storages/StorageReplicatedMergeTree.h>
#include <Core/ExternalTable.h>
#include <Storages/ColumnDefault.h>
#include <DataTypes/DataTypeLowCardinality.h>
2018-12-21 12:17:30 +00:00
#include <Compression/CompressionFactory.h>
2019-07-08 16:11:38 +00:00
#include <common/logger_useful.h>
2019-03-26 18:28:37 +00:00
#include <Processors/Formats/LazyOutputFormat.h>
2012-03-09 15:46:52 +00:00
#include "TCPHandler.h"
2012-03-09 15:46:52 +00:00
namespace DB
{
namespace ErrorCodes
{
2020-02-25 18:02:41 +00:00
extern const int LOGICAL_ERROR;
extern const int ATTEMPT_TO_READ_AFTER_EOF;
extern const int CLIENT_HAS_CONNECTED_TO_WRONG_PORT;
extern const int UNKNOWN_DATABASE;
extern const int UNKNOWN_EXCEPTION;
extern const int UNKNOWN_PACKET_FROM_CLIENT;
extern const int POCO_EXCEPTION;
extern const int SOCKET_TIMEOUT;
extern const int UNEXPECTED_PACKET_FROM_CLIENT;
}
2012-03-09 15:46:52 +00:00
void TCPHandler::runImpl()
{
2018-08-31 00:59:48 +00:00
setThreadName("TCPHandler");
ThreadStatus thread_status;
2018-08-31 00:59:48 +00:00
2017-08-09 11:57:09 +00:00
connection_context = server.context();
2019-07-08 00:51:43 +00:00
connection_context.makeSessionContext();
/// These timeouts can be changed after receiving query.
auto global_receive_timeout = connection_context.getSettingsRef().receive_timeout;
auto global_send_timeout = connection_context.getSettingsRef().send_timeout;
socket().setReceiveTimeout(global_receive_timeout);
socket().setSendTimeout(global_send_timeout);
socket().setNoDelay(true);
in = std::make_shared<ReadBufferFromPocoSocket>(socket());
out = std::make_shared<WriteBufferFromPocoSocket>(socket());
if (in->eof())
{
LOG_WARNING(log, "Client has not sent any data.");
return;
}
/// User will be authenticated here. It will also set settings from user profile into connection_context.
try
{
receiveHello();
}
catch (const Exception & e) /// Typical for an incorrect username, password, or address.
{
if (e.code() == ErrorCodes::CLIENT_HAS_CONNECTED_TO_WRONG_PORT)
{
LOG_DEBUG(log, "Client has connected to wrong port.");
return;
}
if (e.code() == ErrorCodes::ATTEMPT_TO_READ_AFTER_EOF)
{
LOG_WARNING(log, "Client has gone away.");
return;
}
try
{
/// We try to send error information to the client.
2018-08-24 07:30:53 +00:00
sendException(e, connection_context.getSettingsRef().calculate_text_stack_trace);
}
catch (...) {}
throw;
}
/// When connecting, the default database can be specified.
if (!default_database.empty())
{
if (!DatabaseCatalog::instance().isDatabaseExist(default_database))
{
Exception e("Database " + backQuote(default_database) + " doesn't exist", ErrorCodes::UNKNOWN_DATABASE);
LOG_ERROR(log, "Code: " << e.code() << ", e.displayText() = " << e.displayText()
<< ", Stack trace:\n\n" << e.getStackTraceString());
2018-08-24 07:30:53 +00:00
sendException(e, connection_context.getSettingsRef().calculate_text_stack_trace);
return;
}
connection_context.setCurrentDatabase(default_database);
}
Settings connection_settings = connection_context.getSettings();
sendHello();
connection_context.setProgressCallback([this] (const Progress & value) { return this->updateProgress(value); });
2020-03-08 21:29:00 +00:00
while (true)
{
2019-07-04 22:57:26 +00:00
/// We are waiting for a packet from the client. Thus, every `poll_interval` seconds check whether we need to shut down.
2019-07-04 22:23:45 +00:00
{
Stopwatch idle_time;
2019-07-30 22:27:14 +00:00
while (!server.isCancelled() && !static_cast<ReadBufferFromPocoSocket &>(*in).poll(
std::min(connection_settings.poll_interval, connection_settings.idle_connection_timeout) * 1000000))
2019-07-04 22:23:45 +00:00
{
if (idle_time.elapsedSeconds() > connection_settings.idle_connection_timeout)
2019-07-04 23:03:20 +00:00
{
LOG_TRACE(log, "Closing idle connection");
2019-07-04 22:23:45 +00:00
return;
2019-07-04 23:03:20 +00:00
}
2019-07-04 22:23:45 +00:00
}
}
/// If we need to shut down, or client disconnects.
2017-08-09 14:33:07 +00:00
if (server.isCancelled() || in->eof())
break;
2019-07-04 23:03:20 +00:00
/// Set context of request.
query_context = connection_context;
Stopwatch watch;
state.reset();
/// Initialized later.
std::optional<CurrentThread::QueryScope> query_scope;
/** An exception during the execution of request (it must be sent over the network to the client).
* The client will be able to accept it, if it did not happen while sending another packet and the client has not disconnected yet.
*/
std::optional<DB::Exception> exception;
bool network_error = false;
bool send_exception_with_stack_trace = true;
2018-08-24 07:30:53 +00:00
try
{
/// If a user passed query-local timeouts, reset socket to initial state at the end of the query
SCOPE_EXIT({state.timeout_setter.reset();});
/** If Query - process it. If Ping or Cancel - go back to the beginning.
* There may come settings for a separate query that modify `query_context`.
*/
if (!receivePacket())
continue;
query_scope.emplace(*query_context);
send_exception_with_stack_trace = query_context->getSettingsRef().calculate_text_stack_trace;
2018-08-24 07:30:53 +00:00
/// Should we send internal logs to client?
2019-07-10 12:19:17 +00:00
const auto client_logs_level = query_context->getSettingsRef().send_logs_level;
if (client_revision >= DBMS_MIN_REVISION_WITH_SERVER_LOGS
2019-08-09 13:02:01 +00:00
&& client_logs_level != LogsLevel::none)
{
state.logs_queue = std::make_shared<InternalTextLogsQueue>();
2019-07-10 12:19:17 +00:00
state.logs_queue->max_priority = Poco::Logger::parseLevel(client_logs_level.toString());
2019-08-09 13:02:01 +00:00
CurrentThread::attachInternalTextLogsQueue(state.logs_queue, client_logs_level);
}
query_context->setExternalTablesInitializer([&connection_settings, this] (Context & context)
2018-08-24 00:07:25 +00:00
{
if (&context != &*query_context)
throw Exception("Unexpected context in external tables initializer", ErrorCodes::LOGICAL_ERROR);
/// Get blocks of temporary tables
readData(connection_settings);
/// Reset the input stream, as we received an empty block while receiving external table data.
/// So, the stream has been marked as cancelled and we can't read from it anymore.
state.block_in.reset();
2018-11-26 00:56:50 +00:00
state.maybe_compressed_in.reset(); /// For more accurate accounting by MemoryTracker.
state.temporary_tables_read = true;
});
2019-05-28 18:30:10 +00:00
/// Send structure of columns to client for function input()
query_context->setInputInitializer([this] (Context & context, const StoragePtr & input_storage)
{
if (&context != &query_context.value())
2019-05-28 18:30:10 +00:00
throw Exception("Unexpected context in Input initializer", ErrorCodes::LOGICAL_ERROR);
state.need_receive_data_for_input = true;
/// Send ColumnsDescription for input storage.
if (client_revision >= DBMS_MIN_REVISION_WITH_COLUMN_DEFAULTS_METADATA
&& query_context->getSettingsRef().input_format_defaults_for_omitted_fields)
{
sendTableColumns(input_storage->getColumns());
}
/// Send block to the client - input storage structure.
state.input_header = input_storage->getSampleBlock();
sendData(state.input_header);
});
query_context->setInputBlocksReaderCallback([&connection_settings, this] (Context & context) -> Block
2019-05-28 18:30:10 +00:00
{
if (&context != &query_context.value())
2019-05-28 18:30:10 +00:00
throw Exception("Unexpected context in InputBlocksReader", ErrorCodes::LOGICAL_ERROR);
size_t poll_interval;
int receive_timeout;
std::tie(poll_interval, receive_timeout) = getReadTimeouts(connection_settings);
2019-05-28 18:30:10 +00:00
if (!readDataNext(poll_interval, receive_timeout))
{
state.block_in.reset();
state.maybe_compressed_in.reset();
return Block();
2019-05-28 20:16:24 +00:00
}
2019-05-28 18:30:10 +00:00
return state.block_for_input;
});
customizeContext(*query_context);
bool may_have_embedded_data = client_revision >= DBMS_MIN_REVISION_WITH_CLIENT_SUPPORT_EMBEDDED_DATA;
/// Processing Query
state.io = executeQuery(state.query, *query_context, false, state.stage, may_have_embedded_data);
if (state.io.out)
state.need_receive_data_for_insert = true;
after_check_cancelled.restart();
after_send_progress.restart();
/// Does the request require receive data from client?
if (state.need_receive_data_for_insert)
processInsertQuery(connection_settings);
2019-05-28 18:30:10 +00:00
else if (state.need_receive_data_for_input)
{
/// It is special case for input(), all works for reading data from client will be done in callbacks.
/// state.io.in is NullAndDoCopyBlockInputStream so read it once.
state.io.in->read();
state.io.onFinish();
}
2019-03-26 18:28:37 +00:00
else if (state.io.pipeline.initialized())
processOrdinaryQueryWithProcessors();
else
processOrdinaryQuery();
/// Do it before sending end of stream, to have a chance to show log message in client.
query_scope->logPeakMemoryUsage();
sendLogs();
sendEndOfStream();
query_scope.reset();
state.reset();
}
catch (const Exception & e)
{
state.io.onException();
exception.emplace(e);
if (e.code() == ErrorCodes::UNKNOWN_PACKET_FROM_CLIENT)
throw;
/// If a timeout occurred, try to inform client about it and close the session
if (e.code() == ErrorCodes::SOCKET_TIMEOUT)
network_error = true;
}
catch (const Poco::Net::NetException & e)
{
/** We can get here if there was an error during connection to the client,
* or in connection with a remote server that was used to process the request.
* It is not possible to distinguish between these two cases.
* Although in one of them, we have to send exception to the client, but in the other - we can not.
* We will try to send exception to the client in any case - see below.
*/
state.io.onException();
exception.emplace(Exception::CreateFromPoco, e);
}
catch (const Poco::Exception & e)
{
state.io.onException();
exception.emplace(Exception::CreateFromPoco, e);
}
catch (const std::exception & e)
{
state.io.onException();
exception.emplace(Exception::CreateFromSTD, e);
}
catch (...)
{
state.io.onException();
exception.emplace("Unknown exception", ErrorCodes::UNKNOWN_EXCEPTION);
}
try
{
if (exception)
{
try
{
/// Try to send logs to client, but it could be risky too
/// Assume that we can't break output here
sendLogs();
}
catch (...)
{
tryLogCurrentException(log, "Can't send logs to client");
}
2018-08-24 07:30:53 +00:00
sendException(*exception, send_exception_with_stack_trace);
}
}
catch (...)
{
/** Could not send exception information to the client. */
network_error = true;
LOG_WARNING(log, "Client has gone away.");
}
try
{
if (exception && !state.temporary_tables_read)
query_context->initializeExternalTablesIfSet();
}
catch (...)
{
network_error = true;
LOG_WARNING(log, "Can't read external tables after query failure.");
}
try
{
query_scope.reset();
state.reset();
}
catch (...)
{
/** During the processing of request, there was an exception that we caught and possibly sent to client.
* When destroying the request pipeline execution there was a second exception.
* For example, a pipeline could run in multiple threads, and an exception could occur in each of them.
* Ignore it.
*/
}
watch.stop();
LOG_INFO(log, std::fixed << std::setprecision(3)
<< "Processed in " << watch.elapsedSeconds() << " sec.");
/// It is important to destroy query context here. We do not want it to live arbitrarily longer than the query.
query_context.reset();
if (network_error)
break;
}
2012-03-11 08:52:56 +00:00
}
2019-05-28 18:30:10 +00:00
bool TCPHandler::readDataNext(const size_t & poll_interval, const int & receive_timeout)
{
Stopwatch watch(CLOCK_MONOTONIC_COARSE);
/// We are waiting for a packet from the client. Thus, every `POLL_INTERVAL` seconds check whether we need to shut down.
while (true)
{
if (static_cast<ReadBufferFromPocoSocket &>(*in).poll(poll_interval))
break;
/// Do we need to shut down?
if (server.isCancelled())
return false;
/** Have we waited for data for too long?
* If we periodically poll, the receive_timeout of the socket itself does not work.
* Therefore, an additional check is added.
*/
double elapsed = watch.elapsedSeconds();
if (elapsed > receive_timeout)
{
std::stringstream ss;
ss << "Timeout exceeded while receiving data from client.";
ss << " Waited for " << static_cast<size_t>(elapsed) << " seconds,";
ss << " timeout is " << receive_timeout << " seconds.";
throw Exception(ss.str(), ErrorCodes::SOCKET_TIMEOUT);
}
}
/// If client disconnected.
if (in->eof())
return false;
/// We accept and process data. And if they are over, then we leave.
if (!receivePacket())
return false;
sendLogs();
return true;
}
std::tuple<size_t, int> TCPHandler::getReadTimeouts(const Settings & connection_settings)
2012-05-21 06:49:05 +00:00
{
const auto receive_timeout = query_context->getSettingsRef().receive_timeout.value;
/// Poll interval should not be greater than receive_timeout
2019-08-09 13:02:01 +00:00
const size_t default_poll_interval = connection_settings.poll_interval * 1000000;
size_t current_poll_interval = static_cast<size_t>(receive_timeout.totalMicroseconds());
constexpr size_t min_poll_interval = 5000; // 5 ms
size_t poll_interval = std::max(min_poll_interval, std::min(default_poll_interval, current_poll_interval));
2019-05-28 18:30:10 +00:00
return std::make_tuple(poll_interval, receive_timeout.totalSeconds());
}
void TCPHandler::readData(const Settings & connection_settings)
2019-05-28 18:30:10 +00:00
{
size_t poll_interval;
int receive_timeout;
std::tie(poll_interval, receive_timeout) = getReadTimeouts(connection_settings);
2019-05-28 18:30:10 +00:00
sendLogs();
2019-05-28 18:30:10 +00:00
while (true)
if (!readDataNext(poll_interval, receive_timeout))
return;
}
void TCPHandler::processInsertQuery(const Settings & connection_settings)
{
/** Made above the rest of the lines, so that in case of `writePrefix` function throws an exception,
* client receive exception before sending data.
*/
state.io.out->writePrefix();
/// Send ColumnsDescription for insertion table
if (client_revision >= DBMS_MIN_REVISION_WITH_COLUMN_DEFAULTS_METADATA)
{
2020-03-02 20:23:58 +00:00
const auto & table_id = query_context->getInsertionTable();
if (query_context->getSettingsRef().input_format_defaults_for_omitted_fields)
{
2020-03-02 20:23:58 +00:00
if (!table_id.empty())
sendTableColumns(DatabaseCatalog::instance().getTable(table_id)->getColumns());
}
}
/// Send block to the client - table structure.
2018-12-20 14:26:54 +00:00
sendData(state.io.out->getHeader());
readData(connection_settings);
state.io.out->writeSuffix();
state.io.onFinish();
2012-05-21 06:49:05 +00:00
}
void TCPHandler::processOrdinaryQuery()
{
/// Pull query execution result, if exists, and send it to network.
if (state.io.in)
{
/// This allows the client to prepare output format
if (Block header = state.io.in->getHeader())
sendData(header);
/// Use of async mode here enables reporting progress and monitoring client cancelling the query
AsynchronousBlockInputStream async_in(state.io.in);
async_in.readPrefix();
while (true)
{
if (isQueryCancelled())
{
async_in.cancel(false);
break;
}
if (after_send_progress.elapsed() / 1000 >= query_context->getSettingsRef().interactive_delay)
{
2020-03-04 15:16:32 +00:00
/// Some time passed.
after_send_progress.restart();
sendProgress();
}
sendLogs();
if (async_in.poll(query_context->getSettingsRef().interactive_delay / 1000))
{
const auto block = async_in.read();
if (!block)
break;
if (!state.io.null_format)
sendData(block);
}
}
async_in.readSuffix();
2019-12-11 14:54:19 +00:00
/** When the data has run out, we send the profiling data and totals up to the terminating empty block,
* so that this information can be used in the suffix output of stream.
* If the request has been interrupted, then sendTotals and other methods should not be called,
* because we have not read all the data.
*/
if (!isQueryCancelled())
{
sendTotals(state.io.in->getTotals());
sendExtremes(state.io.in->getExtremes());
sendProfileInfo(state.io.in->getProfileInfo());
sendProgress();
}
2019-12-11 14:54:19 +00:00
sendData({});
}
state.io.onFinish();
2020-03-03 19:01:20 +00:00
sendProgress();
2012-07-21 07:02:55 +00:00
}
void TCPHandler::processOrdinaryQueryWithProcessors()
2019-03-26 18:28:37 +00:00
{
auto & pipeline = state.io.pipeline;
/// Send header-block, to allow client to prepare output format for data to send.
{
auto & header = pipeline.getHeader();
if (header)
sendData(header);
}
auto lazy_format = std::make_shared<LazyOutputFormat>(pipeline.getHeader());
pipeline.setOutput(lazy_format);
{
2019-05-14 12:53:20 +00:00
auto thread_group = CurrentThread::getGroup();
2019-07-08 11:55:25 +00:00
ThreadPool pool(1);
2019-05-14 12:53:20 +00:00
auto executor = pipeline.execute();
std::atomic_bool exception = false;
pool.scheduleOrThrowOnError([&]()
2019-05-16 14:57:27 +00:00
{
2019-07-08 11:55:25 +00:00
/// ThreadStatus thread_status;
if (thread_group)
CurrentThread::attachTo(thread_group);
SCOPE_EXIT(
if (thread_group)
CurrentThread::detachQueryIfNotDetached();
);
2019-05-16 14:57:27 +00:00
CurrentMetrics::Increment query_thread_metric_increment{CurrentMetrics::QueryThread};
setThreadName("QueryPipelineEx");
try
{
executor->execute(pipeline.getNumThreads());
2019-05-16 14:57:27 +00:00
}
catch (...)
{
exception = true;
throw;
}
});
2019-05-14 12:53:20 +00:00
2020-01-23 10:04:18 +00:00
/// Wait in case of exception happened outside of pool.
2019-04-18 18:40:55 +00:00
SCOPE_EXIT(
lazy_format->finish();
2019-09-09 14:57:12 +00:00
try
{
pool.wait();
}
catch (...)
{
/// If exception was thrown during pipeline execution, skip it while processing other exception.
2020-01-23 10:04:18 +00:00
tryLogCurrentException(log);
2019-09-09 14:57:12 +00:00
}
2019-04-18 18:40:55 +00:00
);
2020-01-23 10:04:18 +00:00
while (!lazy_format->isFinished() && !exception)
2019-03-26 18:28:37 +00:00
{
2020-01-23 10:04:18 +00:00
if (isQueryCancelled())
2019-03-26 18:28:37 +00:00
{
2020-01-23 10:04:18 +00:00
/// A packet was received requesting to stop execution of the request.
executor->cancel();
break;
}
2019-03-26 18:28:37 +00:00
2020-01-23 10:04:18 +00:00
if (after_send_progress.elapsed() / 1000 >= query_context->getSettingsRef().interactive_delay)
{
/// Some time passed and there is a progress.
after_send_progress.restart();
sendProgress();
}
2019-03-26 18:28:37 +00:00
2020-01-23 10:04:18 +00:00
sendLogs();
2019-04-05 10:52:07 +00:00
2020-01-23 10:04:18 +00:00
if (auto block = lazy_format->getBlock(query_context->getSettingsRef().interactive_delay / 1000))
{
if (!state.io.null_format)
sendData(block);
2019-03-26 18:28:37 +00:00
}
2020-01-23 10:04:18 +00:00
}
2019-03-26 18:28:37 +00:00
2020-01-23 10:04:18 +00:00
/// Finish lazy_format before waiting. Otherwise some thread may write into it, and waiting will lock.
lazy_format->finish();
pool.wait();
2019-04-08 14:55:20 +00:00
2020-01-23 10:04:18 +00:00
/** If data has run out, we will send the profiling data and total values to
* the last zero block to be able to use
* this information in the suffix output of stream.
* If the request was interrupted, then `sendTotals` and other methods could not be called,
* because we have not read all the data yet,
* and there could be ongoing calculations in other threads at the same time.
*/
if (!isQueryCancelled())
{
sendTotals(lazy_format->getTotals());
sendExtremes(lazy_format->getExtremes());
sendProfileInfo(lazy_format->getProfileInfo());
sendProgress();
sendLogs();
2019-05-14 12:53:20 +00:00
}
2020-01-23 10:04:18 +00:00
sendData({});
2019-03-26 18:28:37 +00:00
}
state.io.onFinish();
2020-03-03 19:01:20 +00:00
sendProgress();
2019-03-26 18:28:37 +00:00
}
2012-07-21 07:02:55 +00:00
void TCPHandler::processTablesStatusRequest()
{
TablesStatusRequest request;
request.read(*in, client_revision);
TablesStatusResponse response;
for (const QualifiedTableName & table_name: request.tables)
{
2020-03-04 20:29:52 +00:00
auto resolved_id = connection_context.tryResolveStorageID({table_name.database, table_name.table});
StoragePtr table = DatabaseCatalog::instance().tryGetTable(resolved_id);
if (!table)
continue;
TableStatus status;
if (auto * replicated_table = dynamic_cast<StorageReplicatedMergeTree *>(table.get()))
{
status.is_replicated = true;
status.absolute_delay = replicated_table->getAbsoluteDelay();
}
else
status.is_replicated = false;
response.table_states_by_id.emplace(table_name, std::move(status));
}
writeVarUInt(Protocol::Server::TablesStatusResponse, *out);
response.write(*out, client_revision);
}
2019-09-03 13:55:26 +00:00
void TCPHandler::receiveUnexpectedTablesStatusRequest()
{
TablesStatusRequest skip_request;
skip_request.read(*in, client_revision);
throw NetException("Unexpected packet TablesStatusRequest received from client", ErrorCodes::UNEXPECTED_PACKET_FROM_CLIENT);
}
2019-03-26 18:28:37 +00:00
void TCPHandler::sendProfileInfo(const BlockStreamProfileInfo & info)
2013-05-22 14:57:43 +00:00
{
writeVarUInt(Protocol::Server::ProfileInfo, *out);
2019-03-26 18:28:37 +00:00
info.write(*out);
out->next();
2013-05-22 14:57:43 +00:00
}
2019-03-26 18:28:37 +00:00
void TCPHandler::sendTotals(const Block & totals)
{
if (totals)
{
initBlockOutput(totals);
writeVarUInt(Protocol::Server::Totals, *out);
writeStringBinary("", *out);
state.block_out->write(totals);
state.maybe_compressed_out->next();
out->next();
}
}
2019-03-26 18:28:37 +00:00
void TCPHandler::sendExtremes(const Block & extremes)
{
if (extremes)
{
initBlockOutput(extremes);
writeVarUInt(Protocol::Server::Extremes, *out);
writeStringBinary("", *out);
state.block_out->write(extremes);
state.maybe_compressed_out->next();
out->next();
}
}
2012-05-21 06:49:05 +00:00
void TCPHandler::receiveHello()
2012-05-16 18:03:00 +00:00
{
/// Receive `hello` packet.
UInt64 packet_type = 0;
String user = "default";
String password;
readVarUInt(packet_type, *in);
if (packet_type != Protocol::Client::Hello)
{
/** If you accidentally accessed the HTTP protocol for a port destined for an internal TCP protocol,
* Then instead of the packet type, there will be G (GET) or P (POST), in most cases.
*/
if (packet_type == 'G' || packet_type == 'P')
{
writeString("HTTP/1.0 400 Bad Request\r\n\r\n"
"Port " + server.config().getString("tcp_port") + " is for clickhouse-client program.\r\n"
"You must use port " + server.config().getString("http_port") + " for HTTP.\r\n",
*out);
throw Exception("Client has connected to wrong port", ErrorCodes::CLIENT_HAS_CONNECTED_TO_WRONG_PORT);
}
else
throw NetException("Unexpected packet from client", ErrorCodes::UNEXPECTED_PACKET_FROM_CLIENT);
}
readStringBinary(client_name, *in);
readVarUInt(client_version_major, *in);
readVarUInt(client_version_minor, *in);
// NOTE For backward compatibility of the protocol, client cannot send its version_patch.
readVarUInt(client_revision, *in);
readStringBinary(default_database, *in);
readStringBinary(user, *in);
readStringBinary(password, *in);
LOG_DEBUG(log, "Connected " << client_name
<< " version " << client_version_major
<< "." << client_version_minor
<< "." << client_version_patch
<< ", revision: " << client_revision
<< (!default_database.empty() ? ", database: " + default_database : "")
<< (!user.empty() ? ", user: " + user : "")
<< ".");
connection_context.setUser(user, password, socket().peerAddress(), "");
2012-05-16 18:03:00 +00:00
}
void TCPHandler::receiveUnexpectedHello()
{
UInt64 skip_uint_64;
String skip_string;
readStringBinary(skip_string, *in);
readVarUInt(skip_uint_64, *in);
readVarUInt(skip_uint_64, *in);
readVarUInt(skip_uint_64, *in);
readStringBinary(skip_string, *in);
readStringBinary(skip_string, *in);
readStringBinary(skip_string, *in);
throw NetException("Unexpected packet Hello received from client", ErrorCodes::UNEXPECTED_PACKET_FROM_CLIENT);
}
2012-05-21 06:49:05 +00:00
void TCPHandler::sendHello()
2012-03-11 08:52:56 +00:00
{
writeVarUInt(Protocol::Server::Hello, *out);
writeStringBinary(DBMS_NAME, *out);
writeVarUInt(DBMS_VERSION_MAJOR, *out);
writeVarUInt(DBMS_VERSION_MINOR, *out);
writeVarUInt(ClickHouseRevision::get(), *out);
if (client_revision >= DBMS_MIN_REVISION_WITH_SERVER_TIMEZONE)
writeStringBinary(DateLUT::instance().getTimeZone(), *out);
2018-02-26 06:49:17 +00:00
if (client_revision >= DBMS_MIN_REVISION_WITH_SERVER_DISPLAY_NAME)
writeStringBinary(server_display_name, *out);
if (client_revision >= DBMS_MIN_REVISION_WITH_VERSION_PATCH)
writeVarUInt(DBMS_VERSION_PATCH, *out);
out->next();
2012-03-11 08:52:56 +00:00
}
2012-05-17 19:15:53 +00:00
2012-05-21 06:49:05 +00:00
bool TCPHandler::receivePacket()
2012-03-11 08:52:56 +00:00
{
UInt64 packet_type = 0;
readVarUInt(packet_type, *in);
// std::cerr << "Server got packet: " << Protocol::Client::toString(packet_type) << "\n";
switch (packet_type)
{
case Protocol::Client::Query:
if (!state.empty())
receiveUnexpectedQuery();
receiveQuery();
return true;
case Protocol::Client::Data:
2019-10-19 20:36:35 +00:00
case Protocol::Client::Scalar:
if (state.empty())
receiveUnexpectedData();
2019-10-19 20:36:35 +00:00
return receiveData(packet_type == Protocol::Client::Scalar);
case Protocol::Client::Ping:
writeVarUInt(Protocol::Server::Pong, *out);
out->next();
return false;
case Protocol::Client::Cancel:
return false;
case Protocol::Client::Hello:
receiveUnexpectedHello();
case Protocol::Client::TablesStatusRequest:
if (!state.empty())
2019-09-03 13:55:26 +00:00
receiveUnexpectedTablesStatusRequest();
processTablesStatusRequest();
out->next();
return false;
default:
throw Exception("Unknown packet " + toString(packet_type) + " from client", ErrorCodes::UNKNOWN_PACKET_FROM_CLIENT);
}
2012-03-09 15:46:52 +00:00
}
2012-05-17 19:15:53 +00:00
2012-05-21 06:49:05 +00:00
void TCPHandler::receiveQuery()
2012-03-11 08:52:56 +00:00
{
UInt64 stage = 0;
UInt64 compression = 0;
state.is_empty = false;
readStringBinary(state.query_id, *in);
query_context->setCurrentQueryId(state.query_id);
/// Client info
ClientInfo & client_info = query_context->getClientInfo();
if (client_revision >= DBMS_MIN_REVISION_WITH_CLIENT_INFO)
client_info.read(*in, client_revision);
/// For better support of old clients, that does not send ClientInfo.
if (client_info.query_kind == ClientInfo::QueryKind::NO_QUERY)
{
client_info.query_kind = ClientInfo::QueryKind::INITIAL_QUERY;
client_info.client_name = client_name;
client_info.client_version_major = client_version_major;
client_info.client_version_minor = client_version_minor;
client_info.client_version_patch = client_version_patch;
client_info.client_revision = client_revision;
}
/// Set fields, that are known apriori.
client_info.interface = ClientInfo::Interface::TCP;
if (client_info.query_kind == ClientInfo::QueryKind::INITIAL_QUERY)
{
/// 'Current' fields was set at receiveHello.
client_info.initial_user = client_info.current_user;
client_info.initial_query_id = client_info.current_query_id;
client_info.initial_address = client_info.current_address;
}
else
{
query_context->setInitialRowPolicy();
}
/// Per query settings are also passed via TCP.
/// We need to check them before applying due to they can violate the settings constraints.
2019-11-16 23:12:35 +00:00
auto settings_format = (client_revision >= DBMS_MIN_REVISION_WITH_SETTINGS_SERIALIZED_AS_STRINGS) ? SettingsBinaryFormat::STRINGS
: SettingsBinaryFormat::OLD;
Settings passed_settings;
passed_settings.deserialize(*in, settings_format);
auto settings_changes = passed_settings.changes();
if (client_info.query_kind == ClientInfo::QueryKind::INITIAL_QUERY)
{
/// Throw an exception if the passed settings violate the constraints.
query_context->checkSettingsConstraints(settings_changes);
}
else
{
/// Quietly clamp to the constraints if it's not an initial query.
query_context->clampToSettingsConstraints(settings_changes);
}
query_context->applySettingsChanges(settings_changes);
const Settings & settings = query_context->getSettingsRef();
/// Sync timeouts on client and server during current query to avoid dangling queries on server
/// NOTE: We use settings.send_timeout for the receive timeout and vice versa (change arguments ordering in TimeoutSetter),
/// because settings.send_timeout is client-side setting which has opposite meaning on the server side.
/// NOTE: these settings are applied only for current connection (not for distributed tables' connections)
state.timeout_setter = std::make_unique<TimeoutSetter>(socket(), settings.receive_timeout, settings.send_timeout);
readVarUInt(stage, *in);
state.stage = QueryProcessingStage::Enum(stage);
readVarUInt(compression, *in);
2017-10-03 14:52:08 +00:00
state.compression = static_cast<Protocol::Compression>(compression);
readStringBinary(state.query, *in);
2012-03-11 08:52:56 +00:00
}
void TCPHandler::receiveUnexpectedQuery()
{
UInt64 skip_uint_64;
String skip_string;
readStringBinary(skip_string, *in);
ClientInfo skip_client_info;
if (client_revision >= DBMS_MIN_REVISION_WITH_CLIENT_INFO)
skip_client_info.read(*in, client_revision);
Settings skip_settings;
2019-11-16 23:12:35 +00:00
auto settings_format = (client_revision >= DBMS_MIN_REVISION_WITH_SETTINGS_SERIALIZED_AS_STRINGS) ? SettingsBinaryFormat::STRINGS
: SettingsBinaryFormat::OLD;
skip_settings.deserialize(*in, settings_format);
readVarUInt(skip_uint_64, *in);
readVarUInt(skip_uint_64, *in);
readStringBinary(skip_string, *in);
throw NetException("Unexpected packet Query received from client", ErrorCodes::UNEXPECTED_PACKET_FROM_CLIENT);
}
2012-05-21 06:49:05 +00:00
2019-10-19 20:36:35 +00:00
bool TCPHandler::receiveData(bool scalar)
{
initBlockInput();
/// The name of the temporary table for writing data, default to empty string
2020-02-12 18:14:12 +00:00
auto temporary_id = StorageID::createEmpty();
readStringBinary(temporary_id.table_name, *in);
/// Read one block from the network and write it down
Block block = state.block_in->read();
if (block)
{
2019-10-19 20:36:35 +00:00
if (scalar)
2020-02-12 18:14:12 +00:00
query_context->addScalar(temporary_id.table_name, block);
2019-10-19 20:36:35 +00:00
else
{
2019-10-19 20:36:35 +00:00
/// If there is an insert request, then the data should be written directly to `state.io.out`.
/// Otherwise, we write the blocks in the temporary `external_table_name` table.
if (!state.need_receive_data_for_insert && !state.need_receive_data_for_input)
{
2020-02-12 18:14:12 +00:00
auto resolved = query_context->tryResolveStorageID(temporary_id, Context::ResolveExternal);
2019-10-19 20:36:35 +00:00
StoragePtr storage;
/// If such a table does not exist, create it.
2020-03-10 19:36:17 +00:00
if (resolved)
storage = DatabaseCatalog::instance().getTable(resolved);
else
2019-10-19 20:36:35 +00:00
{
NamesAndTypesList columns = block.getNamesAndTypesList();
2020-03-10 19:36:17 +00:00
auto temporary_table = TemporaryTableHolder(*query_context, ColumnsDescription{columns});
storage = temporary_table.getTable();
query_context->addExternalTable(temporary_id.table_name, std::move(temporary_table));
}
2019-10-19 20:36:35 +00:00
/// The data will be written directly to the table.
state.io.out = storage->write(ASTPtr(), *query_context);
}
2019-10-19 20:36:35 +00:00
if (state.need_receive_data_for_input)
state.block_for_input = block;
else
state.io.out->write(block);
}
return true;
}
else
return false;
}
void TCPHandler::receiveUnexpectedData()
{
String skip_external_table_name;
readStringBinary(skip_external_table_name, *in);
std::shared_ptr<ReadBuffer> maybe_compressed_in;
if (last_block_in.compression == Protocol::Compression::Enable)
maybe_compressed_in = std::make_shared<CompressedReadBuffer>(*in);
else
maybe_compressed_in = in;
auto skip_block_in = std::make_shared<NativeBlockInputStream>(
*maybe_compressed_in,
last_block_in.header,
client_revision);
skip_block_in->read();
throw NetException("Unexpected packet Data received from client", ErrorCodes::UNEXPECTED_PACKET_FROM_CLIENT);
}
void TCPHandler::initBlockInput()
2012-03-11 08:52:56 +00:00
{
if (!state.block_in)
{
if (state.compression == Protocol::Compression::Enable)
state.maybe_compressed_in = std::make_shared<CompressedReadBuffer>(*in);
else
state.maybe_compressed_in = in;
Block header;
if (state.io.out)
header = state.io.out->getHeader();
2019-05-28 18:30:10 +00:00
else if (state.need_receive_data_for_input)
header = state.input_header;
last_block_in.header = header;
last_block_in.compression = state.compression;
state.block_in = std::make_shared<NativeBlockInputStream>(
*state.maybe_compressed_in,
header,
client_revision);
}
}
void TCPHandler::initBlockOutput(const Block & block)
{
if (!state.block_out)
{
if (!state.maybe_compressed_out)
{
std::string method = Poco::toUpper(query_context->getSettingsRef().network_compression_method.toString());
2018-12-21 12:17:30 +00:00
std::optional<int> level;
if (method == "ZSTD")
level = query_context->getSettingsRef().network_zstd_compression_level;
2018-12-21 12:17:30 +00:00
if (state.compression == Protocol::Compression::Enable)
state.maybe_compressed_out = std::make_shared<CompressedWriteBuffer>(
2018-12-21 12:17:30 +00:00
*out, CompressionCodecFactory::instance().get(method, level));
else
state.maybe_compressed_out = out;
}
state.block_out = std::make_shared<NativeBlockOutputStream>(
*state.maybe_compressed_out,
client_revision,
block.cloneEmpty(),
!connection_context.getSettingsRef().low_cardinality_allow_in_native_format);
}
2012-03-19 12:57:56 +00:00
}
2012-03-11 08:52:56 +00:00
void TCPHandler::initLogsBlockOutput(const Block & block)
{
if (!state.logs_block_out)
{
/// Use uncompressed stream since log blocks usually contain only one row
state.logs_block_out = std::make_shared<NativeBlockOutputStream>(
*out,
client_revision,
block.cloneEmpty(),
!connection_context.getSettingsRef().low_cardinality_allow_in_native_format);
}
}
2012-05-21 06:49:05 +00:00
bool TCPHandler::isQueryCancelled()
2012-05-09 08:16:09 +00:00
{
if (state.is_cancelled || state.sent_all_data)
return true;
if (after_check_cancelled.elapsed() / 1000 < query_context->getSettingsRef().interactive_delay)
return false;
after_check_cancelled.restart();
/// During request execution the only packet that can come from the client is stopping the query.
if (static_cast<ReadBufferFromPocoSocket &>(*in).poll(0))
{
UInt64 packet_type = 0;
readVarUInt(packet_type, *in);
switch (packet_type)
{
case Protocol::Client::Cancel:
if (state.empty())
throw NetException("Unexpected packet Cancel received from client", ErrorCodes::UNEXPECTED_PACKET_FROM_CLIENT);
LOG_INFO(log, "Query was cancelled.");
state.is_cancelled = true;
return true;
default:
throw NetException("Unknown packet from client", ErrorCodes::UNKNOWN_PACKET_FROM_CLIENT);
}
}
return false;
2012-05-09 08:16:09 +00:00
}
2012-05-21 06:49:05 +00:00
void TCPHandler::sendData(const Block & block)
2012-03-19 12:57:56 +00:00
{
initBlockOutput(block);
2012-03-11 08:52:56 +00:00
writeVarUInt(Protocol::Server::Data, *out);
/// Send external table name (empty name is the main table)
writeStringBinary("", *out);
2012-05-21 06:49:05 +00:00
state.block_out->write(block);
state.maybe_compressed_out->next();
out->next();
2012-03-19 12:57:56 +00:00
}
2012-05-21 06:49:05 +00:00
void TCPHandler::sendLogData(const Block & block)
{
initLogsBlockOutput(block);
writeVarUInt(Protocol::Server::Log, *out);
/// Send log tag (empty tag is the default tag)
writeStringBinary("", *out);
state.logs_block_out->write(block);
out->next();
}
void TCPHandler::sendTableColumns(const ColumnsDescription & columns)
{
writeVarUInt(Protocol::Server::TableColumns, *out);
/// Send external table name (empty name is the main table)
writeStringBinary("", *out);
writeStringBinary(columns.toString(), *out);
out->next();
}
2018-08-24 07:30:53 +00:00
void TCPHandler::sendException(const Exception & e, bool with_stack_trace)
2012-03-19 12:57:56 +00:00
{
writeVarUInt(Protocol::Server::Exception, *out);
2018-08-24 07:30:53 +00:00
writeException(e, *out, with_stack_trace);
out->next();
2012-03-19 12:57:56 +00:00
}
2012-05-21 06:49:05 +00:00
void TCPHandler::sendEndOfStream()
2012-05-08 11:19:00 +00:00
{
state.sent_all_data = true;
writeVarUInt(Protocol::Server::EndOfStream, *out);
out->next();
2012-05-08 11:19:00 +00:00
}
2012-05-21 06:49:05 +00:00
void TCPHandler::updateProgress(const Progress & value)
2012-03-19 12:57:56 +00:00
{
state.progress.incrementPiecewiseAtomically(value);
}
2012-05-09 15:15:45 +00:00
void TCPHandler::sendProgress()
{
writeVarUInt(Protocol::Server::Progress, *out);
auto increment = state.progress.fetchAndResetPiecewiseAtomically();
increment.write(*out, client_revision);
out->next();
2012-03-11 08:52:56 +00:00
}
2012-03-09 15:46:52 +00:00
void TCPHandler::sendLogs()
{
if (!state.logs_queue)
return;
MutableColumns logs_columns;
MutableColumns curr_logs_columns;
size_t rows = 0;
for (; state.logs_queue->tryPop(curr_logs_columns); ++rows)
{
if (rows == 0)
{
logs_columns = std::move(curr_logs_columns);
}
else
{
for (size_t j = 0; j < logs_columns.size(); ++j)
logs_columns[j]->insertRangeFrom(*curr_logs_columns[j], 0, curr_logs_columns[j]->size());
}
}
if (rows > 0)
{
Block block = InternalTextLogsQueue::getSampleBlock();
block.setColumns(std::move(logs_columns));
sendLogData(block);
}
}
2012-03-09 15:46:52 +00:00
void TCPHandler::run()
{
try
{
runImpl();
LOG_INFO(log, "Done processing connection.");
}
catch (Poco::Exception & e)
{
/// Timeout - not an error.
if (!strcmp(e.what(), "Timeout"))
{
LOG_DEBUG(log, "Poco::Exception. Code: " << ErrorCodes::POCO_EXCEPTION << ", e.code() = " << e.code()
<< ", e.displayText() = " << e.displayText() << ", e.what() = " << e.what());
}
else
throw;
}
2012-03-09 15:46:52 +00:00
}
}