I tried to setup AllowGroups for some partitions and found that some
worked, but others didn't have any valid users in them. I turned on
debugging and played with the code and eventually found that the code in
slurmctld/groups.c is trying to scan the entire list of groups in order
to enumerate its users. Our users and groups come from the campus LDAP
servers which are very large and restrict query sizes.
The getgrent_r function is the problem, it tries to pull the entire list
of group names which then needs to be parsed. That seems pretty
inefficient at best. In our case we end up with about 1/4 of the total
entries so that when all is said and done some of our groups work, but
others don't. All the information we need is in the results of the
getgrnam_r function.
Can this be fixed somehow?
Thanks,
Mike
smime.p7s
Description: S/MIME Cryptographic Signature