[DNR] feat(native): Extend sidecar to support GPU execution

.github/workflows/prestocpp-linux-build-and-unit-test.yml

@@ -353,7 +353,7 @@ jobs:
           github.event_name == 'schedule' || needs.changes.outputs.codechange == 'true'
         run: |
           export PRESTO_SERVER_PATH="${GITHUB_WORKSPACE}/presto-native-execution/_build/release/presto_cpp/main/presto_server"
-          export TESTFILES=`find ./presto-native-tests/src/test -type f -name 'Test*.java'`
+          export TESTFILES=`find ./presto-native-tests/src/test -type f -name 'Test*.java' | grep -v 'cudf/'`
           # Convert file paths to comma separated class names
           export TESTCLASSES=
           for test_file in $TESTFILES

.gitmodules

@@ -1,3 +1,3 @@
 [submodule "presto-native-execution/velox"]
 	path = presto-native-execution/velox
-	url = https://github.com/facebookincubator/velox.git
+	url = https://github.com/pramodsatya/velox.git

presto-native-execution/etc_sidecar/config.properties

@@ -1,7 +1,10 @@
-discovery.uri=http://127.0.0.1:<replace_port>
+discovery.uri=http://presto-coordinator:8080
 presto.version=testversion
 http-server.http.port=7778
 shutdown-onset-sec=1
 runtime-metrics-collection-enabled=true
 native-sidecar=true
 presto.default-namespace=native.default
+cudf.enabled=true
+cudf.debug_enabled=true
+cudf.allow_cpu_fallback=true

presto-native-execution/etc_sidecar/node.properties

@@ -1,3 +1,3 @@
-node.environment=testing
-node.internal-address=127.0.0.1
+node.environment=test
+node.internal-address=172.24.0.3
 node.location=testing-location

presto-native-execution/presto_cpp/main/CMakeLists.txt

@@ -17,10 +17,7 @@ add_subdirectory(thrift)
 add_subdirectory(connectors)
 add_subdirectory(functions)
 add_subdirectory(tool)
-
-add_library(presto_session_properties SessionProperties.cpp)
-
-target_link_libraries(presto_session_properties ${FOLLY_WITH_DEPENDENCIES})
+add_subdirectory(sidecar)
 
 add_library(
   presto_server_lib
@@ -112,7 +109,12 @@ target_link_libraries(
 )
 
 if(PRESTO_ENABLE_CUDF)
-  target_link_libraries(presto_server_lib velox_cudf_exec)
+  target_link_libraries(
+    presto_server_lib
+    presto_cudf_function_metadata
+    presto_cudf_session_properties
+    velox_cudf_exec
+  )
 endif()
 
 # Enabling Parquet causes build errors with missing symbols on MacOS. This is

presto-native-execution/presto_cpp/main/PrestoServer.cpp

@@ -22,7 +22,6 @@
 #include "presto_cpp/main/CoordinatorDiscoverer.h"
 #include "presto_cpp/main/PeriodicMemoryChecker.h"
 #include "presto_cpp/main/PeriodicTaskManager.h"
-#include "presto_cpp/main/SessionProperties.h"
 #include "presto_cpp/main/SignalHandler.h"
 #include "presto_cpp/main/TaskResource.h"
 #include "presto_cpp/main/common/ConfigReader.h"
@@ -31,7 +30,6 @@
 #include "presto_cpp/main/connectors/Registration.h"
 #include "presto_cpp/main/connectors/SystemConnector.h"
 #include "presto_cpp/main/connectors/hive/functions/HiveFunctionRegistration.h"
-#include "presto_cpp/main/functions/FunctionMetadata.h"
 #include "presto_cpp/main/http/HttpConstants.h"
 #include "presto_cpp/main/http/filters/AccessLogFilter.h"
 #include "presto_cpp/main/http/filters/HttpEndpointLatencyFilter.h"
@@ -44,6 +42,8 @@
 #include "presto_cpp/main/operators/ShuffleExchangeSource.h"
 #include "presto_cpp/main/operators/ShuffleRead.h"
 #include "presto_cpp/main/operators/ShuffleWrite.h"
+#include "presto_cpp/main/sidecar/function/NativeFunctionMetadata.h"
+#include "presto_cpp/main/sidecar/properties/SessionProperties.h"
 #include "presto_cpp/main/types/ExpressionOptimizer.h"
 #include "presto_cpp/main/types/PrestoToVeloxQueryPlan.h"
 #include "presto_cpp/main/types/VeloxPlanConversion.h"
@@ -77,6 +77,8 @@
 #include "velox/serializers/UnsafeRowSerializer.h"
 
 #ifdef PRESTO_ENABLE_CUDF
+#include "presto_cpp/main/sidecar/function/CudfFunctionMetadata.h"
+#include "presto_cpp/main/sidecar/properties/CudfSessionProperties.h"
 #include "velox/experimental/cudf/CudfConfig.h"
 #include "velox/experimental/cudf/exec/ToCudf.h"
 #endif
@@ -1817,32 +1819,62 @@ void PrestoServer::handleGracefulShutdown(
 
 void PrestoServer::registerSidecarEndpoints() {
   VELOX_CHECK(httpServer_);
+
   httpServer_->registerGet(
       "/v1/properties/session",
-      [this](
+      [](
           proxygen::HTTPMessage* /*message*/,
           const std::vector<std::unique_ptr<folly::IOBuf>>& /*body*/,
           proxygen::ResponseHandler* downstream) {
+#ifdef PRESTO_ENABLE_CUDF
+        const auto* sessionProperties =
+            facebook::presto::cudf::CudfSessionProperties::instance();
+        http::sendOkResponse(downstream, sessionProperties->serialize());
+#else
         const auto* sessionProperties = SessionProperties::instance();
         http::sendOkResponse(downstream, sessionProperties->serialize());
+#endif
       });
+
   httpServer_->registerGet(
       "/v1/functions",
-      [](proxygen::HTTPMessage* /*message*/,
-         const std::vector<std::unique_ptr<folly::IOBuf>>& /*body*/,
-         proxygen::ResponseHandler* downstream) {
-        http::sendOkResponse(downstream, getFunctionsMetadata());
+        [](
+          proxygen::HTTPMessage* /*message*/,
+          const std::vector<std::unique_ptr<folly::IOBuf>>& /*body*/,
+          proxygen::ResponseHandler* downstream) {
+#ifdef PRESTO_ENABLE_CUDF
+        http::sendOkResponse(
+            downstream,
+            presto::cudf::cudfFunctionMetadataProvider().getFunctionsMetadata(
+                std::nullopt));
+#else
+        http::sendOkResponse(
+            downstream,
+          nativeFunctionMetadata().getFunctionsMetadata(
+                std::nullopt));
+#endif
       });
+
   httpServer_->registerGet(
       R"(/v1/functions/([^/]+))",
-      [](proxygen::HTTPMessage* /*message*/,
-         const std::vector<std::string>& pathMatch) {
+        [](
+          proxygen::HTTPMessage* /*message*/,
+          const std::vector<std::string>& pathMatch) {
         return new http::CallbackRequestHandler(
-            [catalog = pathMatch[1]](
+          [catalog = pathMatch[1]](
                 proxygen::HTTPMessage* /*message*/,
                 std::vector<std::unique_ptr<folly::IOBuf>>& /*body*/,
                 proxygen::ResponseHandler* downstream) {
-              http::sendOkResponse(downstream, getFunctionsMetadata(catalog));
+#ifdef PRESTO_ENABLE_CUDF
+              http::sendOkResponse(
+                  downstream,
+                  presto::cudf::cudfFunctionMetadataProvider()
+                      .getFunctionsMetadata(catalog));
+#else
+              http::sendOkResponse(
+                  downstream,
+                  nativeFunctionMetadata().getFunctionsMetadata(catalog));
+#endif
             });
       });
   httpServer_->registerPost(

presto-native-execution/presto_cpp/main/PrestoToVeloxQueryConfig.cpp

@@ -13,8 +13,8 @@
  */
 
 #include "presto_cpp/main/PrestoToVeloxQueryConfig.h"
-#include "presto_cpp/main/SessionProperties.h"
 #include "presto_cpp/main/common/Configs.h"
+#include "presto_cpp/main/sidecar/properties/SessionProperties.h"
 #include "velox/common/compression/Compression.h"
 #include "velox/core/QueryConfig.h"
 #include "velox/type/tz/TimeZoneMap.h"

presto-native-execution/presto_cpp/main/QueryContextManager.cpp

@@ -15,8 +15,8 @@
 #include "presto_cpp/main/QueryContextManager.h"
 #include <folly/executors/IOThreadPoolExecutor.h>
 #include "presto_cpp/main/PrestoToVeloxQueryConfig.h"
-#include "presto_cpp/main/SessionProperties.h"
 #include "presto_cpp/main/common/Configs.h"
+#include "presto_cpp/main/sidecar/properties/SessionProperties.h"
 #include "velox/connectors/hive/HiveConfig.h"
 #include "velox/core/QueryConfig.h"
 

presto-native-execution/presto_cpp/main/common/Configs.cpp

@@ -624,6 +624,10 @@ bool SystemConfig::prestoNativeSidecar() const {
   return optionalProperty<bool>(kNativeSidecar).value();
 }
 
+std::string SystemConfig::executionMode() const {
+  return optionalProperty<std::string>(kExecutionMode).value_or("cpu");
+}
+
 uint32_t SystemConfig::systemMemLimitGb() const {
   return optionalProperty<uint32_t>(kSystemMemLimitGb).value();
 }

presto-native-execution/presto_cpp/main/common/Configs.h

@@ -339,6 +339,10 @@ class SystemConfig : public ConfigBase {
   /// Indicates if the process is configured as a sidecar.
   static constexpr std::string_view kNativeSidecar{"native-sidecar"};
 
+  /// Execution mode: "cpu" for Velox/CPU execution or "cudf" for CUDF GPU
+  /// execution. Default is "cpu".
+  static constexpr std::string_view kExecutionMode{"execution-mode"};
+
   /// If true, enable memory pushback when the server is under low memory
   /// condition. This only applies if 'system-mem-limit-gb' is set.
   static constexpr std::string_view kSystemMemPushbackEnabled{
@@ -1227,6 +1231,9 @@ class SystemConfig : public ConfigBase {
 
   bool prestoNativeSidecar() const;
 
+  /// Returns the execution mode ("cpu" or "cudf"). Defaults to "cpu".
+  std::string executionMode() const;
+
   std::string prestoDefaultNamespacePrefix() const;
 
   std::string poolType() const;

presto-native-execution/presto_cpp/main/common/Utils.cpp

@@ -146,4 +146,8 @@ const std::vector<std::string> getFunctionNameParts(
       fmt::format("Prefix missing for function {}", registeredFunction));
   return parts;
 }
+
+std::string boolToString(bool value) {
+  return value ? "true" : "false";
+}
 } // namespace facebook::presto::util

presto-native-execution/presto_cpp/main/common/Utils.h

@@ -66,4 +66,7 @@ inline std::string addDefaultNamespacePrefix(
 /// three parts, {catalog, schema, function_name}, from the registered function.
 const std::vector<std::string> getFunctionNameParts(
     const std::string& registeredFunction);
+
+/// Convert boolean to lowercase string representation.
+std::string boolToString(bool value);
 } // namespace facebook::presto::util

presto-native-execution/presto_cpp/main/functions/CMakeLists.txt

@@ -10,15 +10,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-add_library(presto_function_metadata OBJECT FunctionMetadata.cpp)
-target_link_libraries(presto_function_metadata presto_common velox_function_registry)
-
 add_subdirectory(dynamic_registry)
 
 if(PRESTO_ENABLE_REMOTE_FUNCTIONS)
   add_subdirectory(remote)
 endif()
-
-if(PRESTO_ENABLE_TESTING)
-  add_subdirectory(tests)
-endif()