Without this change a high priority batch job may not start at submit
time. In addtion, a pending job with mutltiple partitions be cancelled
when the scheduler runs if any of it's partitions can not be used by
the job.

(removed some unused variables but many remain)

Fix to some un/pack logic
Fix to test12.5 for new sacct help format
Address various compiler warnings

Attached is the energy accounting patch that Martin and Yiannis have been working.  The framework is there, but the functionality it currently not working.  They are both on vacation this week and then are back a week before the conference.  I thought it would be better to send in order to get the framework and the structures in place for an official 2.5.0 instead of waiting.  If you disagree, just let us know and we can send it again when the low level functionality working.    Here is a short summary of our test results.

1. jobacct_gather/none + energy_accounting/none

Looks OK.  Did not find any errors.

2.  jobacct_gather/linux or cgroup + energy_accounting/none

Looks OK.  Did not find any errors.

3.  jobacct_gather/linux or cgroup + energy_accounting/rapl

Slurmd aborts when you run a job that uses a node that does not support RAPL.  This appears to be because of the error()/pexit() at line# 150/151 in energy_accounting_rapl.c.  We need to change this code to just issue a debug message and return.  For now, energy_accounting must not be configured if the cluster includes any nodes that do not support RAPL.

The cpu frequency values reported by jobacct_gather are not correct.

Again, there are obviously some problems, so if it would be better to wait for full functionality just let us know.  It may be three weeks before they are able to spend some time on this to fix the problems, so that is why I thought you may prefer to have something that has the correctly data structures in sooner rather than later.

previous 20 minute time limit.  The previous behavior would fail for
large files 20 minutes into the transfer.

previous 20 minute time limit.  The previous behavior would fail for
large files 20 minutes into the transfer.

Adding a reset level in reset_stats for controlling values to reset

algorithm

This addresses trouble ticket 85

Add support for advanced reservation for specific cores rather than whole
nodes. Current limiations: homogeneous cluster, nodes idle when reservation
created, and no more than one reservation per node. Code is still under
development. Work by Alejandro Lucero Palau, et. al, BSC.

Format is "name:used/total"

This eliminates one of four RPCs needed for each job.

Hi,

With slurm 2.3.2 (or 2.3.3), I encounter the following error when
trying to launch as root a command attached to a running user's job
even if I use the --uid=<user> option :

sila@suse112:~> squeue
   JOBID PARTITION     NAME     USER    STATE      TIME TIMELIMIT
NODES   CPUS NODELIST(REASON)
     551     debug mysleep.     sila  RUNNING      0:02 UNLIMITED
1      1 n1

root@suse112:~ # srun --jobid=551 hostname
srun: error: Unable to create job step: Access/permission denied
<--normal behaviour

root@suse112:~ # srun --jobid=551 --uid=sila hostname
srun: error: Unable to create job step: Invalid user id <--problem

By increasing slurmctld verbosity, the log files displays the follwing
error :

slurmctld: debug2: Processing RPC: REQUEST_JOB_ALLOCATION_INFO_LITE from
uid=0
slurmctld: debug:  _slurm_rpc_job_alloc_info_lite JobId=551 NodeList=n1
usec=1442
slurmctld: debug2: Processing RPC: REQUEST_JOB_STEP_CREATE from uid=0
slurmctld: error: Security violation, JOB_STEP_CREATE RPC from uid=0 to
run as uid 1001

which occurs in function : _slurm_rpc_job_step_create
(src/slurmctld/proc_req.c)

Here's my patch to prevent the command from failing (but I'm not sure
that there is no side effects) :

all jobs would be returned even if the flag was set.
Patch from Bill Brophy, Bull.