[fix](memory safe) Fix block-reader memory alloc failed make be core …

…dump. (apache#36299) We should leave the construction of Arena of BlockReader to its init function to avoid problem like below: ```text *** Query id: 221bb894cfc945fb-ba8f584e7668a36c *** *** is nereids: 1 *** *** tablet id: 0 *** *** Aborted at 1718335148 (unix time) try "date -d @1718335148" if you are using GNU date *** *** Current BE git commitID: ac0568acff *** *** SIGSEGV address not mapped to object (@0x0) received by PID 1330723 (TID 1332580 OR 0x7fc6c7af5700) from PID 0; stack trace: *** 0# doris::signal::(anonymous namespace)::FailureSignalHandler(int, siginfo_t*, void*) at /mnt/disk1/hezhiqiang/doris/be/src/common/signal_handler.h:421 1# 0x00007FCF485C3B50 in /lib64/libc.so.6 2# doris::vectorized::NewOlapScanner::open(doris::RuntimeState*) at /mnt/disk1/hezhiqiang/doris/be/src/vec/exec/scan/new_olap_scanner.cpp:218 3# doris::vectorized::ScannerScheduler::_scanner_scan(std::shared_ptr<doris::vectorized::ScannerContext>, std::shared_ptr<doris::vectorized::ScanTask>) at /mnt/disk1/hezhiqiang/doris/be/src/vec/exec/scan/scanner_scheduler.cpp:249 4# doris::vectorized::ScannerScheduler::submit(std::shared_ptr<doris::vectorized::ScannerContext>, std::shared_ptr<doris::vectorized::ScanTask>)::$_0::operator()() const::{lambda()#1}::operator()() const at /mnt/disk1/hezhiqiang/doris/be/src/vec/exec/scan/scanner_scheduler.cpp:141 5# doris::vectorized::ScannerScheduler::submit(std::shared_ptr<doris::vectorized::ScannerContext>, std::shared_ptr<doris::vectorized::ScanTask>)::$_0::operator()() const at /mnt/disk1/hezhiqiang/doris/be/src/vec/exec/scan/scanner_scheduler.cpp:140 6# void std::__invoke_impl<void, doris::vectorized::ScannerScheduler::submit(std::shared_ptr<doris::vectorized::ScannerContext>, std::shared_ptr<doris::vectorized::ScanTask>)::$_0&>(std::__invoke_other, doris::vectorized::ScannerScheduler::submit(std::shared_ptr<doris::vectorized::ScannerContext>, std::shared_ptr<doris::vectorized::ScanTask>)::$_0&) at /mnt/disk1/hezhiqiang/toolchains/bin/../lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/invoke.h:61 7# std::enable_if<is_invocable_r_v<void, doris::vectorized::ScannerScheduler::submit(std::shared_ptr<doris::vectorized::ScannerContext>, std::shared_ptr<doris::vectorized::ScanTask>)::$_0&>, void>::type std::__invoke_r<void, doris::vectorized::ScannerScheduler::submit(std::shared_ptr<doris::vectorized::ScannerContext>, std::shared_ptr<doris::vectorized::ScanTask>)::$_0&>(doris::vectorized::ScannerScheduler::submit(std::shared_ptr<doris::vectorized::ScannerContext>, std::shared_ptr<doris::vectorized::ScanTask>)::$_0&) at /mnt/disk1/hezhiqiang/toolchains/bin/../lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/invoke.h:117 8# std::_Function_handler<void (), doris::vectorized::ScannerScheduler::submit(std::shared_ptr<doris::vectorized::ScannerContext>, std::shared_ptr<doris::vectorized::ScanTask>)::$_0>::_M_invoke(std::_Any_data const&) at /mnt/disk1/hezhiqiang/toolchains/bin/../lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/std_function.h:291 9# std::function<void ()>::operator()() const at /mnt/disk1/hezhiqiang/toolchains/bin/../lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/std_function.h:560 10# doris::FunctionRunnable::run() at /mnt/disk1/hezhiqiang/doris/be/src/util/threadpool.cpp:48 11# doris::ThreadPool::dispatch_thread() at /mnt/disk1/hezhiqiang/doris/be/src/util/threadpool.cpp:543 12# void std::__invoke_impl<void, void (doris::ThreadPool::*&)(), doris::ThreadPool*&>(std::__invoke_memfun_deref, void (doris::ThreadPool::*&)(), doris::ThreadPool*&) at /mnt/disk1/hezhiqiang/toolchains/bin/../lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/invoke.h:74 13# std::__invoke_result<void (doris::ThreadPool::*&)(), doris::ThreadPool*&>::type std::__invoke<void (doris::ThreadPool::*&)(), doris::ThreadPool*&>(void (doris::ThreadPool::*&)(), doris::ThreadPool*&) at /mnt/disk1/hezhiqiang/toolchains/bin/../lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/invoke.h:96 14# void std::_Bind<void (doris::ThreadPool::*(doris::ThreadPool*))()>::__call<void, , 0ul>(std::tuple<>&&, std::_Index_tuple<0ul>) at /mnt/disk1/hezhiqiang/toolchains/bin/../lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/functional:420 15# void std::_Bind<void (doris::ThreadPool::*(doris::ThreadPool*))()>::operator()<, void>() at /mnt/disk1/hezhiqiang/toolchains/bin/../lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/functional:503 16# void std::__invoke_impl<void, std::_Bind<void (doris::ThreadPool::*(doris::ThreadPool*))()>&>(std::__invoke_other, std::_Bind<void (doris::ThreadPool::*(doris::ThreadPool*))()>&) at /mnt/disk1/hezhiqiang/toolchains/bin/../lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/invoke.h:61 17# std::enable_if<is_invocable_r_v<void, std::_Bind<void (doris::ThreadPool::*(doris::ThreadPool*))()>&>, void>::type std::__invoke_r<void, std::_Bind<void (doris::ThreadPool::*(doris::ThreadPool*))()>&>(std::_Bind<void (doris::ThreadPool::*(doris::ThreadPool*))()>&) at /mnt/disk1/hezhiqiang/toolchains/bin/../lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/invoke.h:117 18# std::_Function_handler<void (), std::_Bind<void (doris::ThreadPool::*(doris::ThreadPool*))()> >::_M_invoke(std::_Any_data const&) at /mnt/disk1/hezhiqiang/toolchains/bin/../lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/std_function.h:291 19# std::function<void ()>::operator()() const at /mnt/disk1/hezhiqiang/toolchains/bin/../lib/gcc/x86_64-linux-gnu/11/../../../../include/c++/11/bits/std_function.h:560 20# doris::Thread::supervise_thread(void*) at /mnt/disk1/hezhiqiang/doris/be/src/util/thread.cpp:498 21# start_thread in /lib64/libpthread.so.0 22# __clone in /lib64/libc.so.6 ```
morningman · Jun 14, 2024 · 4bb46a8 · 4bb46a8
1 parent c1fef47
commit 4bb46a8
Show file tree

Hide file tree

Showing 4 changed files with 21 additions and 6 deletions.
diff --git a/be/src/vec/olap/block_reader.cpp b/be/src/vec/olap/block_reader.cpp
@@ -209,6 +209,8 @@ Status BlockReader::_init_agg_state(const ReaderParams& read_params) {
 Status BlockReader::init(const ReaderParams& read_params) {
     RETURN_IF_ERROR(TabletReader::init(read_params));
 
+    _arena = std::make_unique<Arena>();
+
     int32_t return_column_size = read_params.origin_return_columns->size();
     _return_columns_loc.resize(read_params.return_columns.size());
     for (int i = 0; i < return_column_size; ++i) {
@@ -511,6 +513,10 @@ size_t BlockReader::_copy_agg_data() {
 }
 
 void BlockReader::_update_agg_value(MutableColumns& columns, int begin, int end, bool is_close) {
+    if (!_arena) [[unlikely]] {
+        return;
+    }
+
     for (int i = 0; i < _agg_columns_idx.size(); i++) {
         auto idx = _agg_columns_idx[i];
 
@@ -520,7 +526,7 @@ void BlockReader::_update_agg_value(MutableColumns& columns, int begin, int end,
 
         if (begin <= end) {
             function->add_batch_range(begin, end, place, const_cast<const IColumn**>(&column_ptr),
-                                      &_arena, _stored_has_null_tag[idx]);
+                                      _arena.get(), _stored_has_null_tag[idx]);
         }
 
         if (is_close) {
@@ -529,8 +535,9 @@ void BlockReader::_update_agg_value(MutableColumns& columns, int begin, int end,
             function->reset(place);
         }
     }
+
     if (is_close) {
-        _arena.clear();
+        _arena->clear();
     }
 }
 

diff --git a/be/src/vec/olap/block_reader.h b/be/src/vec/olap/block_reader.h
@@ -119,7 +119,8 @@ class BlockReader final : public TabletReader {
 
     bool _is_rowsets_overlapping = true;
 
-    Arena _arena;
+    // Use pointer to avoid allocing memory during construction
+    std::unique_ptr<Arena> _arena;
 };
 
 } // namespace vectorized

diff --git a/be/src/vec/olap/vertical_block_reader.cpp b/be/src/vec/olap/vertical_block_reader.cpp
@@ -208,6 +208,8 @@ Status VerticalBlockReader::init(const ReaderParams& read_params) {
     _reader_context.batch_size = opts.block_row_max;
     RETURN_IF_ERROR(TabletReader::init(read_params));
 
+    _arena = std::make_unique<Arena>();
+
     auto status = _init_collect_iter(read_params);
     if (!status.ok()) [[unlikely]] {
         if (!config::is_cloud_mode()) {
@@ -298,14 +300,17 @@ void VerticalBlockReader::_update_agg_data(MutableColumns& columns) {
 
 void VerticalBlockReader::_update_agg_value(MutableColumns& columns, int begin, int end,
                                             bool is_close) {
+    if (!_arena) [[unlikely]] {
+        return;
+    }
     for (size_t idx = 0; idx < _return_columns.size(); ++idx) {
         AggregateFunctionPtr function = _agg_functions[idx];
         AggregateDataPtr place = _agg_places[idx];
         auto* column_ptr = _stored_data_columns[idx].get();
 
         if (begin <= end) {
             function->add_batch_range(begin, end, place, const_cast<const IColumn**>(&column_ptr),
-                                      &_arena, _stored_has_null_tag[idx]);
+                                      _arena.get(), _stored_has_null_tag[idx]);
         }
 
         if (is_close) {
@@ -315,7 +320,7 @@ void VerticalBlockReader::_update_agg_value(MutableColumns& columns, int begin,
         }
     }
     if (is_close) {
-        _arena.clear();
+        _arena->clear();
     }
 }
 

diff --git a/be/src/vec/olap/vertical_block_reader.h b/be/src/vec/olap/vertical_block_reader.h
@@ -107,7 +107,9 @@ class VerticalBlockReader final : public TabletReader {
     // for agg mode
     std::vector<AggregateFunctionPtr> _agg_functions;
     std::vector<AggregateDataPtr> _agg_places;
-    Arena _arena;
+
+    // Use pointer to avoid memory allocation during construction
+    std::unique_ptr<Arena> _arena;
 
     std::vector<int> _normal_columns_idx;
     std::vector<int> _agg_columns_idx;