When powering up a node to change it's state (e.g. KNL NUMA or MCDRAM mode)
    then pass to the ResumeProgram the job ID assigned to the nodes in the
    SLURM_JOB_ID environment variable.
bug 3100

Don't log error for job end_time being zero if node health check is still
    running.
bug 3053

Previous logic duplicated checking error_codes returned from
job_allocate. job_allocate() will set job state to FAILED if there
was an actual issue.

Was just checking for ESLURM_REQUESTED_PART_CONFIG_UNAVAILABLE and
ENFORCE_ALL however in _slurm_rpc_allocate_resources() and
_slurm_rpc_submit_batch_job() both check for ANY and ALL.

to siblings (If not already connected).
This will happen when the next message is sent to them.

sibling clusters.

back a message to the caller.

a federation connection, someone adding and removing the cluster
from the federation lots of times at the same time the cluster could
be not found.

Fixes build issue caused by 844830d4.

connection

error.

at startup.  Starting it up when you get a connection from another
cluster could cause delays in processing the request.

want to only wait for message_timeout instead of forever.  Otherwise
we could hit deadlock if the other person is trying to do the same
thing.

processed at a time.  Otherwise you could get issues if you are
rapidly adding and removing a cluster from a federation.  Probably
not likely in real life, but in testing that is a different story.

slurmctld.

scenario when first added to a federation.

the same logic that was found in the slurmdbd.  Now both functionalities
share the same code.

This was done with the merge right before this commit.

update is sent to a slurmctld.

with real persistent connections.

the fd is set to -1 so we don't try to actually use it.

when querying if there are runaway jobs.

out in, but it is back now!