feat(functions): Add support for REST based remote functions #10911

Joe-Abraham · 2024-09-02T05:36:22Z

netlify · 2024-09-02T05:36:42Z

✅ Deploy Preview for meta-velox canceled.

Name	Link
🔨 Latest commit	`305b0a1`
🔍 Latest deploy log	https://app.netlify.com/sites/meta-velox/deploys/67a44f92a4ff1b00086ffbea

pedroerp · 2024-09-09T15:19:16Z

Pretty cool! I see the PR is still as draft, but I can help review when it's ready. Would also be nice to add some documentation on how to use it, the configs parameters, etc.

assignUser · 2025-01-21T03:16:37Z

CMakeLists.txt

+  set(cpr_SOURCE BUNDLED)
+  velox_resolve_dependency(cpr)


#10911 (comment)

velox/functions/remote/client/Remote.cpp

CMakeLists.txt

velox/functions/remote/client/Remote.cpp

velox/functions/remote/client/RestClient.h

czentgr · 2025-02-03T20:21:47Z

velox/functions/remote/client/Remote.cpp

+      // Because location_ is a variant, we must get the string:
+      const auto& url = boost::get<std::string>(location_);
+      const std::string fullUrl = fmt::format(
+          "{}/v1/functions/{}/{}/{}/{}",


This REST API makes an assumption about the endpoint. The backend side, is that specified somewhere and would we need to mention what this is as this must be specific to a service that is targeted.

The rest server is following this presto rest function server, Can you help me guide where to document this?

This remote function is tied to a Presto rest function server. So this isn't really generic from a Velox perspective. Velox could be used with Gluten among other engines.

Can you make the fullUrl a configuration parameter of the Remote function ?

czentgr · 2025-02-03T20:24:11Z

velox/functions/remote/client/tests/RemoteFunctionRestTest.cpp

+#include <folly/init/Init.h>
+#include <gmock/gmock.h>
+#include <gtest/gtest.h>
+#include <cstdio>


As a system include it should come before boost. This seems to not have changed yet.

czentgr · 2025-02-03T20:34:57Z

velox/functions/remote/client/tests/RemoteFunctionRestTest.cpp

+  std::unique_ptr<std::thread> serverThread_;
+  boost::asio::io_context ioc_{1};
+
+  std::string location_{("http://127.0.0.1:8321")};


Instead of hard coding the port we can get free ports for use in the test.
See function getFreePorts in the test utils.

Thanks for the feedback, updated the code accordingly

czentgr · 2025-02-03T20:39:21Z

velox/functions/remote/server/CMakeLists.txt

+  velox_memory
+  velox_functions_prestosql)
+
+add_executable(velox_functions_remote_server_rest_main


Do we need this executable?
The test uses a separate thread to create the listener. Am I missing something?

Yes, you are correct there is not actual need for this executable unless someone want to run a local rest server with buildin velox function registered. I have removed the executable. I was just trying to keep the rest based remote functions as similar to the thrift implementation

czentgr · 2025-02-03T20:41:11Z

velox/functions/remote/server/RemoteFunctionServiceRestMain.cpp

+
+DEFINE_string(service_host, "127.0.0.1", "Host to bind the service to");
+
+DEFINE_int32(service_port, 8321, "Port to bind the service to");


If we need this we can pick a free port to use based on the port utils.

I have removed this code.

aditi-pandit · 2025-02-05T08:04:46Z

velox/functions/remote/client/Remote.cpp

@@ -71,6 +112,50 @@ class RemoteFunction : public exec::VectorFunction {
  }

 private:
+  void applyRestRemote(
+      const SelectivityVector& rows,
+      std::vector<VectorPtr>& args,


const& for args.

Thanks for the feedback, updated the code accordingly

aditi-pandit · 2025-02-05T08:06:34Z

velox/functions/remote/client/Remote.cpp

+
+      // Because location_ is a variant, we must get the string:
+      const auto& url = boost::get<std::string>(location_);
+      const std::string fullUrl = fmt::format(


Can you do fullUrl setup once during construction instead of for each apply ?

Under the current design, it isn’t feasible to construct the fullUrl once in the constructor and reuse it. The fullUrl encapsulates not just the base path, but also function-specific details that can change over time. By hard-coding it in the constructor, we lose the ability to dynamically add or modify functions. Keeping the fullUrl generation flexible ensures it can adapt to new or updated functions without requiring changes to the constructor logic.

Ah... So then it means that the RemoteFunction isn't really tied to a particular function but can be used for any function ? Please document that. Then there isn't a need for functionName_ member variable.

Don't think the Remote function is designed this way. The Thrift part is assuming the functionName is fixed in member variable functionName_.

The RemoteFunction is a VectorFunction meaning that it is invoked in the context of an expression. This is limited to a single fragment execution. If the user wants to use another function, the query expression would be different and so another RemoteFunction object would be constructed. It will not be this same one.

Co-authored-by: Wills Feng <[email protected]>

aditi-pandit · 2025-02-05T19:14:05Z

velox/functions/remote/client/CMakeLists.txt

 velox_add_library(velox_functions_remote Remote.cpp)
 velox_link_libraries(
  velox_functions_remote
  PUBLIC velox_expression
+         velox_memory


Its a bit strange that these Velox exec dependencies came in because of REST. Can you remove them and check what happens ?

aditi-pandit · 2025-02-05T19:20:15Z

velox/functions/remote/server/CMakeLists.txt

+         velox_functions_remote_get_serde
+         velox_functions_remote_utils
+         velox_type_fbhive
+         velox_memory)

 add_executable(velox_functions_remote_server_main RemoteFunctionServiceMain.cpp)


Can you add an executable for RestRemoteFunctionService as well or enhance the code in this main for Rest service ?

aditi-pandit · 2025-02-05T19:32:18Z

velox/functions/remote/client/Remote.cpp

+
+      // Because location_ is a variant, we must get the string:
+      const auto& url = boost::get<std::string>(location_);
+      const std::string fullUrl = fmt::format(


Ah... So then it means that the RemoteFunction isn't really tied to a particular function but can be used for any function ? Please document that. Then there isn't a need for functionName_ member variable.

aditi-pandit · 2025-02-05T19:34:20Z

velox/functions/remote/client/Remote.cpp

@@ -62,7 +99,11 @@ class RemoteFunction : public exec::VectorFunction {
      exec::EvalCtx& context,
      VectorPtr& result) const override {
    try {
-      applyRemote(rows, args, outputType, context, result);
+      if ((metadata_.location.type() == typeid(SocketAddress))) {


There isn't a need to do this typing at this point then. You can write a condition based on whether you have a thriftClient_ or restClient_.

aditi-pandit · 2025-02-05T23:56:43Z

velox/functions/remote/client/Remote.cpp

+      // Because location_ is a variant, we must get the string:
+      const auto& url = boost::get<std::string>(location_);
+      const std::string fullUrl = fmt::format(
+          "{}/v1/functions/{}/{}/{}/{}",


This remote function is tied to a Presto rest function server. So this isn't really generic from a Velox perspective. Velox could be used with Gluten among other engines.

Can you make the fullUrl a configuration parameter of the Remote function ?

aditi-pandit · 2025-02-06T03:07:03Z

velox/functions/remote/utils/CMakeLists.txt

@@ -12,7 +12,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.

-add_library(velox_functions_remote_utils RemoteFunctionServiceProvider.cpp)
+add_library(velox_functions_remote_utils RemoteFunctionHelper.h


.h files are not compiled in add_library. Its better to split this into a .h and .cpp file.

aditi-pandit · 2025-02-06T03:08:00Z

velox/functions/remote/server/RemoteFunctionRestService.h

@@ -0,0 +1,92 @@
+/*


Can you also add a program with a main class that starts a Rest Function server (similar to that for Thrift in this directory) ?

aditi-pandit · 2025-02-06T03:12:18Z

velox/functions/remote/server/RemoteFunctionRestService.h

+/// @brief Listens for incoming TCP connections and creates sessions.
+/// Sets up a TCP acceptor to listen for client connections,
+/// creating a new session for each accepted connection.
+class listener : public std::enable_shared_from_this<listener> {


I found https://medium.com/@AlexanderObregon/building-restful-apis-with-c-4c8ac63fe8a7 that has similar code.

Can we provide that as an example of implementing a Rest server with boost::beast library ?

aditi-pandit · 2025-02-06T03:16:33Z

velox/functions/remote/server/RemoteFunctionRestService.cpp

+  std::string returnType;
+};
+
+std::map<std::string, InternalFunctionSignature> internalFunctionSignatureMap =


Don't think we should be doing such manual registration. The Thrift service is registering all PrestoSQL scalar functions. You can do the same for this Rest service for a start. But we should enhance this to use the dynamic function loading using #11439

aditi-pandit · 2025-02-06T03:20:04Z

velox/functions/remote/server/RemoteFunctionRestService.cpp

+    const auto& functionSignature =
+        internalFunctionSignatureMap.at(functionName);
+
+    auto inputType = deserializeArgTypes(functionSignature.argumentTypes);


The Thrift and Rest servers would have such an implementation. Can you abstract a common function for this ?

aditi-pandit · 2025-02-06T03:24:34Z

velox/functions/remote/client/tests/RemoteFunctionRestTest.cpp

+        {remotePrefix_ + ".remote_substr"});
+  }
+
+  void initializeServer(uint16_t servicePort) {


Might be better to abstract a RestFunctionServiceRunner class for this.

facebook-github-bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Sep 2, 2024

Joe-Abraham changed the title ~~Add support for REST based remote functions~~ [WIP] Add support for REST based remote functions Sep 2, 2024

Joe-Abraham force-pushed the udf branch 12 times, most recently from b88d136 to b85e0e6 Compare September 4, 2024 09:32

Yuhta requested review from pedroerp and mbasmanova September 4, 2024 18:28

Joe-Abraham force-pushed the udf branch 6 times, most recently from 0cd4510 to 74023dc Compare September 9, 2024 08:10

Joe-Abraham force-pushed the udf branch 3 times, most recently from abe87e1 to 6c1606e Compare September 13, 2024 05:06

Joe-Abraham force-pushed the udf branch 3 times, most recently from 05115f4 to 2ffec26 Compare September 20, 2024 05:18

Joe-Abraham force-pushed the udf branch from 0efa665 to 36f7abb Compare January 20, 2025 04:14

assignUser requested changes Jan 21, 2025

View reviewed changes

CMakeLists.txt

Comment on lines +524 to +527

set(cpr_SOURCE BUNDLED)

velox_resolve_dependency(cpr)

Copy link

Collaborator

assignUser Jan 21, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

#10911 (comment)

Joe-Abraham force-pushed the udf branch 4 times, most recently from f69837a to 11c7ad2 Compare January 29, 2025 05:58

Joe-Abraham requested a review from assignUser January 29, 2025 05:59

Joe-Abraham force-pushed the udf branch 4 times, most recently from 32b272b to 322d042 Compare January 30, 2025 07:35

Joe-Abraham requested a review from aditi-pandit January 30, 2025 07:37

Joe-Abraham force-pushed the udf branch 3 times, most recently from a312e97 to c8e8424 Compare February 3, 2025 12:15

czentgr reviewed Feb 3, 2025

View reviewed changes

aditi-pandit reviewed Feb 5, 2025

View reviewed changes

Joe-Abraham force-pushed the udf branch 4 times, most recently from bb18b7e to 7cdcaae Compare February 5, 2025 12:45

Joe-Abraham requested review from aditi-pandit and czentgr February 5, 2025 12:46

Joe-Abraham force-pushed the udf branch from 7cdcaae to 4950b9f Compare February 5, 2025 15:54

feat(functions): Add support for REST based remote functions

8f454f0

Co-authored-by: Wills Feng <[email protected]>

Joe-Abraham force-pushed the udf branch from 4950b9f to 8f454f0 Compare February 5, 2025 16:05

aditi-pandit reviewed Feb 6, 2025

View reviewed changes

Joe-Abraham added 2 commits February 6, 2025 11:00

review comments - helper function

6f75362

review comments - fullURL

305b0a1

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(functions): Add support for REST based remote functions #10911

feat(functions): Add support for REST based remote functions #10911

Joe-Abraham commented Sep 2, 2024 •

edited

Loading

netlify bot commented Sep 2, 2024 •

edited

Loading

pedroerp commented Sep 9, 2024

assignUser Jan 21, 2025

czentgr Feb 3, 2025

Joe-Abraham Feb 5, 2025

aditi-pandit Feb 5, 2025

czentgr Feb 3, 2025

czentgr Feb 3, 2025

Joe-Abraham Feb 5, 2025

czentgr Feb 3, 2025

Joe-Abraham Feb 5, 2025

czentgr Feb 3, 2025

Joe-Abraham Feb 5, 2025

aditi-pandit Feb 5, 2025

Joe-Abraham Feb 5, 2025

aditi-pandit Feb 5, 2025

Joe-Abraham Feb 5, 2025

aditi-pandit Feb 5, 2025

aditi-pandit Feb 5, 2025

aditi-pandit Feb 5, 2025

aditi-pandit Feb 5, 2025

aditi-pandit Feb 5, 2025

aditi-pandit Feb 5, 2025

aditi-pandit Feb 5, 2025

aditi-pandit Feb 6, 2025

aditi-pandit Feb 6, 2025

aditi-pandit Feb 6, 2025

aditi-pandit Feb 6, 2025

aditi-pandit Feb 6, 2025

aditi-pandit Feb 6, 2025


		DEFINE_string(service_host, "127.0.0.1", "Host to bind the service to");

		DEFINE_int32(service_port, 8321, "Port to bind the service to");

feat(functions): Add support for REST based remote functions #10911

Are you sure you want to change the base?

feat(functions): Add support for REST based remote functions #10911

Conversation

Joe-Abraham commented Sep 2, 2024 • edited Loading

netlify bot commented Sep 2, 2024 • edited Loading

✅ Deploy Preview for meta-velox canceled.

pedroerp commented Sep 9, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Joe-Abraham commented Sep 2, 2024 •

edited

Loading

netlify bot commented Sep 2, 2024 •

edited

Loading