It turned out that the unknown error messages were produced by process 32058 that was not in ctl ps list at all. So it must have survived alive from a previous restart of K.
Question: how it is possible that a tcp worker that does not belong to the current running K instance gets to service a json rpc request?
-- Juha