service::waitForService stuck in ros::Duration::sleep when use_sim_time true without /clock publisher, because using ros::Duration::sleep instead of ros::WallDuration::sleep

[alrros]

The function service::waitForService in file libros/service.cpp
uses ros::Duration::sleep instead of ros::WallDuration::sleep,
which leads to the side effect of beeing stuck in ros::Duration::sleep
under the following conditions:

• use_sim_time is set as ros parameter
• no simulated time publisher is active (no /clock publisher)
• the service provider is not yet available

This case is significantly critical, when:

• using a nodelet as /clock publisher
• starting nodelet manager and nodelet loader in 1 roslaunch
  --> in 99/100 cases everything is fine
  --> in 1/100 cases the nodelet loader gets stuck in service::waitForService -> ros::Duration::sleep

Although nodelet is used as an example, the main problem is the usage of ros::Duration::sleep at service client initialization instead of ros::WallDuration::sleep.

Quick Reproduction:
The problem can be quickly reproduced the following way by using the service "load_nodelet" of the nodelet manager as an example:
case A) everything is fine

• first: terminal 1: roscore
• then: terminal 2: rosparam set use_sim_time "true"
• then: terminal 3: rosrun nodelet nodelet manager __name:=nodelet_manager
• then: (wait 1s)
• then: terminal 4: rosrun nodelet nodelet load nodelet_tutorial_math/Plus nodelet_manager __name:=nodelet1 nodelet1/in:=foo _value:=1.1
• then: terminal 5: rosservice call /nodelet_manager/list
  ==> nodelet is loaded

case B) stuck, no nodelet loaded

• first: terminal 1: roscore
• then: terminal 2: rosparam set use_sim_time "true"
• then: terminal 3: rosrun nodelet nodelet load nodelet_tutorial_math/Plus nodelet_manager __name:=nodelet1 nodelet1/in:=foo _value:=1.1
• then: (wait 1s)
• then: terminal 4: rosrun nodelet nodelet manager __name:=nodelet_manager
• then: terminal 5: rosservice call /nodelet_manager/list
  ==> no nodelet loaded, empty list

(note: no /clock publisher is active, but ros::Duration::sleep is never executed in case A)

---

---

function

• bool service::waitForService(const std::string& service_name, ros::Duration timeout)

in file

• ros_comm/clients/roscpp/src/libros/service.cpp

uses

• ros::Duration::sleep

instead of

• ros::WallDuration::sleep

under the following conditions

• use_sim_time is set as ros parameter
• no simulated time publisher is active (no /clock publisher)
• the service provider is not yet available

causing critical side effect:

• beeing stuck in service::waitForService -> ros::Duration::sleep

[dirk-thomas]

Please consider to provide a PR with the proposed change.

[khssnv]

Faced the same in ROS Kinetic / rospy.

[cwecht]

@khssnv your problem seems to have a different cause. rospys wait_for_service function relies not on rostime (as far as I can see).