If alert syncer stops before alert manager it can crash alert manager.

This can happen if manager receives and processes syncer client disconnecting event. It will release syncer client which will free it and then it will attempt to close the syncer client later.

Mangers (not just alert, but also others) store they worker client connection without incrementing its reference counter, which is wrong. If the client connection is kept then its reference counter must be incremented too. It might be that alerting is the only place where worker properly closes its connection on shutdown and manager deinitializes all clients causing this crash, but it should be fixed in all managers.

Need to investigate if this is reproducible in earlier versions - it might be that alert syncer or manager processes are just killed without properly stopping them.