[ARROW-15881] [c++] When link to libavrocpp.so, call parquet::arrow::WriteTable in a child thread get segmentation fault

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Not A Bug
• Affects Version/s: 7.0.0
• Component/s: C++
• Labels: avro newbie thread
• Environment: CentOS7,gcc7.0+,C++17 or C++11
• External issue URL: https://github.com/apache/arrow/issues/31313
• Assignee: Unassigned
• Reporter: zzh
• Created: 09/Mar/22 12:45
• Updated: 11/Jan/23 11:40
• Resolved: 14/Mar/22 08:23

Description

I resolve this problem. I use rpm package to install arrow before. After I build arrow from source code (use same gcc I build avro) and install, there is no segmentation fault. 

----------------------------------------------------------------------------------------------

When I try to use Arrow to write parquet files, I encounter a error. The parquet::arrow::WriteTable out child thread can call successful, but  parquet::arrow::WriteTable in child thread while cause Segmentation fault.

In my CMakeList.txt,I link libavrocpp.so because I want to use avro in project after finish this test. And I find that if I remove the link of libavrocpp.so, code while run successful.

The code like this:

arrow::Int64Builder test_a;
for (int i = 0; i < 1e7; ++i) {
  PARQUET_THROW_NOT_OK(test_a.Append(i));
}
auto sc = arrow::schema({arrow::field("A", arrow::int64())});
auto table = arrow::Table::Make(sc,{test_a.Finish().ValueOrDie()});
const string &fileid = sole::uuid4().str();
string filename = "test.parq";
try {
  std::shared_ptr<arrow::io::FileOutputStream> outfile;
  PARQUET_ASSIGN_OR_THROW(
          outfile,arrow::io::FileOutputStream::Open(filename)
  );
  PARQUET_THROW_NOT_OK(
          parquet::arrow::WriteTable(*table, arrow::default_memory_pool(), outfile, table->num_rows())
  );
} catch (exception &ex) {
  cout << ex.what() << endl;
}
shared_ptr<std::thread> thread = make_shared<std::thread>([=]() {
    arrow::Int64Builder test_a;
      for (int i = 0; i < 1e7; ++i) {
        PARQUET_THROW_NOT_OK(test_a.Append(i));
    }
    auto sc = arrow::schema({arrow::field("A", arrow::int64())});
    auto table = arrow::Table::Make(sc,{test_a.Finish().ValueOrDie()});
    const string &fileid = sole::uuid4().str();
    string filename = "test.parq";
    try {
        std::shared_ptr<arrow::io::FileOutputStream> outfile;
        PARQUET_ASSIGN_OR_THROW(
                outfile,arrow::io::FileOutputStream::Open(filename)
        );
        PARQUET_THROW_NOT_OK(
                parquet::arrow::WriteTable(*table, arrow::default_memory_pool(), outfile, table->num_rows())
        );
    } catch (exception &ex) {
        cout << ex.what() << endl;
    }
};
thread->join();
return 0;

 

The stack message is in the picture in attachment.