Files
werewulf/docs/07-operations.md
ilker 29bd39c135 Warewulf HPC test cluster: provisioning, identity, docs and overlay UI
A complete, verified HPC lab on Proxmox (hv01) with its own isolated subnet,
central identity and a web console for overlay management.

Infrastructure
- vmbr6: L2-only cluster fabric, deliberately without an address on the
  hypervisor so Warewulf owns DHCP/TFTP/PXE for the whole subnet.
- head01 (VM 300): Warewulf 4.7.0, dnsmasq-free DHCP/TFTP/iPXE chain,
  Slurm controller, MUNGE, NAT gateway, /opt exported read-only.
- wwnode1-4 (VM 301-304): diskless, PXE-booted, rootfs unpacked into RAM.
- idm1 (VM 305): FreeIPA (LDAP/Kerberos/Dogtag CA/DNS) plus the NFS home
  server. Dual-homed on purpose: idm1.hpc.local resolves to the cluster
  fabric so Kerberos and NFS never traverse head01's masquerade.

Stack
Rocky Linux 9.8, OpenHPC 3, Slurm 25.11.4, OpenMPI 5.0.7, FreeIPA 4.13.1.

Stateless domain join
Nodes are rebuilt in RAM every boot, so ipa-client-install cannot be used.
Enrolment happens once and is distributed through Warewulf overlays: shared
config as templates rendered per node via {{.Id}}, and a separate keytab per
node so wwnode1 cannot impersonate wwnode2.

Overlay web UI (webui/)
A Go service on the head node serving a browser interface over the Warewulf
REST API. It exists as a server rather than a static page because the API is
root-equivalent and bound to loopback; the service holds the credentials and
proxies same-origin, so the browser never sees one. Published behind Caddy
with bcrypt logins, server-side sessions and per-user/per-IP lockout.

Documentation
README plus docs/01-10 covering architecture, installation, Warewulf,
identity, storage, Slurm/MPI, operations, troubleshooting and reference,
with a 22-page PDF edition. Section 08 records twelve concrete failures hit
during the build, including a wwctl flag that fails silently.

No credentials are stored here. Every secret lives on the servers; the
documentation refers only to its path.
2026-08-29 09:07:51 +02:00

5.0 KiB

7. Operations Runbook

7.1 Access

ssh head01      # head node   10.0.0.30, user ilker, sudo NOPASSWD
ssh idm1        # identity    10.0.0.31, user ilker, sudo NOPASSWD
ssh proxmox     # hypervisor  10.0.0.10, root

Compute nodes sit on the isolated subnet and are reached through the head node:

ssh head01
sudo ssh root@wwnode1        # or root@10.10.10.11

Root SSH from head01 to the nodes uses /root/.ssh/cluster, a dedicated key pair distributed by the ssh.authorized_keys overlay.

Serial console (useful when a node will not finish booting):

ssh proxmox 'qm terminal 301'      # wwnode1 = 301, wwnode2 = 302, …
# exit with Ctrl-O

7.2 Power management

ssh proxmox 'bash -s' < scripts/99-cluster-ctl.sh status
ssh proxmox 'bash -s' < scripts/99-cluster-ctl.sh start     # idm1, head01, nodes
ssh proxmox 'bash -s' < scripts/99-cluster-ctl.sh reset     # reboot nodes only
ssh proxmox 'bash -s' < scripts/99-cluster-ctl.sh stop      # stop nodes only
ssh proxmox 'bash -s' < scripts/99-cluster-ctl.sh stopall   # everything

Start order matters. idm1 must be up before nodes boot — they need DNS from it during the overlay stage and will otherwise come up without working identity. head01 must be up before nodes for PXE. 99-cluster-ctl.sh start enforces this order.

idm1 and head01 have onboot=1 and start with the hypervisor. Nodes have onboot=0 and are started deliberately.

7.3 Health check

./scripts/A99-verify.sh

Checks, in order: IPA services (expect 9 running), NFS exports, IPA user count, head01 hostname/DNS//home/services/sinfo, then per node sssd, slurmd, munge, /home mount and a real kinit -k against the KDC — finally submitting an MPI job and printing its output.

krb5=OK means the node actually obtained a ticket with its host keytab, not merely that the file exists.

7.4 Common tasks

Add a user

ssh idm1
sudo bash -c '. /root/.ipa-credentials
  echo $IPA_ADMIN_PASSWORD | kinit admin
  ipa user-add ayse --first=Ayse --last=Yilmaz \
      --homedir=/home/ayse --shell=/bin/bash --password
  ipa group-add-member hpcusers --users=ayse'

Home directory is created on first login. No node reboot needed — SSSD picks up new users immediately.

Change something in the node image

ssh head01
sudo wwctl image exec rocky-9 -- /bin/bash
  dnf install -y <package>
  exit
ssh proxmox 'bash -s' < scripts/99-cluster-ctl.sh reset

Costs a full reboot cycle (~5 min) since nodes re-download the image.

Change configuration without touching the image

ssh head01
sudo wwctl overlay import -p -o hpc /etc/slurm/slurm.conf /etc/slurm/slurm.conf
sudo wwctl overlay chmod hpc /etc 0755        # always re-check after import -p
sudo wwctl overlay build
  • System overlay change → node reboot required.
  • Runtime overlay change → picked up within 60 s automatically.

Add a compute node

  1. Create the VM on Proxmox (copy the 00-proxmox-provision.sh node block, next free VMID and MAC BC:24:11:E0:0A:1<n>).
  2. Register it:
    sudo wwctl node add wwnode5 --ipaddr 10.10.10.15 \
         --hwaddr BC:24:11:E0:0A:15 --netdev eth0 --discoverable=false
    
  3. Create its host entry and keytab on idm1 (A40 pattern), copy the keytab to head01, build an ipa.wwnode5 overlay, assign it:
    sudo wwctl node set --yes wwnode5 -O ipa.wwnode5
    
  4. Add DNS A and PTR records on idm1.
  5. Extend slurm.conf to NodeName=wwnode[1-5], re-import into the hpc overlay, scontrol reconfigure.
  6. wwctl overlay build, then boot the node.

Restart services

# head
sudo systemctl restart warewulfd dhcpd slurmctld

# identity
ssh idm1 'sudo ipactl restart'

# a node
ssh head01 'sudo ssh root@wwnode1 systemctl restart sssd'

7.5 Logs

What Where
Warewulf provisioning journalctl -u warewulfd on head01
DHCP / PXE journalctl -u dhcpd on head01
Slurm controller /var/log/slurmctld.log on head01
Slurm node daemon /var/log/slurmd.log on the node
FreeIPA (all) idm1:/var/log/ — dirsrv/, krb5kdc.log, httpd/
SSSD journalctl -u sssd on the client
Node boot serial console (qm terminal <vmid>)

For a node that will not boot, capture the serial output to a file:

ssh proxmox 'systemd-run --unit=wwnode1-serial --collect \
  socat -u UNIX-CONNECT:/var/run/qemu-server/301.serial0 CREATE:/tmp/wwnode1.log'
ssh proxmox 'qm reset 301; tail -f /tmp/wwnode1.log'

This is how the original 4 GB RAM failure was diagnosed — the node was dying silently with nothing in any server-side log.

7.6 Shutting the lab down cleanly

ssh proxmox 'bash -s' < scripts/99-cluster-ctl.sh stop      # nodes first
ssh head01 'sudo umount /home'                              # release NFS
ssh proxmox 'qm stop 300; qm stop 305'

Stopping idm1 before unmounting /home on head01 leaves a hung NFS mount and makes the head node's shutdown hang.