Bug 3431

burst_buffer/cray - Fix race condition that could cause multiple batch job
    launch requests resulting in downed nodes.
bug 3366

This reverts commit b9bff82f.

Bug 1599

slurmctld/agent race condition fix: Prevent job launch while PrologSlurmctld
    daemon is running or node boot in progress.
bug 3366

If a lower version client would try to communicate with a higher version
controller the dbd would return the controller's version and the client
would use that version to talk to the controller. When the controller
would respond, the client wouldn't know how to unpack the higher version
msg.

Bug 2508

If job is allocated nodes which are powered down, then reset job start time
    when the nodes are ready and do not charge the job for power up time.
bug 3411

Add logic to monitor Uncorrectable Memory Errors (UME) and notify
  active jobs in case they run for a while afterwards. This copies
  logic from knl_generic to knl_cray. There may be a different UME
  monitoring system for Cray systems in the future. The original
  knl_generic development is in commit 56ff27da
bug 3341

bug 3390

ensure eio objects get explicitly shutdown when
eio_handle_mainloop exits. currently depending on
whether the order the eio_handle_mainloop and
eio_signal_shutdown get called relative to each other

when stepd is instructed to shut down the socket use
SHUT_RDWR instead of SHUT_RD. just using SHUT_RD can
cause srun to receive ECONNRESET if there's outstanding
data that's been sent to stepd that the task has not
read.

bug 3166

Bug 3398

bug 3099

This reverts commit e92b49d3.

instead of also in the backfill scheduler.

Avoid allocating resources to a job in the event that its run time
    plus boot time (if needed) extent into an advanced reservation.
Note that if resources have not yet been selected (e.g. determining
  availability of licenses or burst buffers with respect to advanced
  reservations) then assume a reboot will be required.
bug 3360

Bug 3405.

acct_policy_job_runnable_pre_select() calls assoc_mgr_set_qos_tres_cnt()
without tres READ_LOCK.

Note that existing code does not modify the tres structures, so this
cannot currently lead to a race condition.

Bug 3406.

job_submit/cnode was previously removed by commit 63bc71ed.

Bug 3403.

Add BootTime configuration parameter to knl.conf file to optimize resource
    allocations with respect to required node reboots.
Add node_features_p_boot_time() to node_features plugin to optimize
    scheduling with respect to node reboots.
bug 3360

pack_bit_fmt can exceed 0xfffe characters on large systems and thus
truncate leading to:
"error: Credential signature check: Credential data size mismatch"

Add warning to the pack_bit_fmt macro to highlight issue.

Bug 3257.

scancel modified to note that no jobs satisfy the filter options when the
    --verbose option is used along with one or more job filters (e.g. "--qos=").
bug 3072

scancel would treat a non-numeric argument as the name of jobs to be
    cancelled (a non-documented feature). Cancelling jobs by name now require
    the "--jobname=" command line argument.
bug 3072

Replace version borrowed from Linux 2.4.9 with newer version in 4.9.

burst_buffer/cray - Avoid "pre_run" operation if not using buffer (i.e.
    just creating or deleting a persistent burst buffer).
bug 3391

Previous job state information was "PENDING" rather than "REQUEUED"
  for each job requeued due to a burst buffer error.
bug 3388

scheduling a Datawarp job.

The assoc_mgr lock needs to happen before the bb_state.bb_mutex.  One place
this could cause deadlock is from src/slurmctld/controller.c
_accounting_cluster_ready() which calls clusteracct_storage_g_cluster_tres
which inturn calls bb_g_job_set_tres_cnt which calls bb_p_job_set_tres_cnt
which will lock the bb_muxtex after the assoc_mgr is already locked.

Bug 3389

Bug 3331

Cache results of bit_set_count() calls.

Bug 3393.

The old logic would result in test16.4 failing some of the time.
  The failure was caused by the sattach command attaching to a
  job step before the original srun command received a
  RESPONSE_LAUNCH_TASKS message. That messsage  would then be sent
  to the salloc command. Since srun never got the message, it
  would hang. This change does not mark the job step as RUNNING
  until after the original srun gets sent the RESPONSE_LAUNCH_TASKS
  message and sattach requests are blocked until that time.

backfill scheduler: Stop trying to determine expected start time for a job
    after 2 seconds of wall time. This can happen if there are many running jobs
    and a pending job can not be started soon.
byg 3373

This reverts commit 17549a03.

Bug 3364.