alibaba / hiactor Goto Github PK

When building multi-platform docker image with Hiactor v0.1.1 installed, error occurred on arm64 arch.

docker buildx build --platform linux/amd64,linux/arm64 -f interactive-base.Dockerfile -t registry.cn-hongkong.aliyuncs.com/graphscope/interactive-base --push .

# install hiactor
RUN cd /tmp && git clone https://github.com/alibaba/hiactor.git -b v0.1.1 --single-branch && cd hiactor && \
    git submodule update --init --recursive && ./seastar/seastar/install-dependencies.sh && mkdir build && cd build && \
    cmake -DCMAKE_INSTALL_PREFIX=/opt/flex -DHiactor_DEMOS=OFF -DHiactor_TESTING=OFF -DHiactor_DPDK=OFF -DHiactor_CXX_DIALECT=gnu++17 -DSeastar_CXX_FLAGS="-DSEASTAR_DEFAULT_ALLOCATOR -mno-avx512" .. && \
    make -j && make install && rm -rf /tmp/hiactor

building with gcc7 will result in the following error:

[  4%] Building CXX object build/seastar/CMakeFiles/seastar.dir/src/core/future.cc.o
/opt/hiactor/build/seastar/src/core/future.cc:105:50: error: explicit qualification in declaration of 'seastar::future<> seastar::internal::current_exception_as_future()'
 future<> internal::current_exception_as_future() noexcept;
                                                  ^~~~~~~~
build/seastar/CMakeFiles/seastar.dir/build.make:245: recipe for target 'build/seastar/CMakeFiles/seastar.dir/src/core/future.cc.o' failed
make[2]: *** [build/seastar/CMakeFiles/seastar.dir/src/core/future.cc.o] Error 1
CMakeFiles/Makefile2:670: recipe for target 'build/seastar/CMakeFiles/seastar.dir/all' failed
make[1]: *** [build/seastar/CMakeFiles/seastar.dir/all] Error 2
Makefile:162: recipe for target 'all' failed
make: *** [all] Error 2

As I create two hqps_actor_autogen targets in my CMakeLists.txt, cmake error occurs

CMake Error at /usr/local/bin/hiactor_codegen/ActorAutoGen.cmake:42 (add_custom_target):
  add_custom_target cannot create target "installed_libclang" because another
  target with the same name already exists.  The existing target is a custom
  target created in source directory
  "/root/GraphScope/balabala/graph_db".  See documentation for policy
  CMP0002 for more details.
Call Stack (most recent call first):
  engines/balabala/CMakeLists.txt:39 (include)

Currently, when a hiactor cluster needs to be started, the order in which the nodes are started must be strictly in accordance with their machine id order. The reason is: The current connection rule is that a node with a larger machine id will act as a client to connect to a node with a smaller machine id. If the node with a smaller machine id has not been started at this time, a connection error will occur.

When max_concurrency is set to 1, and two tasks are send to a same actor, the actor will only execute the first submitted taks, while the second one got stuck and will never be executed.

At present, when it is necessary to call an actor method across machines, hiactor will look for a local shard connected to the target machine on the current machine to forward it on its behalf.
However, the calculation of this proxy shard adopts a round-robin method, so multiple calls to an actor's method may be forwarded by different shards, which leads to an inconsistency at the dataflow level and cannot guarantee that the actor method called first will be certain executed first.

in seastar/core/httpd.cc

195   static future<std::unique_ptr<httpd::request>>
196   read_request_body(input_stream<char>& buf, std::unique_ptr<httpd::request> req) {
197     if (!req->content_length) {
198          return make_ready_future<std::unique_ptr<httpd::request>>(std::move(req));
199     }
200      return buf.read_exactly(req->content_length).then([req = std::move(req)] (temporary_buffer<char> body) mutable {
201         req->content = seastar::to_sstring(std::move(body));
202          return make_ready_future<std::unique_ptr<httpd::request>>(std::move(req));
203     });
204   }

the req->content_length is undefined at line 200, as req has been moved.