cred address will always be 0.0.0.0 for IPv6 (for now) so the error is expected.

Bug 8350

Bug 8350

Bug 8350

Bug 8350

Bug 8350

Bug 8350

Bug 8350

move bit_or() before while()

Bug 9537

minor code performance improvement

Bug 9537

Bug 9537

Call _add_preemptable_job() only if job uses max_start_delay reservation or
system has enabled preeamption.

Bug 9537

Create function to determine if a job uses a max_start_delay reservation.

Bug 9537

job_overlap_and_running() calls bit_overlap_any() which can be expensive
operation on big systems. _is_job_preempt_exempt() can also call it but it
is a less common use-case.

Bug 9537

Bitstrings are handled as 64bits and bit_ffs()/bit_fls() are faster on
whole words.

Bug 9537

Bug 8350

The one location using this already performs the NULL check anyways.

And make all the necessary changes for the switch. Start anticipating
IPv6 handling while revisiting all these code paths as well.

Bug 8350.

Which will make it simpler to switch the typedef later.

Bug 8833

See also 9ea6c946

Bug 8833

Bug 9617

Also added the word test in the suffix to make the backup file name a
little more unique from backups generated by other sources

Bug 9617

When a very high number of reconfigure RPCs were served by slurmctld,
a segfault could happen if one thread was trying to save the state
while another one was reading the config.

Serving reconfigure RPCs in parallel doesn't make sense, so fix it by
changing it to be sequential instead.

Bug 8680

Bug 9306

_rpc_batch_job->slurm_cred_handle_reissue() uses pointers from
_find_job_state() without lock.

Bug 9306

Bug 9167

Bug 8112

The race was possible when the newly registered node got a job to start.
Lock initialization can potentially be done by two threads one handling
the registration reply in _handle_node_reg_resp and second commit from
job being started _send_slurmstepd_init.

Bug 9808