niklas/rclcpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Workaround for wait_for_service lasting the full timeout with connext (#476)

Browse source 
* Limit wait_for_graph_change timeout as alternative workaround for connext

* Increase max wait time to 100ms
This commit is contained in:
dhood 2018-05-22 21:01:47 -04:00 • committed by GitHub
parent 15d505ec1f
commit f9a78df9fe
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 12 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

17
rclcpp/src/rclcpp/client.cpp
View file

@ -14,6 +14,7 @@
#include "rclcpp/client.hpp" #include "rclcpp/client.hpp"
#include <algorithm>
#include <chrono> #include <chrono>
#include <cstdio> #include <cstdio>
#include <memory> #include <memory>
@ -132,12 +133,18 @@ ClientBase::wait_for_service_nanoseconds(std::chrono::nanoseconds timeout)
if (!rclcpp::ok()) { if (!rclcpp::ok()) {
return false; return false;
} }
node_ptr->wait_for_graph_change(event, time_to_wait); // Limit each wait to 100ms to workaround an issue specific to the Connext RMW implementation.
event->check_and_clear(); // reset the event // A race condition means that graph changes for services becoming available may trigger the
// wait set to wake up, but then not be reported as ready immediately after the wake up
// always check if the service is ready, even if the graph event wasn't triggered
// this is needed to avoid a race condition that is specific to the Connext RMW implementation
// (see https://github.com/ros2/rmw_connext/issues/201) // (see https://github.com/ros2/rmw_connext/issues/201)
// If no other graph events occur, the wait set will not be triggered again until the timeout
// has been reached, despite the service being available, so we artificially limit the wait
// time to limit the delay.
node_ptr->wait_for_graph_change(
event, std::min(time_to_wait, std::chrono::nanoseconds(RCL_MS_TO_NS(100))));
// Because of the aforementioned race condition, we check if the service is ready even if the
// graph event wasn't triggered.
event->check_and_clear();
if (this->service_is_ready()) { if (this->service_is_ready()) {
return true; return true;
} }