[BUG] ORC writer hit cudaErrorInvalidValue exception with ZSTD compression

[ttnghia]

With some changes to the current ORC unit tests, I got cuda exception:

unknown file: Failure
C++ exception with description "CUDA error encountered at: ../../src/io/orc/writer_impl.cu:1591: 1
 cudaErrorInvalidValue invalid argument" thrown in the test body.
[  FAILED  ] OrcWriterNumericTypeTest/0.SingleColumn, where TypeParam = signed char (382 ms)

The changes are below:

diff --git a/cpp/tests/io/orc_test.cpp b/cpp/tests/io/orc_test.cpp
index 2ae6edc6c7..a8bfed6400 100644
--- a/cpp/tests/io/orc_test.cpp
+++ b/cpp/tests/io/orc_test.cpp
@@ -224,16 +224,17 @@ struct SkipRowTest {
 
 TYPED_TEST(OrcWriterNumericTypeTest, SingleColumn)
 {
-  auto sequence = cudf::detail::make_counting_transform_iterator(0, [](auto i) { return i; });
+  auto sequence = cudf::detail::make_counting_transform_iterator(0, [](auto i) { return i % 100; });
 
-  constexpr auto num_rows = 100;
+  constexpr auto num_rows = 1000000;
   column_wrapper<TypeParam, typename decltype(sequence)::value_type> col(sequence,
                                                                          sequence + num_rows);
   table_view expected({col});
 
   auto filepath = temp_env->get_temp_filepath("OrcSingleColumn.orc");
   cudf::io::orc_writer_options out_opts =
-    cudf::io::orc_writer_options::builder(cudf::io::sink_info{filepath}, expected);
+    cudf::io::orc_writer_options::builder(cudf::io::sink_info{filepath}, expected)
+      .compression(cudf::io::compression_type::ZSTD);
   cudf::io::write_orc(out_opts);

[ttnghia]

CC @vuule.

[ttnghia]

Notice that the exception is thrown with compression ZSTD, not with other compression like SNAPPY. I didn't test all compression types.

[ttnghia]

I also tested that roudtrip example with Parquet writer/reader without any issues. So the problem should be due to ORC writer itself.

  cudf::io::parquet_writer_options args =
    cudf::io::parquet_writer_options::builder(cudf::io::sink_info{filepath}, expected)
      .compression(cudf::io::compression_type::ZSTD);
  cudf::io::write_parquet(args);

  cudf::io::parquet_reader_options read_opts =
    cudf::io::parquet_reader_options::builder(cudf::io::source_info{filepath});
  auto result = cudf::io::read_parquet(read_opts);

  CUDF_TEST_EXPECT_TABLES_EQUAL(expected, result.tbl->view());

[vuule]

Thank you for filing the issue. I'll try to get a local repro today.

[vuule]

Got local repro, will look into this tomorrow.