Fixed executor conflict #126
Conversation
Signed-off-by: Bitterisland6 <bitterisland6@gmail.com>
Signed-off-by: Bitterisland6 <bitterisland6@gmail.com>
|
CI:
|
|
I'm not sure if overriding the rclpy method is the best way to go here, is there any other method that could serve this purpose, implementing |
| event = Event() | ||
| future = client.call_async(request) | ||
| rclpy.spin_until_future_complete(self._node, future, None, timeout) | ||
| future.add_done_callback(lambda _: event.set()) | ||
|
|
||
| event.wait(timeout) |
There was a problem hiding this comment.
This is what client.call() does. I'd recommend this instead:
try:
return client.call(request)
except:
raise AsyncServiceCallFailed(hint='the target node may not be spinning')
There was a problem hiding this comment.
@Bitterisland6 Would you prefer rlcpy client class function that Shane recommended or keep using python events?
There was a problem hiding this comment.
@quarkytale the problem with solution recommended by Shane is that in the call method the waiting for event is started without timeout, so there is a risk that it'll never end. Therefore I'd prefer to keep on using python events.
There was a problem hiding this comment.
Okay I see, using the rclpy function would be preferred but I don't see any reason to block this. Thanks for your patience, I'll run CI.
|
Any update on this? |
|
CI:
|
|
@quarkytale Could you please backport it to humble and iron? |
|
@Mergifyio backport iron humble |
✅ Backports have been created
|
* fixed executor conflict Signed-off-by: Bitterisland6 <bitterisland6@gmail.com> * fixed formatting errors Signed-off-by: Bitterisland6 <bitterisland6@gmail.com> --------- Signed-off-by: Bitterisland6 <bitterisland6@gmail.com> (cherry picked from commit e199a0d)
* fixed executor conflict Signed-off-by: Bitterisland6 <bitterisland6@gmail.com> * fixed formatting errors Signed-off-by: Bitterisland6 <bitterisland6@gmail.com> --------- Signed-off-by: Bitterisland6 <bitterisland6@gmail.com> (cherry picked from commit e199a0d)
Fixed executor conflict (#126) * fixed executor conflict Signed-off-by: Bitterisland6 <bitterisland6@gmail.com> * fixed formatting errors Signed-off-by: Bitterisland6 <bitterisland6@gmail.com> --------- Signed-off-by: Bitterisland6 <bitterisland6@gmail.com> (cherry picked from commit e199a0d) Co-authored-by: Aleksander Szymański <bitterisland6@gmail.com>
Fixed executor conflict (#126) * fixed executor conflict Signed-off-by: Bitterisland6 <bitterisland6@gmail.com> * fixed formatting errors Signed-off-by: Bitterisland6 <bitterisland6@gmail.com> --------- Signed-off-by: Bitterisland6 <bitterisland6@gmail.com> (cherry picked from commit e199a0d) Co-authored-by: Aleksander Szymański <bitterisland6@gmail.com>
Rqt spawns a multi threaded executor, and provides all open plugins in a single node which is spun inside it. But in the
rqt_reconfigureplugin when calling service, there was added waiting for the end of the call withspin_until_future_complete. This method was giving the control of the rqt to the default rclpy executor.This issue caused problems like:
To fix this issue it's enough to wait for the service request without the
spin_until_future_completemethod. One solution is to use python events andadd_done_callbackmethod from the future object returned bycall_asyncmethod. The waiting is done with eventwaitfunction and provided timeout, and can be interrupted by setting the event flag in the callback specified in theadd_done_callback.