mapleFU commented on PR #38466:
URL: https://github.com/apache/arrow/pull/38466#issuecomment-1810477979
```c++
class DelayedBufferReader : public ::arrow::io::BufferReader {
public:
explicit DelayedBufferReader(const std::shared_ptr<::arrow::Buffer>&
buffer)
: ::arrow::io::BufferReader(buffer) {}
::arrow::Future<std::shared_ptr<Buffer>> ReadAsync(const
::arrow::io::IOContext& io_context, int64_t position,
int64_t nbytes)
override {
read_async_count.fetch_add(1);
auto self =
std::dynamic_pointer_cast<DelayedBufferReader>(shared_from_this());
return DeferNotOk(::arrow::io::internal::SubmitIO(
io_context, [self, position, nbytes]() ->
Result<std::shared_ptr<Buffer>> {
std::this_thread::sleep_for(std::chrono::seconds(1));
return self->DoReadAt(position, nbytes);
}));
}
std::atomic<int> read_async_count{0};
};
TEST_F(TestParquetFileFormat, MultithreadedScanUnsafe) {
auto reader = MakeGeneratedRecordBatch(schema({field("utf8", utf8())}),
10000, 100);
ASSERT_OK_AND_ASSIGN(auto buffer,
ParquetFormatHelper::Write(reader.get()));
std::vector<Future<>> completes;
std::vector<std::shared_ptr<arrow::internal::ThreadPool>> pools;
for (int idx = 0; idx < 3; ++idx) {
auto buffer_reader = std::make_shared<DelayedBufferReader>(buffer);
auto source = std::make_shared<FileSource>(buffer_reader,
buffer->size());
auto fragment = MakeFragment(*source);
std::shared_ptr<Scanner> scanner;
{
auto options = std::make_shared<ScanOptions>();
ASSERT_OK_AND_ASSIGN(auto thread_pool,
arrow::internal::ThreadPool::Make(1));
pools.emplace_back(thread_pool);
options->io_context =
::arrow::io::IOContext(::arrow::default_memory_pool(), pools.back().get());
auto fragment_scan_options =
std::make_shared<ParquetFragmentScanOptions>();
fragment_scan_options->arrow_reader_properties->set_pre_buffer(true);
options->fragment_scan_options = fragment_scan_options;
options->use_threads = true;
ScannerBuilder builder(ArithmeticDatasetFixture::schema(), fragment,
options);
ASSERT_OK(builder.UseThreads(true));
ASSERT_OK(builder.BatchSize(10000));
ASSERT_OK_AND_ASSIGN(scanner, builder.Finish());
}
ASSERT_OK_AND_ASSIGN(auto batch, scanner->Head(10000));
auto fut = scanner->ScanBatchesUnorderedAsync();
// ASSERT_OK_AND_ASSIGN(batch, scanner->Head(8000 * 10));
// Random ReadAsync calls
for (int yy = 0; yy < 16; yy++) {
completes.emplace_back(buffer_reader->ReadAsync(::arrow::io::IOContext(), 0,
1001));
}
scanner = nullptr;
}
for (auto& f: completes) {
f.Wait();
}
}
```
Finally I re-produce the issue successfully with same stack I meet, should I
use this @bkietz ? (this test might runs for a long time...)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
