This section covers deploying the NVIDIA Minerva platform on Docker Swarm. Choose the installation that fits your deployment:
|
Installation |
Use it when |
|---|---|
|
All platform services run on one host. Use the |
|
|
The platform services are spread across a manager node and worker nodes, for templates sized for higher ingestion rates, for example |
|
|
A multi-node installation that keeps running when an HA node fails. It requires shared NFS storage on every node. |
For the hardware each template needs, see Resource Requirements.
Stacks
The svm script manages the platform services in five stacks. The stacks start in this order and stop in the reverse order:
|
Stack |
What it runs |
|---|---|
|
|
Shared infrastructure: Kafka and ZooKeeper, the databases (MySQL, MongoDB, Redis, and Neo4j), Consul, Keycloak, and the nginx front end |
|
|
The Druid time-series database: broker, coordinator, router, historical, and middlemanager |
|
|
Telemetry ingestion and processing, anomaly detection and correlation, notifications, the API, licensing, high-frequency telemetry storage, and device provisioning |
|
|
NVLink management: service registration, inventory, partitions, and scheduling |
|
|
The NVIDIA Minerva web UI |
To check the stacks, run ./svm stack-status in ~/minerva.
Last updated: