log level is DEBUG

(because the ActiveMQ implementation was crashing when it tried to look
up the number itself)

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)