static.

This reverts commit afdd8778.

Turns out MsgAggregation hits a race condition here where the
assoc_mgr_tres_list might not be set up yet.  Reverting this commit
makes everything work as expected.  It was only added in the first
place as convenience.

Otherwise incorrect status could be registered in the database and completed
jobs be seen as pending, running, and so on.

Bug 4847

Add infrastructure for per-job and per-step TRES parameters: tres-per-job,
    tres-per-node, tres-per-socket, tres-per-task, cpus-per-tres, mem-per-tres,
    tres-bind and tres-freq. These new parameters are not currently used, but
    have been added to the appropriate RPCs.

This does not add the new fields, but just duplicates the v17.11
  logic. The next commit will indentify what changes.

Non-pack jobs were getting their pack_offset set to -1, which resulting
in a bad pack_group value being reported for non-pack steps which were
killed.

We already use an unguarded #define _GNU_SOURCE throughout the code,
so clean these out #ifndef guards out just for consistency.

_acct_policy_validate was always set.  But that could override
QOS_FLAG_DENY_LIMIT which is really what we want to be going with.

We now are honoring it fully.  If the job doesn't have a QOS we will
deny on limit by default.

Bug 4749.

some reason and we never init query.

of MB.

one doing something different.

Rank 0 will allocate memory, 1 will write to disk and 2 will read.

from the assoc_mgr.

precision.

TRES variables.

Bug 2782

tres columns.

Bug 2782

TRES variables.

Bug 2782

TRES columns.

Bug 2782

which is when the problem actually exists.