GitLab

Instead of enumerating all groups on the system and comparing, check members of specific queue groups. This change makes CQM compatible with sssd.conf: enumerate = False, and overall is doing less work.

We need to log and end the task timers on both __rc_unknown and
__rc_success on __task_finalize.  The only outcome that doesn't
officially end a task is __rc_retry, as it will go into communication
retries at that point.

Only set the timers/emit messages once, despite retry attempts.

start.

This was a simple type mismatch that would result in a failure of
scaling dep fracs based on job size.  Now the float value is forced.

This adds the limiter for maximum jobs overall running in queue.  Useful
for profiling machines with noisy network environments.  This also adds
output to cqadm for this information, and an entry in the cqadm manpage.

Adds feature requested for #4320.  This will signal the head pid rather
than the process group.  Not recommended for use with csh scripts.

When upgrading you would lose all dep-held jobs.  This makes sure that
the appropriate flag is added and set on reinitialization post-upgrade.

I was getting a dep-release message when using qsub -h.  This fixes the
message.  Dependency handling and reporting appears to be working.

Restart tests are pending.

Statefile tests and multi-chain tests pending.

Some systems are configured such that the user is only associated with
their default group in /etc/pwd as opposed to /etc/groups.  This allows
the groups check to function in an expected manner, and pass in that
circumstance.

If groups and users are set on a queue, you have to be both in users and
have a group in groups.

Adam Scovel authored Mar 13, 2015 and Paul Rich committed Mar 13, 2015

This makes dep_hold a proper hold state (as opposed to an iterative
check)  dep_holds should now follow the same score accrual rules as
other holds (don't gain score).  Verification required, and integration
testing must be done.  Also, upgrade tests are pending.

There is simple scaling for dep_frac that can be enabled in the
cobalt.conf that gives a lower amount of fraction for dependent jobs if
the dependent job is smaller in nodecount relative to it's parent.

This is based on a patch submitted by Adam Scovel <ascovel@alcf.anl.gov>

Adam Scovel authored Mar 11, 2015 and Paul Rich committed Mar 11, 2015

This lets a Unix group be set on a queue as a restriction.  If the
submitter isn't a part of the group, the job is rejected.

Jobs now only inherit if they share a project

When users release holds, total_etime should be reset.  This is a patch
from Adam Scovel.

This fixes bad defaults in ion kernel options and ensures proper
operation with alt-kernel support disabled.

Also, fixed local simulator setup script.

1. PBS-style directive for script jobs
2. Multiple env unions
3. Added appropriate tests

Also adding in geometry and passthrough blocking information to the
logging database.  We can backfill data later.

Fixed typo

Have some more work to do with suppressing reboots if we're still on the
default kernel, but this appears to be doing the right thing for the
most part.

This rev is likely broken.  Checked in to archive changes with testing
pending.

Flag to qsub/qalter to request that a script job not be prebooted.  This
gets ignored for cN mode jobs.

Also adding in a client that lets us get the bootable blocks that are
children of a given block.  For use by scripts so they have an easier
time splitting up blocks for ensemble jobs.

Still require qalter, and display support.  The basic cases are working, though.

Also need to add validation code to make sure that qsub gets rid of the obviously
wrong cases.

This allows a user to state: only run this job on this geometry.

This is a problem we saw on Vesta where a resource would not be released if
a user put a job on hold while the job was being scheduled by bgsched.

cqm run_job will set a resource reservation in the system component if
there isn't one already when starting a job.