Post-installation

Passwords

What are the passwords?

By default, the TrinityX installer will generate random passwords for all services that require one. You can find all of the generated passwords on the controller(s) at /etc/trinity/passwords/ where every password lives in its own file that’s named after the service that uses it.

See Passwords for details.

Verifying the installation

Before tailoring the cluster it is worth confirming that the controller came up healthy. A few quick checks:

• Confirm the core services are running with the bundled diagnostic, which reports the status of the Luna daemon, the database, DNS, DHCP, LDAP, the workload manager and the monitoring stack in one place:

# trix-diag

This is the same tool described under Operational troubleshooting; every listed service should report as active.

• Check that the Luna daemon itself is up:

# systemctl status luna2-daemon

• Get a per-node overview of provisioning, IPMI and workload-manager state with the cluster utility:

# lcluster

See Luna utilities for how to read its output. Once a node has booted, its installer status should show as successful.

If any of these report a problem, the Operational troubleshooting page covers the most common causes before you continue with the tasks below.

Typical tasks

Once the controller has been installed and the first image has been generated, the cluster will run with the defaults. Although this is sufficient to get started, some tailoring may be in place, such as:

• Installing applications and modules,
• Adding users and groups (See user management),
• Tailoring the compute image (See image management),
• Tailoring the compute node partition layout (default is diskless) (See node and group management),
• Adding more nodes, removing or modifying the default set (See node and group management),
• Configuring the workload manager (See Enabling Slurm nodes for jobs below),
• Tailoring the monitoring and dashboards.

Enabling Slurm nodes for jobs

The installer lays down /etc/slurm/slurm-nodes.conf with a catch-all NodeName=DEFAULT line that is intentionally a placeholder: it does not describe real hardware, so nodes are not enabled for jobs until it is corrected. Do one of the following:

• Edit the first line to match the hardware of your default node, then let Luna generate the per-node configuration from it, for example:

NodeName=DEFAULT Boards=1 SocketsPerBoard=2 CoresPerSocket=24 ThreadsPerCore=2 RealMemory=257000 State=UNKNOWN

• Or configure the host(s) with the Slurm Graphical Configurator in Open OnDemand, which writes the same configuration for you.

Apply the change so the controller picks it up:

# scontrol reconfigure

See workload management for how the managed blocks and the default-node line work in detail.