>>>> "J" == Jarett DeAngelis
J> Hi folks, I was able to get my head node installed and can now
J> boot it via the YUMI Linux installer USB. (I still don’t know why
J> this is necessary, but hey, it works.) Now I have two issues:
J> How do I add nodes to the cluster? I tried doing this with
J> qluman-qt. I go to Manage Hosts and New Hosts, where it sees the
J> MAC from one of my nodes. If I try to add it to the cluster I get
J> a SQL error from Python about how it cannot insert a new record
J> without the “status” field. Like so:
J> If I click either “Add Selected” or “Add Host” here, I get this:
J> Which seems to be generated from this SQL insert:
do you use the latest qluman-qt version (126.96.36.199)? The error suggests
you are using something older.
J> The procedure for adding hosts actually doesn’t seem to be
J> documented anywhere.
J> I thought maybe I could try it from the command line, but there
J> also doesn’t seem to be any documentation for the qluman-cli
qluman-cli has only restricted purposes after installing (bootstrapping
QluMan, which is done just once) and generating initial one-time tokens:
J> I’m not entirely sure how the host abstraction works in
J> Qlustar. My front-end/head node’s hostname is “argo,” yet the
J> only thing in the host database visible in qluman-qt that seems
J> to map to this is “beosrv-c.” Where did that name come from?
It's the hostname of the head-node in the cluster-internal network.
Please read up on the docs, then things will become clearer. Setting up a
cluster properly even with Qlustar is still a complex thing and just
going by 'trial&error' will not be a good approach.
J> The next steps I would like to get to are as follows:
J> * Add my two compute nodes to the cluster
J> * Add the NFS NAS I built to be a “scratch” filesystem
J> available across all nodes
J> * Install some bioinformatics software (a good resource for
J> this in Qlustar would be a big help)
J> * Make the cluster able to authenticate users via an Active
J> Directory instance — I have a hard time understanding users
J> in Qlustar. There is a very brief blurb in the administration
J> guide about sssd.conf being replicated to each machine in the
J> cluster, but the hostname in SSSD would have to be different
J> for each node, so I don’t think that works for AD unless
J> Qlustar is doing some kind of templating with the names. The
J> other issue here is that when you join Linux machines to
J> Active Directory and users log in, no entry is created in
J> /etc/passwd, so if user process work is being kept track of
J> with those, it won’t work. Unless the typical method for
J> authenticating users is to use “plain” LDAP somehow without
J> actually performing an AD join and just manually creating
J> users that are then only authenticated against LDAP, and
J> replicated between hosts normally because they’re “regular”
J> UNIX users? That would also be fine;
AD integration is custom stuff and not Qlustar specific. If you have it
working on the head-node with the correct sssd.conf, it will also work
on the nodes booting from network, since that sssd.conf is integrated
into the node OS images automatically. You might still have to add
certificates on the netboot nodes by hand. Apart from that, there is
nothing more I can help on this topic.
J> the problem here is that I can’t figure out how users are
J> supposed to work. I added a user via qluman-qt and it appears not
J> to have been added to /etc/passwd either,
They are in the NIS database. Please check the admin guide.
J> and since I have not yet configured SSSD I can’t figure out what
J> those users are for. I can’t log in to SSH with them.
On the head-node(s), only root is allowed to login via ssh. Normal users should
only be allowed on FE nodes.