Skip to main content

High Availability

This article refers to Platform v3.1.0. The current Platform version is v3.2.0.

Overview​

Barbara's High Availability (HA) support brings fault-tolerant, resilient operation to your edge workloads. It is built on Docker Swarm for cluster orchestration and KeepAlived for the floating cluster IP. Clusters give you continuous service even when individual nodes fail or are taken offline for maintenance.

This article is the entry point to the HA reference. For day-to-day operations, follow the dedicated guides:

Core features​

Cluster creation​

Create a cluster from one or more edge nodes. Docker Swarm orchestrates the cluster so that services can run redundantly across nodes and resist single-node failure.

Persistent volumes​

Use highly available volumes managed at the cluster level so that critical application state survives node restarts and failovers.

Virtual IP (KeepAlived)​

Assign one or more virtual IP addresses (VIPs) to a cluster. KeepAlived migrates the VIP between healthy nodes automatically, so clients keep reaching the cluster through a single address even when the underlying owner changes.

Swarm overlay network​

Cluster traffic flows over a secure, isolated Swarm overlay network. Services on different nodes communicate as if they were on a single LAN.

Dynamic node management​

Add nodes to a cluster with extra configuration on top of the regular node settings:

  • Node labels for fine-grained scheduling and placement.
  • Swarm traffic restriction — pin Swarm traffic to a specific network interface for performance or security.

Non-disruptive firmware upgrades​

Put a node into Drain before updating it. Drain mode reschedules its workloads to other cluster members, so services stay up while the node updates. See Cluster Info.

Cluster-wide configuration sync​

Barbara propagates the Global Config across every node in the cluster. A change made at the cluster level lands on every node automatically.

Cluster visibility​

  • Swarm configuration overview — leader node, members, tokens, status, all in one place.
  • Centralized secrets management — Global Secrets defined at the cluster level are distributed to services via native Docker Swarm encryption.
  • Docker registry credentials — set authentication for private registries at the cluster level.

Application and model deployment​

Deploy workloads to a cluster through the 6-step Adding applications wizard. The wizard exposes the four Docker Swarm deployment modes:

  • Replicated — runs a specified number of identical tasks across the cluster.
  • Global — runs exactly one task on every eligible node.
  • Replicated Job — runs a replicated task until it completes.
  • Global Job — runs a global task until it completes.

In addition, placement constraints based on node labels let you control which nodes a service may run on — useful for matching hardware capabilities, locations, or roles.

Summary​

HA in Barbara is Docker Swarm + KeepAlived under a Panel UI: create clusters, attach nodes, install apps and models through the 6-step cluster wizard, and use Drain to keep services available while you update nodes underneath them.