DOCSEARCHbyAlgolia
Submit Docs

ClusterVision TrinityX Documentation

website·Indexed 33 days ago

https://docs.clustervision.com

Documentation for TrinityX HPC cluster management, featuring the Luna CLI and Luna2 REST API. Covers cluster installation, upgrading, and operational troubleshooting, as well as node and group management, BIOS configuration, power (BMC) management, and high availability (HA) using Pacemaker. Includes guides for workload management with SLURM, AlertX monitoring, zero-touch provisioning (ZTP) for switches, and image management including Any-Image and diskfull provisioning.

Tokens
434.1K
Snippets
2.6K
Records
3.7K
Agent score
50%

What's inside TrinityX

  1. Overview of TrinityX monitoring architecture

    TrinityX monitoring is built on the Prometheus and Grafana stack. It uses a pull-based architecture where the Prometheus Server periodically scrapes HTTP endpoints exposed by systems and services. The monitoring framework consists of four pillars: Metrics exporting, Metrics collection & storing, Dashboards & Graphing, and Alerting.

  2. Overview of TrinityX High Availability Architecture

    TrinityX HA uses Corosync and Pacemaker to ensure central services remain available if a controller fails. It typically employs an active/passive setup where one controller runs single-instance services and the shared virtual IP, while the other remains as a standby. Managed resources are grouped to ensure sequential startup and movement:

    • Trinity: Floating (virtual) IP (trinity-ip).
    • Trinity-zfs-trinityx: Shared storage (DRBD, ZFS).
    • Trinity-stack: Single-instance services including MariaDB, OpenLDAP/389 Directory Server, slurmctld, and luna2-master.

  3. Overview of TrinityX Any-Image functionality

    The Any-Image approach in TrinityX allows for the creation of alternative OS images for compute nodes, enabling scenarios where the headnode and compute nodes run different distributions (e.g., a Rocky-8 image on a system where the headnode runs Alma-9).

    There are four supported approaches to creating these images:

    1. Default: The base image is generated based on the controller's distribution.
    2. Base: Uses a pre-packaged base image to build upon.
    3. Docker: Uses images from the Docker Hub repository as a base.
    4. OSgrab: Captures an already installed node and converts it into a TrinityX image.

  4. Overview of TrinityX Any-Image functionality

    The Any-Image approach in TrinityX allows the creation of alternative OS images for nodes, enabling scenarios where the headnode and compute nodes run different distributions (e.g., a Rocky-8 image on an Alma-9 headnode).

    There are three supported methods for providing the image base:

    1. Default: The base is generated based on the controller's own distribution.
    2. Base: Uses a pre-packaged base image to build upon.
    3. Docker: Uses images from Docker Hub (official or custom) to provide the base.

  5. Overview of TrinityX image management

    Images are stored as file/directory structures in /trinity/images (default trix_images). These directories represent the node's filesystem after installation. Because the filesystem image is not the one distributed to nodes, it must be 'packed' into a tarball using the Luna provisioning system before changes are propagated to the cluster. Image manipulation must be performed on the active controller in high-availability setups.

  6. Overview of TrinityX Slurm (cv-slurm)

    TrinityX Slurm (also known as cv-slurm) is the default workload management offering for TrinityX 16, replacing the deprecated OpenHPC Slurm stack. It provides cross-distribution support (RHEL-family and Ubuntu), stays current with Slurm releases independently of OpenHPC, and integrates with TrinityX's graphical configurator and automation tools.

  7. Overview of TrinityX HPC and AI Infrastructure Management

    TrinityX is an open-source, on-premises infrastructure management solution for HPC and AI systems, featuring CloudBursting capabilities. It provides essential services for modern HPC/AI systems with full installation customization and optional modules for compute node fine-tuning. A standard installation typically includes Luna (provisioning and configuration), Slurm, MariaDB, and OpenLDAP. TrinityX is compatible with Enterprise Linux versions 8, 9, and 10.

  8. Overview of TrinityX High Availability (HA)

    TrinityX HA ensures central services remain available if a controller fails or undergoes maintenance. It uses Corosync and Pacemaker to manage an active/passive setup where one controller runs single-instance services and the shared virtual IP, while the other remains as a standby. Managed resources include the virtual IP, shared filesystems (ZFS/DRBD), MariaDB, directory services (OpenLDAP/389 DS), and the Luna image master service. These services must not be started or stopped manually via systemctl; they are controlled exclusively by Pacemaker.

  9. Overview of AlertX Rule Domains

    AlertX organizes monitoring rules into three distinct domains:

    • Generic rules: Typically used for operating system level rules.
    • Services rules: Used to observe required services such as sshd or sssd.
    • Hardware rules: Automatically generated upon first boot based on the node's hardware configuration. These can be reset to match the current state after hardware changes.

  10. Overview of AlertX Rule Domains

    AlertX organizes monitoring rules into three distinct domains:

    • Generic rules: Typically used for operating system level rules.
    • Services rules: Used to observe required services such as sshd or sssd.
    • Hardware rules: Automatically generated upon the first boot based on the node's hardware configuration. These can be reset to match the current state after hardware changes.

  11. Overview of node image management and packing

    Node installation images are stored as file/directory structures in /trinity/images (default trix_images). These directories represent the state of a node after installation.

    Packing: Changes made to the filesystem image are not automatically distributed. Images must be "packed" into a tarball before the provisioning system (Luna) can distribute them to nodes via torrents, HTTP(S), or other methods. Packing is required after any modification before nodes can see the changes.

  12. Overview of TrinityX HPC and AI Infrastructure Management

    TrinityX is an open-source, on-premises infrastructure management solution for HPC and AI systems, featuring CloudBursting capabilities. It provides essential services for modern HPC/AI systems and allows for full installation customization and fine-tuning of compute nodes via optional modules. A standard installation typically includes Luna (for provisioning and configuration), Slurm, MariaDB, and OpenLDAP. TrinityX is compatible with Enterprise Linux versions 8, 9, and 10.