All glibc nss interfaces need to be installed as revision 2 at this time,
thus the hard coded -version-info.

After 083c48b8 these are sent as part of the launch cred, and only sent
as part of launch_tasks_request_msg_t / batch_job_launch_msg_t is send_gids
is disabled. (In which case the slurmd will populate them instead through
these structs, since it cannot alter the launch cred.)

Bug 6215

Change references to the "micro" release in rpc.html and troubleshoot.html
as well; SchedMD refers to the last part of the version number as the
"maintenance" release.

Bug 6833.

Bug 6832.

Bug 6832.

Bug 6832.

No functional change.

Bug 6832.

Otherwise, we could send communication packets bigger than max_allowed_packet.

Bug 6832.

Co-authored-by: Tim Wickberg <tim@schedmd.com>

as part of testing hetjobs ability to be started, the accounting
functions _add_usage_node_bitmap() and _rm_usage_node_bitmap()
were being called before nodes were selected for the jobs. Logic
is added here to prevent those error messages specifically for
pending hetjobs.

Regression introduced in 8d643e79.

Bug 6832.

The problem was freeing an interior pointer to buffer contents before
the call to FREE_NULL_BUFFER. The issue was only triggered when loading
an archived data with protocol version < 17.11.

Regression introduced in 8d643e79.

Bug 6832.

Provide a way to bypass this during testing with 'scontrol getent' by
sending GETPW_MATCH_ALL in. Change nss_slurm to use GETPW_MATCH_PID
to require the pid to match for getpwent().

Only supports getpwuid / getpwnam at the moment.

Used to fetch all 'getent passwd' info from all steps on node,
making it easier to verify that nss_slurm will be able to operate
correctly.

Will be used to service getpwuid / getpwnam requests from nss_slurm.

And out of the slurmctld code. Remove send_groups_in_cred handling
now that it's all handled inside slurm_cred.c.

There is a minor difference in functionality after this: the pw_gecos,
pw_dir, and pw_shell fields will be populated when they were not
previously. As the expensive part of the lookup is the underlying
getpwuid() syscall, this should not make a real difference.

Add a copy_gr_names() utility function to group_cache.c;
these lookups should be cached at some point to avoid the
performance hit from calling into getgrgid() constantly.

Lookup and send over the required struct passwd fields.

Match the other struct passwd fields added previously.

And handle pack/unpack for 19.05.

No functional change.

And always print a colon after the function name.

Without this change a heterogeneous job allocation could not run
a step on any single component of a hetjob, but only across multiple
hetjob components. Note that by default a step runs on only the
first hetjob component, so the default mode of operation running
steps on a hetjob was broken. This bug caused tests 38.7 and 38.11
to fail consistently. The bug was introduced in commit 2f162b7a

Commit 2987c5db (Convert "Admin Priority" to "Site Priority")
broke this test

Bug 6849.

Use retry_list_size() instead of list_count(retry_list) to avoid any asserts if list is NULL.

Bug 6849.