Introspection Functions

You can use functions described in this chapter to introspect ELF and DWARF for query profiling.

Warning

These functions are slow and may impose security considerations.

For proper operation of introspection functions:

  • Install the clickhouse-common-static-dbg package.

  • Set the allow_introspection_functions setting to 1.

    1. For security reasons introspection functions are disabled by default.

ClickHouse saves profiler reports to the trace_log system table. Make sure the table and profiler are configured properly.

addressToLine

Converts virtual memory address inside ClickHouse server process to the filename and the line number in ClickHouse source code.

If you use official ClickHouse packages, you need to install the clickhouse-common-static-dbg package.

Syntax

  1. addressToLine(address_of_binary_instruction)

Parameters

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

Returned value

  • Source code filename and the line number in this file delimited by colon.

    1. For example, `/build/obj-x86_64-linux-gnu/../src/Common/ThreadPool.cpp:199`, where `199` is a line number.

  • Name of a binary, if the function couldn’t find the debug information.

  • Empty string, if the address is not valid.

Type: String.

Example

Enabling introspection functions:

  1. SET allow_introspection_functions=1

Selecting the first string from the trace_log system table:

  1. SELECT * FROM system.trace_log LIMIT 1 \G
  1. Row 1:
  2. ──────
  3. event_date:              2019-11-19
  4. event_time:              2019-11-19 18:57:23
  5. revision:                54429
  6. timer_type:              Real
  7. thread_number:           48
  8. query_id:                421b6855-1858-45a5-8f37-f383409d6d72
  9. trace:                   [140658411141617,94784174532828,94784076370703,94784076372094,94784076361020,94784175007680,140658411116251,140658403895439]

The trace field contains the stack trace at the moment of sampling.

Getting the source code filename and the line number for a single address:

  1. SELECT addressToLine(94784076370703) \G
  1. Row 1:
  2. ──────
  3. addressToLine(94784076370703): /build/obj-x86_64-linux-gnu/../src/Common/ThreadPool.cpp:199

Applying the function to the whole stack trace:

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

The arrayMap function allows to process each individual element of the trace array by the addressToLine function. The result of this processing you see in the trace_source_code_lines column of output.

  1. Row 1:
  2. ──────
  3. trace_source_code_lines: /lib/x86_64-linux-gnu/libpthread-2.27.so
  4. /usr/lib/debug/usr/bin/clickhouse
  5. /build/obj-x86_64-linux-gnu/../src/Common/ThreadPool.cpp:199
  6. /build/obj-x86_64-linux-gnu/../src/Common/ThreadPool.h:155
  7. /usr/include/c++/9/bits/atomic_base.h:551
  8. /usr/lib/debug/usr/bin/clickhouse
  9. /lib/x86_64-linux-gnu/libpthread-2.27.so
  10. /build/glibc-OTsEL5/glibc-2.27/misc/../sysdeps/unix/sysv/linux/x86_64/clone.S:97

addressToSymbol

Converts virtual memory address inside ClickHouse server process to the symbol from ClickHouse object files.

Syntax

  1. addressToSymbol(address_of_binary_instruction)

Parameters

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

Returned value

  • Symbol from ClickHouse object files.
  • Empty string, if the address is not valid.

Type: String.

Example

Enabling introspection functions:

  1. SET allow_introspection_functions=1

Selecting the first string from the trace_log system table:

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

The trace field contains the stack trace at the moment of sampling.

Getting a symbol for a single address:

  1. SELECT addressToSymbol(94138803686098) \G
  1. Row 1:
  2. ──────
  3. addressToSymbol(94138803686098): _ZNK2DB24IAggregateFunctionHelperINS_20AggregateFunctionSumImmNS_24AggregateFunctionSumDataImEEEEE19addBatchSinglePlaceEmPcPPKNS_7IColumnEPNS_5ArenaE

Applying the function to the whole stack trace:

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

The arrayMap function allows to process each individual element of the trace array by the addressToSymbols function. The result of this processing you see in the trace_symbols column of output.

  1. Row 1:
  2. ──────
  3. trace_symbols: _ZNK2DB24IAggregateFunctionHelperINS_20AggregateFunctionSumImmNS_24AggregateFunctionSumDataImEEEEE19addBatchSinglePlaceEmPcPPKNS_7IColumnEPNS_5ArenaE
  4. _ZNK2DB10Aggregator21executeWithoutKeyImplERPcmPNS0_28AggregateFunctionInstructionEPNS_5ArenaE
  5. _ZN2DB10Aggregator14executeOnBlockESt6vectorIN3COWINS_7IColumnEE13immutable_ptrIS3_EESaIS6_EEmRNS_22AggregatedDataVariantsERS1_IPKS3_SaISC_EERS1_ISE_SaISE_EERb
  6. _ZN2DB10Aggregator14executeOnBlockERKNS_5BlockERNS_22AggregatedDataVariantsERSt6vectorIPKNS_7IColumnESaIS9_EERS6_ISB_SaISB_EERb
  7. _ZN2DB10Aggregator7executeERKSt10shared_ptrINS_17IBlockInputStreamEERNS_22AggregatedDataVariantsE
  8. _ZN2DB27AggregatingBlockInputStream8readImplEv
  9. _ZN2DB17IBlockInputStream4readEv
  10. _ZN2DB26ExpressionBlockInputStream8readImplEv
  11. _ZN2DB17IBlockInputStream4readEv
  12. _ZN2DB26ExpressionBlockInputStream8readImplEv
  13. _ZN2DB17IBlockInputStream4readEv
  14. _ZN2DB28AsynchronousBlockInputStream9calculateEv
  15. _ZNSt17_Function_handlerIFvvEZN2DB28AsynchronousBlockInputStream4nextEvEUlvE_E9_M_invokeERKSt9_Any_data
  16. _ZN14ThreadPoolImplI20ThreadFromGlobalPoolE6workerESt14_List_iteratorIS0_E
  17. _ZZN20ThreadFromGlobalPoolC4IZN14ThreadPoolImplIS_E12scheduleImplIvEET_St8functionIFvvEEiSt8optionalImEEUlvE1_JEEEOS4_DpOT0_ENKUlvE_clEv
  18. _ZN14ThreadPoolImplISt6threadE6workerESt14_List_iteratorIS0_E
  19. execute_native_thread_routine
  20. start_thread
  21. clone

demangle

Converts a symbol that you can get using the addressToSymbol function to the C++ function name.

Syntax

  1. demangle(symbol)

Parameters

  • symbol (String) — Symbol from an object file.

Returned value

  • Name of the C++ function.
  • Empty string if a symbol is not valid.

Type: String.

Example

Enabling introspection functions:

  1. SET allow_introspection_functions=1

Selecting the first string from the trace_log system table:

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

The trace field contains the stack trace at the moment of sampling.

Getting a function name for a single address:

  1. SELECT demangle(addressToSymbol(94138803686098)) \G
  1. Row 1:
  2. ──────
  3. 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

Applying the function to the whole stack trace:

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

The arrayMap function allows to process each individual element of the trace array by the demangle function. The result of this processing you see in the trace_functions column of output.

  1. Row 1:
  2. ──────
  3. trace_functions: DB::IAggregateFunctionHelper<DB::AggregateFunctionSum<unsigned long, unsigned long, DB::AggregateFunctionSumData<unsigned long> > >::addBatchSinglePlace(unsigned long, char*, DB::IColumn const**, DB::Arena*) const
  4. DB::Aggregator::executeWithoutKeyImpl(char*&, unsigned long, DB::Aggregator::AggregateFunctionInstruction*, DB::Arena*) const
  5. 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&)
  6. 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&)
  7. DB::Aggregator::execute(std::shared_ptr<DB::IBlockInputStream> const&, DB::AggregatedDataVariants&)
  8. DB::AggregatingBlockInputStream::readImpl()
  9. DB::IBlockInputStream::read()
  10. DB::ExpressionBlockInputStream::readImpl()
  11. DB::IBlockInputStream::read()
  12. DB::ExpressionBlockInputStream::readImpl()
  13. DB::IBlockInputStream::read()
  14. DB::AsynchronousBlockInputStream::calculate()
  15. std::_Function_handler<void (), DB::AsynchronousBlockInputStream::next()::{lambda()#1}>::_M_invoke(std::_Any_data const&)
  16. ThreadPoolImpl<ThreadFromGlobalPool>::worker(std::_List_iterator<ThreadFromGlobalPool>)
  17. 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
  18. ThreadPoolImpl<std::thread>::worker(std::_List_iterator<std::thread>)
  19. execute_native_thread_routine
  20. start_thread
  21. clone