Log cuIO warnings using the libcudf logger (#13043)

Log warning messages in the following scenarios:

- cuIO reader benchmark is ran without dropping the L3 cache.
- An nvCOMP call fails and we fall back to older API.
- Mangling of duplicated column names is disabled and some columns are dropped.
- Default compression fails in ORC writer and we fall back to uncompressed output.
- Parquet writer is writing a decimal column with precision < 10 as int64.

Authors:
  - Vukasin Milovanovic (https://github.com/vuule)

Approvers:
  - Nghia Truong (https://github.com/ttnghia)
  - Mark Harris (https://github.com/harrism)

URL: #13043

cpp/benchmarks/io/cuio_common.cpp

@@ -15,6 +15,7 @@
  */
 
 #include <benchmarks/io/cuio_common.hpp>
+#include <cudf/detail/utilities/logger.hpp>
 
 #include <cstdio>
 #include <fstream>
@@ -173,10 +174,24 @@ std::string exec_cmd(std::string_view cmd)
   return error_out;
 }
 
+void log_l3_warning_once()
+{
+  static bool is_logged = false;
+  if (not is_logged) {
+    CUDF_LOG_WARN(
+      "Running benchmarks without dropping the L3 cache; results may not reflect file IO "
+      "throughput");
+    is_logged = true;
+  }
+}
+
 void try_drop_l3_cache()
 {
   static bool is_drop_cache_enabled = std::getenv("CUDF_BENCHMARK_DROP_CACHE") != nullptr;
-  if (not is_drop_cache_enabled) { return; }
+  if (not is_drop_cache_enabled) {
+    log_l3_warning_once();
+    return;
+  }
 
   std::array drop_cache_cmds{"/sbin/sysctl vm.drop_caches=3", "sudo /sbin/sysctl vm.drop_caches=3"};
   CUDF_EXPECTS(std::any_of(drop_cache_cmds.cbegin(),

cpp/src/io/comp/nvcomp_adapter.cpp

@@ -269,6 +269,9 @@ size_t batched_compress_temp_size(compression_type compression,
       compression, num_chunks, max_uncomp_chunk_size, max_total_uncomp_size);
   } catch (...) {
     // Ignore errors in the expanded version; fall back to the old API in case of failure
+    CUDF_LOG_WARN(
+      "CompressGetTempSizeEx call failed, falling back to CompressGetTempSize; this may increase "
+      "the memory usage");
   }
 #endif
 

cpp/src/io/csv/reader_impl.cu

@@ -712,7 +712,9 @@ table_with_metadata read_csv(cudf::io::datasource* source,
     if (!reader_opts.is_enabled_mangle_dupe_cols()) {
       for (auto& col_name : column_names) {
         if (++col_names_counts[col_name] > 1) {
-          // All duplicate columns will be ignored; First appearance is parsed
+          CUDF_LOG_WARN("Multiple columns with name {}; only the first appearance is parsed",
+                        col_name);
+
           const auto idx    = &col_name - column_names.data();
           column_flags[idx] = column_parse::disabled;
         }

cpp/src/io/orc/stripe_enc.cu

@@ -1334,6 +1334,7 @@ void CompressOrcDataStreams(uint8_t* compressed_data,
         [] __device__(compression_result & stat) { stat.status = compression_status::FAILURE; });
       // Since SNAPPY is the default compression (may not be explicitly requested), fall back to
       // writing without compression
+      CUDF_LOG_WARN("ORC writer: compression failed, writing uncompressed data");
     }
   } else if (compression == ZLIB) {
     if (auto const reason = nvcomp::is_compression_disabled(nvcomp::compression_type::DEFLATE);

cpp/src/io/parquet/writer_impl.cu

@@ -488,6 +488,9 @@ struct leaf_schema_fn {
     if (col_meta.is_decimal_precision_set()) {
       CUDF_EXPECTS(col_meta.get_decimal_precision() >= col_schema.decimal_scale,
                    "Precision must be equal to or greater than scale!");
+      if (col_schema.type == Type::INT64 and col_meta.get_decimal_precision() < 10) {
+        CUDF_LOG_WARN("Parquet writer: writing a decimal column with precision < 10 as int64");
+      }
       col_schema.decimal_precision = col_meta.get_decimal_precision();
     }
   }