the messaging plugin as well

timeouts to the name used by Apache Kafka

partitioner callback

appropriate -- the additional WARNING in MQ_Send() is superfluous

unrecoverable (since rdkafka attempts recovery); just log WARNINGs

changes in the list of brokers

and add a summary stats line to the log

  error states; wrk objects have private data for error states and stats;
  stats callback emits wrk object stats

errors

This reverts commit 34058e973866ecc8e96c6f4a67257142198a4c8b.

worker number as a parameter

(but the ActiveMQ implementation does not use sharding)

           by calling MQ_WorkerShutdown() after resend/reconnect
           failures (may have been leaking threads and connections)

           interface and the ActiveMQ plugin (fixing up some obsolete
           docs along the way), and add a new README/manpage for the
           ActiveMQ plugin

on connection shutdown