The connection's memory wasn't getting free'd which caused issues when
the connection needed to be restablished.

data_size is used when sending buffer messages (e.g. RESPONSE_JOB_INFO).

This protects the fed_mgr_fed_rec.

Reverted 315cff15. This is a cleaner patch.

This reverts commit 315cff15.

This reverts commit e5f2c720.

Continuation of 315cff15

Explicitly test WIFEXITED() rather than assume an exit code if WIFSIGNALED is false.
bug 3562

This will avoid filling the slurmctld logs with communication error
  messages when a cluster in the federation is down.

bug 3610

No change to logic

Fix some recently introduced memory leaks related to lists of RPCs.

Correct spelling of persistent in various comments and log messages.
No changed to logic.

Log the RPCs queued for processing by the federation agent
which remain unprocessed at slurmctld shutdown

The broadcast was temporarily removed for testing purposes

# Conflicts:
#	src/slurmctld/proc_req.c

Bug 3097

Value stored to 'rc' is never read.

Add message type strings and job IDs where possible

Failed RPCs get requeued and retried with exponential backup.
Repeatedly failing RPCs get logged (need to improve that).

Fix correct macro this time, Tim Wickberg's fault for prior failed patch.

Bug 3605.

This reverts commit 53523c54.

Bug 3605.

We were changing the buffers' processed offset to the beginning of the
body part of the offset to pack just the body part of the buffer but
once this happened we lost how much actual data was in the buffer. So we
were packing the whole buffer which could be larger than what was
actually being used in the buffer.

Changed job id print formats from %d to %u in several places