Improve disconnection handling

Samuel GAIST requested to merge improve_disconnection_handling into master

When reloading a worker node, the broker will notice after the usual timeout that it was gone. This is the same as if a node loses its network connection or gets killed.

This wasn't an issue before because the restart of the worker script was "slow" due to the enumeration of all BEAT related docker images. If a node uses the new labeled images, the restart is way faster and thus the broker will have an invalid state.

This patch fixes that by sending a notification to the broker that the worker node is disconnecting. This will make the broker cleanup the entry and have it ready once the worker is back online.

Merge request reports