Skip to main content

内省功能

您可以使用本章中描述的函数来反省 ELFDWARF 用于查询分析。

!!! warning "警告" 这些功能很慢,可能会强加安全考虑。

对于内省功能的正确操作:

  • 安装 clickhouse-common-static-dbg 包。

  • 设置 allow_introspection_functions 设置为1。

    出于安全考虑,内省函数默认是关闭的。

ClickHouse将探查器报告保存到 trace_log 系统表. 确保正确配置了表和探查器。

addressToLine

将ClickHouse服务器进程内的虚拟内存地址转换为ClickHouse源代码中的文件名和行号。

如果您使用官方的ClickHouse软件包,您需要安装 clickhouse-common-static-dbg 包。

语法

addressToLine(address_of_binary_instruction)

参数

  • address_of_binary_instruction (UInt64) — 正在运行进程的指令地址。

返回值

  • 源代码文件名和行号(用冒号分隔的行号)

    示例, `/build/obj-x86_64-linux-gnu/../src/Common/ThreadPool.cpp:199`, where `199` is a line number.
  • 如果函数找不到调试信息,返回二进制文件的名称。

  • 如果地址无效,返回空字符串。

类型: 字符串.

示例

启用内省功能:

SET allow_introspection_functions=1

从中选择第一个字符串 trace_log 系统表:

SELECT * FROM system.trace_log LIMIT 1 \G
Row 1:
──────
event_date: 2019-11-19
event_time: 2019-11-19 18:57:23
revision: 54429
timer_type: Real
thread_number: 48
query_id: 421b6855-1858-45a5-8f37-f383409d6d72
trace: [140658411141617,94784174532828,94784076370703,94784076372094,94784076361020,94784175007680,140658411116251,140658403895439]

trace 字段包含采样时的堆栈跟踪。

获取单个地址的源代码文件名和行号:

SELECT addressToLine(94784076370703) \G
Row 1:
──────
addressToLine(94784076370703): /build/obj-x86_64-linux-gnu/../src/Common/ThreadPool.cpp:199

将函数应用于整个堆栈跟踪:

SELECT
arrayStringConcat(arrayMap(x -> addressToLine(x), trace), '\n') AS trace_source_code_lines
FROM system.trace_log
LIMIT 1
\G

arrayMap 功能允许处理的每个单独的元素 trace 阵列由 addressToLine 功能。 这种处理的结果,你在看 trace_source_code_lines 列的输出。

Row 1:
──────
trace_source_code_lines: /lib/x86_64-linux-gnu/libpthread-2.27.so
/usr/lib/debug/usr/bin/clickhouse
/build/obj-x86_64-linux-gnu/../src/Common/ThreadPool.cpp:199
/build/obj-x86_64-linux-gnu/../src/Common/ThreadPool.h:155
/usr/include/c++/9/bits/atomic_base.h:551
/usr/lib/debug/usr/bin/clickhouse
/lib/x86_64-linux-gnu/libpthread-2.27.so
/build/glibc-OTsEL5/glibc-2.27/misc/../sysdeps/unix/sysv/linux/x86_64/clone.S:97

addressToSymbol

将ClickHouse服务器进程内的虚拟内存地址转换为ClickHouse对象文件中的符号。

语法

addressToSymbol(address_of_binary_instruction)

参数

  • address_of_binary_instruction (UInt64) — Address of instruction in a running process.

返回值

  • 来自ClickHouse对象文件的符号。
  • 如果地址无效,返回空字符串。

类型: 字符串.

示例

启用内省功能:

SET allow_introspection_functions=1

从中选择第一个字符串 trace_log 系统表:

SELECT * FROM system.trace_log LIMIT 1 \G
Row 1:
──────
event_date: 2019-11-20
event_time: 2019-11-20 16:57:59
revision: 54429
timer_type: Real
thread_number: 48
query_id: 724028bf-f550-45aa-910d-2af6212b94ac
trace: [94138803686098,94138815010911,94138815096522,94138815101224,94138815102091,94138814222988,94138806823642,94138814457211,94138806823642,94138814457211,94138806823642,94138806795179,94138806796144,94138753770094,94138753771646,94138753760572,94138852407232,140399185266395,140399178045583]

trace 字段包含采样时的堆栈跟踪。

获取单个地址的符号:

SELECT addressToSymbol(94138803686098) \G
Row 1:
──────
addressToSymbol(94138803686098): _ZNK2DB24IAggregateFunctionHelperINS_20AggregateFunctionSumImmNS_24AggregateFunctionSumDataImEEEEE19addBatchSinglePlaceEmPcPPKNS_7IColumnEPNS_5ArenaE

将函数应用于整个堆栈跟踪:

SELECT
arrayStringConcat(arrayMap(x -> addressToSymbol(x), trace), '\n') AS trace_symbols
FROM system.trace_log
LIMIT 1
\G

arrayMap 功能允许处理的每个单独的元素 trace 阵列由 addressToSymbols 功能。 这种处理的结果,你在看 trace_symbols 列的输出。

Row 1:
──────
trace_symbols: _ZNK2DB24IAggregateFunctionHelperINS_20AggregateFunctionSumImmNS_24AggregateFunctionSumDataImEEEEE19addBatchSinglePlaceEmPcPPKNS_7IColumnEPNS_5ArenaE
_ZNK2DB10Aggregator21executeWithoutKeyImplERPcmPNS0_28AggregateFunctionInstructionEPNS_5ArenaE
_ZN2DB10Aggregator14executeOnBlockESt6vectorIN3COWINS_7IColumnEE13immutable_ptrIS3_EESaIS6_EEmRNS_22AggregatedDataVariantsERS1_IPKS3_SaISC_EERS1_ISE_SaISE_EERb
_ZN2DB10Aggregator14executeOnBlockERKNS_5BlockERNS_22AggregatedDataVariantsERSt6vectorIPKNS_7IColumnESaIS9_EERS6_ISB_SaISB_EERb
_ZN2DB10Aggregator7executeERKSt10shared_ptrINS_17IBlockInputStreamEERNS_22AggregatedDataVariantsE
_ZN2DB27AggregatingBlockInputStream8readImplEv
_ZN2DB17IBlockInputStream4readEv
_ZN2DB26ExpressionBlockInputStream8readImplEv
_ZN2DB17IBlockInputStream4readEv
_ZN2DB26ExpressionBlockInputStream8readImplEv
_ZN2DB17IBlockInputStream4readEv
_ZN2DB28AsynchronousBlockInputStream9calculateEv
_ZNSt17_Function_handlerIFvvEZN2DB28AsynchronousBlockInputStream4nextEvEUlvE_E9_M_invokeERKSt9_Any_data
_ZN14ThreadPoolImplI20ThreadFromGlobalPoolE6workerESt14_List_iteratorIS0_E
_ZZN20ThreadFromGlobalPoolC4IZN14ThreadPoolImplIS_E12scheduleImplIvEET_St8functionIFvvEEiSt8optionalImEEUlvE1_JEEEOS4_DpOT0_ENKUlvE_clEv
_ZN14ThreadPoolImplISt6threadE6workerESt14_List_iteratorIS0_E
execute_native_thread_routine
start_thread
clone

demangle

转换一个符号,您可以使用 addressToSymbol 函数到C++函数名。

语法

demangle(symbol)

参数

  • symbol (字符串) — Symbol from an object file.

返回值

  • C++函数的名称。
  • 如果符号无效,则为空字符串。

类型: 字符串.

示例

启用内省功能:

SET allow_introspection_functions=1

从中选择第一个字符串 trace_log 系统表:

SELECT * FROM system.trace_log LIMIT 1 \G
Row 1:
──────
event_date: 2019-11-20
event_time: 2019-11-20 16:57:59
revision: 54429
timer_type: Real
thread_number: 48
query_id: 724028bf-f550-45aa-910d-2af6212b94ac
trace: [94138803686098,94138815010911,94138815096522,94138815101224,94138815102091,94138814222988,94138806823642,94138814457211,94138806823642,94138814457211,94138806823642,94138806795179,94138806796144,94138753770094,94138753771646,94138753760572,94138852407232,140399185266395,140399178045583]

trace 字段包含采样时的堆栈跟踪。

获取单个地址的函数名称:

SELECT demangle(addressToSymbol(94138803686098)) \G
Row 1:
──────
demangle(addressToSymbol(94138803686098)): DB::IAggregateFunctionHelper<DB::AggregateFunctionSum<unsigned long, unsigned long, DB::AggregateFunctionSumData<unsigned long> > >::addBatchSinglePlace(unsigned long, char*, DB::IColumn const**, DB::Arena*) const

将函数应用于整个堆栈跟踪:

SELECT
arrayStringConcat(arrayMap(x -> demangle(addressToSymbol(x)), trace), '\n') AS trace_functions
FROM system.trace_log
LIMIT 1
\G

arrayMap 功能允许处理的每个单独的元素 trace 阵列由 demangle 功能。 这种处理的结果,你在看 trace_functions 列的输出。

Row 1:
──────
trace_functions: DB::IAggregateFunctionHelper<DB::AggregateFunctionSum<unsigned long, unsigned long, DB::AggregateFunctionSumData<unsigned long> > >::addBatchSinglePlace(unsigned long, char*, DB::IColumn const**, DB::Arena*) const
DB::Aggregator::executeWithoutKeyImpl(char*&, unsigned long, DB::Aggregator::AggregateFunctionInstruction*, DB::Arena*) const
DB::Aggregator::executeOnBlock(std::vector<COW<DB::IColumn>::immutable_ptr<DB::IColumn>, std::allocator<COW<DB::IColumn>::immutable_ptr<DB::IColumn> > >, unsigned long, DB::AggregatedDataVariants&, std::vector<DB::IColumn const*, std::allocator<DB::IColumn const*> >&, std::vector<std::vector<DB::IColumn const*, std::allocator<DB::IColumn const*> >, std::allocator<std::vector<DB::IColumn const*, std::allocator<DB::IColumn const*> > > >&, bool&)
DB::Aggregator::executeOnBlock(DB::Block const&, DB::AggregatedDataVariants&, std::vector<DB::IColumn const*, std::allocator<DB::IColumn const*> >&, std::vector<std::vector<DB::IColumn const*, std::allocator<DB::IColumn const*> >, std::allocator<std::vector<DB::IColumn const*, std::allocator<DB::IColumn const*> > > >&, bool&)
DB::Aggregator::execute(std::shared_ptr<DB::IBlockInputStream> const&, DB::AggregatedDataVariants&)
DB::AggregatingBlockInputStream::readImpl()
DB::IBlockInputStream::read()
DB::ExpressionBlockInputStream::readImpl()
DB::IBlockInputStream::read()
DB::ExpressionBlockInputStream::readImpl()
DB::IBlockInputStream::read()
DB::AsynchronousBlockInputStream::calculate()
std::_Function_handler<void (), DB::AsynchronousBlockInputStream::next()::{lambda()#1}>::_M_invoke(std::_Any_data const&)
ThreadPoolImpl<ThreadFromGlobalPool>::worker(std::_List_iterator<ThreadFromGlobalPool>)
ThreadFromGlobalPool::ThreadFromGlobalPool<ThreadPoolImpl<ThreadFromGlobalPool>::scheduleImpl<void>(std::function<void ()>, int, std::optional<unsigned long>)::{lambda()#3}>(ThreadPoolImpl<ThreadFromGlobalPool>::scheduleImpl<void>(std::function<void ()>, int, std::optional<unsigned long>)::{lambda()#3}&&)::{lambda()#1}::operator()() const
ThreadPoolImpl<std::thread>::worker(std::_List_iterator<std::thread>)
execute_native_thread_routine
start_thread
clone