I have improved the situation for the common "Client is not started" error.
The error is actually an error in Curator library, in its usage of Zookeeper connection.
To keep the impacts to the minimum this is what I have done:
- modified a Curator class that now lives in our codebase to suppress the log entry each time the error appeared. Now it logs at DEBUG level
- Improved the handling in some scenarios where the error is actually a possible outcome, like in FabricWebRegistrationHandler that needs to unpublish a servlet from ZK registry, but it cannot do it if the connection to zk is not available.
The issue in FabricWebRegistrationHandler has just currently be wrapped but not entirely solved, since I'm only wrapping the exception in warn log call, but the underlying problem of having an invalid endpoint published might still be there.
I'll treat this possibility as a separate issue.