[GitHub] [arrow] pitrou commented on a change in pull request #12239: ARROW-15374: [C++][FlightRPC] Add support for MemoryManager in data methods

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

pitrou commented on a change in pull request #12239:
URL: https://github.com/apache/arrow/pull/12239#discussion_r793461781



##########
File path: cpp/src/arrow/flight/flight_cuda_test.cc
##########
@@ -0,0 +1,201 @@
+// Licensed to the Apache Software Foundation (ASF) under one
+// or more contributor license agreements.  See the NOTICE file
+// distributed with this work for additional information
+// regarding copyright ownership.  The ASF licenses this file
+// to you under the Apache License, Version 2.0 (the
+// "License"); you may not use this file except in compliance
+// with the License.  You may obtain a copy of the License at
+//
+//   http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing,
+// software distributed under the License is distributed on an
+// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+// KIND, either express or implied.  See the License for the
+// specific language governing permissions and limitations
+// under the License.
+
+#include <gtest/gtest.h>
+
+#include <memory>
+
+#include "arrow/array.h"
+#include "arrow/flight/client.h"
+#include "arrow/flight/server.h"
+#include "arrow/flight/test_util.h"
+#include "arrow/gpu/cuda_api.h"
+#include "arrow/table.h"
+#include "arrow/testing/gtest_util.h"
+
+namespace arrow {
+namespace flight {
+
+Status CheckBuffersOnDevice(const Array& array, const Device& device) {
+  for (const auto& buffer : array.data()->buffers) {
+    if (!buffer) continue;
+    if (!buffer->device()->Equals(device)) {
+      return Status::Invalid("Expected buffer on device: ", device.ToString(),
+                             ". Was allocated on device: ", buffer->device()->ToString());
+    }
+  }

Review comment:
       Should we also recurse into child data here?

##########
File path: cpp/src/arrow/flight/flight_cuda_test.cc
##########
@@ -0,0 +1,201 @@
+// Licensed to the Apache Software Foundation (ASF) under one
+// or more contributor license agreements.  See the NOTICE file
+// distributed with this work for additional information
+// regarding copyright ownership.  The ASF licenses this file
+// to you under the Apache License, Version 2.0 (the
+// "License"); you may not use this file except in compliance
+// with the License.  You may obtain a copy of the License at
+//
+//   http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing,
+// software distributed under the License is distributed on an
+// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+// KIND, either express or implied.  See the License for the
+// specific language governing permissions and limitations
+// under the License.
+
+#include <gtest/gtest.h>
+
+#include <memory>
+
+#include "arrow/array.h"
+#include "arrow/flight/client.h"
+#include "arrow/flight/server.h"
+#include "arrow/flight/test_util.h"
+#include "arrow/gpu/cuda_api.h"
+#include "arrow/table.h"
+#include "arrow/testing/gtest_util.h"
+
+namespace arrow {
+namespace flight {
+
+Status CheckBuffersOnDevice(const Array& array, const Device& device) {
+  for (const auto& buffer : array.data()->buffers) {
+    if (!buffer) continue;
+    if (!buffer->device()->Equals(device)) {
+      return Status::Invalid("Expected buffer on device: ", device.ToString(),
+                             ". Was allocated on device: ", buffer->device()->ToString());
+    }
+  }
+  return Status::OK();
+}
+
+class CudaTestServer : public FlightServerBase {
+ public:
+  explicit CudaTestServer(std::shared_ptr<Device> device) : device_(std::move(device)) {}
+
+  Status DoGet(const ServerCallContext&, const Ticket&,
+               std::unique_ptr<FlightDataStream>* data_stream) override {
+    BatchVector batches;
+    RETURN_NOT_OK(ExampleIntBatches(&batches));
+    auto batch_reader = std::make_shared<BatchIterator>(batches[0]->schema(), batches);
+    *data_stream = std::unique_ptr<FlightDataStream>(new RecordBatchStream(batch_reader));
+    return Status::OK();
+  }
+
+  Status DoPut(const ServerCallContext&, std::unique_ptr<FlightMessageReader> reader,
+               std::unique_ptr<FlightMetadataWriter> writer) override {
+    BatchVector batches;
+    RETURN_NOT_OK(reader->ReadAll(&batches));
+    for (const auto& batch : batches) {
+      for (const auto& column : batch->columns()) {
+        RETURN_NOT_OK(CheckBuffersOnDevice(*column, *device_));
+      }
+    }
+    return Status::OK();
+  }
+
+  Status DoExchange(const ServerCallContext& context,
+                    std::unique_ptr<FlightMessageReader> reader,
+                    std::unique_ptr<FlightMessageWriter> writer) override {
+    FlightStreamChunk chunk;
+    bool begun = false;
+    while (true) {
+      RETURN_NOT_OK(reader->Next(&chunk));
+      if (!chunk.data) break;
+      if (!begun) {
+        begun = true;
+        RETURN_NOT_OK(writer->Begin(chunk.data->schema()));
+      }
+      for (const auto& column : chunk.data->columns()) {
+        RETURN_NOT_OK(CheckBuffersOnDevice(*column, *device_));
+      }
+      RETURN_NOT_OK(writer->WriteRecordBatch(*chunk.data));
+    }
+    return Status::OK();
+  }
+
+ private:
+  std::shared_ptr<Device> device_;
+};
+
+class TestCuda : public ::testing::Test {
+ public:
+  void SetUp() {
+    ASSERT_OK_AND_ASSIGN(manager_, cuda::CudaDeviceManager::Instance());
+    ASSERT_OK_AND_ASSIGN(device_, manager_->GetDevice(0));
+    ASSERT_OK_AND_ASSIGN(context_, device_->GetContext());
+
+    ASSERT_OK(MakeServer<CudaTestServer>(
+        &server_, &client_,
+        [this](FlightServerOptions* options) {
+          options->memory_manager = device_->default_memory_manager();
+          return Status::OK();
+        },
+        [](FlightClientOptions* options) { return Status::OK(); }, device_));
+  }
+  void TearDown() { ASSERT_OK(server_->Shutdown()); }
+
+ protected:
+  cuda::CudaDeviceManager* manager_;
+  std::shared_ptr<cuda::CudaDevice> device_;
+  std::shared_ptr<cuda::CudaContext> context_;
+
+  std::unique_ptr<FlightClient> client_;
+  std::unique_ptr<FlightServerBase> server_;
+};
+
+TEST_F(TestCuda, DoGet) {
+  // Check that we can allocate the results of DoGet with a custom
+  // memory manager.
+  FlightCallOptions options;
+  options.memory_manager = device_->default_memory_manager();
+
+  Ticket ticket{""};
+  std::unique_ptr<FlightStreamReader> stream;
+  ASSERT_OK(client_->DoGet(options, ticket, &stream));
+  std::shared_ptr<Table> table;
+  ASSERT_OK(stream->ReadAll(&table));
+
+  for (const auto& column : table->columns()) {
+    for (const auto& chunk : column->chunks()) {
+      ASSERT_OK(CheckBuffersOnDevice(*chunk, *device_));
+    }
+  }
+}
+
+TEST_F(TestCuda, DoPut) {
+  // Check that we can send a record batch containing references to
+  // GPU buffers.
+  BatchVector batches;
+  ASSERT_OK(ExampleIntBatches(&batches));
+
+  std::unique_ptr<FlightStreamWriter> writer;
+  std::unique_ptr<FlightMetadataReader> reader;
+  auto descriptor = FlightDescriptor::Path({""});
+  ASSERT_OK(client_->DoPut(descriptor, batches[0]->schema(), &writer, &reader));
+
+  ipc::DictionaryMemo memo;
+  for (const auto& batch : batches) {
+    ASSERT_OK_AND_ASSIGN(auto buffer, cuda::SerializeRecordBatch(*batch, context_.get()));
+    ASSERT_OK_AND_ASSIGN(auto cuda_batch,
+                         cuda::ReadRecordBatch(batch->schema(), &memo, buffer));
+
+    for (const auto& column : cuda_batch->columns()) {
+      ASSERT_OK(CheckBuffersOnDevice(*column, *device_));
+    }
+
+    ASSERT_OK(writer->WriteRecordBatch(*cuda_batch));
+  }
+  ASSERT_OK(writer->Close());
+}
+
+TEST_F(TestCuda, DoExchange) {
+  // Check that we can send a record batch containing references to
+  // GPU buffers.
+  FlightCallOptions options;
+  options.memory_manager = device_->default_memory_manager();
+
+  BatchVector batches;
+  ASSERT_OK(ExampleIntBatches(&batches));
+
+  std::unique_ptr<FlightStreamWriter> writer;
+  std::unique_ptr<FlightStreamReader> reader;
+  auto descriptor = FlightDescriptor::Path({""});
+  ASSERT_OK(client_->DoExchange(options, descriptor, &writer, &reader));
+  ASSERT_OK(writer->Begin(batches[0]->schema()));
+
+  ipc::DictionaryMemo memo;
+  for (const auto& batch : batches) {
+    ASSERT_OK_AND_ASSIGN(auto buffer, cuda::SerializeRecordBatch(*batch, context_.get()));
+    ASSERT_OK_AND_ASSIGN(auto cuda_batch,
+                         cuda::ReadRecordBatch(batch->schema(), &memo, buffer));
+
+    for (const auto& column : cuda_batch->columns()) {
+      ASSERT_OK(CheckBuffersOnDevice(*column, *device_));
+    }
+
+    ASSERT_OK(writer->WriteRecordBatch(*cuda_batch));
+
+    FlightStreamChunk chunk;
+    ASSERT_OK(reader->Next(&chunk));
+    for (const auto& column : chunk.data->columns()) {

Review comment:
       Should we also check the contents of the returned data here?

##########
File path: cpp/src/arrow/flight/serialization_internal.cc
##########
@@ -272,7 +282,13 @@ grpc::Status FlightDataSerialize(const FlightPayload& msg, ByteBuffer* out,
         // entries are invalid.
         if (!buffer) continue;
 
-        slices.push_back(SliceFromBuffer(buffer));
+        grpc::Slice slice;
+        auto status = SliceFromBuffer(buffer).Value(&slice);
+        if (ARROW_PREDICT_FALSE(!status.ok())) {
+          // This will likely lead to abort as gRPC cannot recover from an error here
+          return ToGrpcStatus(status);
+        }
+        slices.push_back(slice);

Review comment:
       (not sure how `grpc::Slice` behaves wrt copies)
   
   ```suggestion
           slices.push_back(std::move(slice));
   ```

##########
File path: cpp/src/arrow/flight/flight_cuda_test.cc
##########
@@ -0,0 +1,201 @@
+// Licensed to the Apache Software Foundation (ASF) under one
+// or more contributor license agreements.  See the NOTICE file
+// distributed with this work for additional information
+// regarding copyright ownership.  The ASF licenses this file
+// to you under the Apache License, Version 2.0 (the
+// "License"); you may not use this file except in compliance
+// with the License.  You may obtain a copy of the License at
+//
+//   http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing,
+// software distributed under the License is distributed on an
+// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+// KIND, either express or implied.  See the License for the
+// specific language governing permissions and limitations
+// under the License.
+
+#include <gtest/gtest.h>
+
+#include <memory>
+
+#include "arrow/array.h"
+#include "arrow/flight/client.h"
+#include "arrow/flight/server.h"
+#include "arrow/flight/test_util.h"
+#include "arrow/gpu/cuda_api.h"
+#include "arrow/table.h"
+#include "arrow/testing/gtest_util.h"
+
+namespace arrow {
+namespace flight {
+
+Status CheckBuffersOnDevice(const Array& array, const Device& device) {
+  for (const auto& buffer : array.data()->buffers) {
+    if (!buffer) continue;
+    if (!buffer->device()->Equals(device)) {
+      return Status::Invalid("Expected buffer on device: ", device.ToString(),
+                             ". Was allocated on device: ", buffer->device()->ToString());
+    }
+  }

Review comment:
       Ok, I realize the test might only use primitive data, in which case it doesn't matter.




-- 
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: github-unsubscribe@arrow.apache.org

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