Open SimonHeybrock opened 8 years ago
The way this should be implemented basically depends on whether or not we can accept (1) latency and (2) blocking behavior.
ParameterControlServer
can request the info from all ranks via the heartbeat and wait for the reply.The latter option might actually be one of the easier ones. The getter would be trivial, we only need to implement the gathering. But how to define what is to be included in the status gather?
Basically we need to implement a gather of information from all ranks (via
BackendHeartbeat
). The root rank is then exposing this via theParameterControlServer
.We need this in particular for monitoring queue status and memory consumption.