Resolve "Agent panics and crashes when Kubernetes is not available"
The problem was that nilableCfg
channel was still assigned to a closed
w.cfg2module
channel after a leader module stop (via w.stop()
).
This caused an attempt to either send nil
or an agent configuration
to the nilableCfg
channel which was nil
at that point causing the
panic.
This problem only appears for leader modules that return from Run
before it's signaled to stop (known one is the remote_development
module) - which is a breach of contract, but
anyways, agentk musn't panic in that case and gracefully continue to
run.
Closes #434 (closed)
Edited by Timo Furrer