[GitHub] [arrow] bkietz commented on a change in pull request #7233: ARROW-8860: [C++] Fix IPC/Feather decompression for nested types (with child_data)

[GitBox <gi...@apache.org>]

bkietz commented on a change in pull request #7233:
URL: https://github.com/apache/arrow/pull/7233#discussion_r434704664



##########
File path: cpp/src/arrow/ipc/reader.cc
##########
@@ -344,50 +344,73 @@ class ArrayLoader {
   ArrayData* out_;
 };
 
+Result<std::shared_ptr<Buffer>> DecompressBuffer(const std::shared_ptr<Buffer>& buf,
+                                                 const IpcReadOptions& options,
+                                                 util::Codec* codec) {
+  if (buf == nullptr || buf->size() == 0) {
+    return buf;
+  }
+
+  if (buf->size() < 8) {
+    return Status::Invalid(
+        "Likely corrupted message, compressed buffers "
+        "are larger than 8 bytes by construction");
+  }
+
+  const uint8_t* data = buf->data();
+  int64_t compressed_size = buf->size() - sizeof(int64_t);
+  int64_t uncompressed_size = BitUtil::FromLittleEndian(util::SafeLoadAs<int64_t>(data));
+
+  ARROW_ASSIGN_OR_RAISE(auto uncompressed,
+                        AllocateBuffer(uncompressed_size, options.memory_pool));
+
+  ARROW_ASSIGN_OR_RAISE(
+      int64_t actual_decompressed,
+      codec->Decompress(compressed_size, data + sizeof(int64_t), uncompressed_size,
+                        uncompressed->mutable_data()));
+  if (actual_decompressed != uncompressed_size) {
+    return Status::Invalid("Failed to fully decompress buffer, expected ",
+                           uncompressed_size, " bytes but decompressed ",
+                           actual_decompressed);
+  }
+
+  return std::move(uncompressed);
+}
+
 Status DecompressBuffers(Compression::type compression, const IpcReadOptions& options,
                          std::vector<std::shared_ptr<ArrayData>>* fields) {
-  std::unique_ptr<util::Codec> codec;
-  ARROW_ASSIGN_OR_RAISE(codec, util::Codec::Create(compression));
+  struct BufferAccumulator {
+    using BufferPtrVector = std::vector<std::shared_ptr<Buffer>*>;
 
-  auto DecompressOne = [&](int i) {
-    ArrayData* arr = (*fields)[i].get();
-    for (size_t i = 0; i < arr->buffers.size(); ++i) {
-      if (arr->buffers[i] == nullptr) {
-        continue;
-      }
-      if (arr->buffers[i]->size() == 0) {
-        continue;
-      }
-      if (arr->buffers[i]->size() < 8) {
-        return Status::Invalid(
-            "Likely corrupted message, compressed buffers "
-            "are larger than 8 bytes by construction");
+    void AppendFrom(const std::vector<std::shared_ptr<ArrayData>>& fields) {
+      for (const auto& field : fields) {
+        for (auto& buffer : field->buffers) {
+          buffers_.push_back(&buffer);
+        }
+        AppendFrom(field->child_data);
       }
-      const uint8_t* data = arr->buffers[i]->data();
-      int64_t compressed_size = arr->buffers[i]->size() - sizeof(int64_t);
-      int64_t uncompressed_size =
-          BitUtil::FromLittleEndian(util::SafeLoadAs<int64_t>(data));
-
-      ARROW_ASSIGN_OR_RAISE(auto uncompressed,
-                            AllocateBuffer(uncompressed_size, options.memory_pool));
+    }
 
-      int64_t actual_decompressed;
-      ARROW_ASSIGN_OR_RAISE(
-          actual_decompressed,
-          codec->Decompress(compressed_size, data + sizeof(int64_t), uncompressed_size,
-                            uncompressed->mutable_data()));
-      if (actual_decompressed != uncompressed_size) {
-        return Status::Invalid("Failed to fully decompress buffer, expected ",
-                               uncompressed_size, " bytes but decompressed ",
-                               actual_decompressed);
-      }
-      arr->buffers[i] = std::move(uncompressed);
+    BufferPtrVector Get(const std::vector<std::shared_ptr<ArrayData>>& fields) && {
+      AppendFrom(fields);
+      return std::move(buffers_);
     }
-    return Status::OK();
+
+    BufferPtrVector buffers_;
   };
 
+  // flatten all buffers
+  auto buffers = BufferAccumulator{}.Get(*fields);

Review comment:
       Accumulating all buffers to decompress should provide better utilization of a thread pool. Previously, in the case of a single nested field all but one thread would be idle regardless of how many buffers needed to be decompressed.




----------------------------------------------------------------
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.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org