A single-node Tableau Server is a single point of failure. High availability deployments use multiple nodes across process types to eliminate that risk. This guide covers the HA architecture options, the minimum viable configuration, and the operational requirements.
A single-node Tableau Server is a single point of failure. If that node goes down, every user loses access to analytics simultaneously. For organisations where Tableau is a critical business system, a single-node deployment is an operational risk that eventually manifests as an outage at the worst possible time. High availability deployments distribute processes across multiple nodes so that any single node failure does not take the environment offline.
This guide covers Tableau Server's HA architecture, the minimum viable HA configuration, and the operational requirements.
Tableau Server process model
Tableau Server is composed of multiple processes, each responsible for different functions:
- **VizQL Server**: Renders views and handles interactive queries. CPU-intensive. The primary bottleneck under high user load.
- **Backgrounder**: Executes background jobs — extract refreshes, subscriptions, flows. CPU-intensive.
- **Application Server (Tomcat)**: Handles web requests, authentication, REST API calls.
- **Data Server**: Manages published data source connections and query routing.
- **Repository (PostgreSQL)**: Stores Tableau Server's metadata — workbook info, user accounts, permissions, schedules.
- **File Store**: Stores workbook files, extract files, and flow outputs.
- **Cache Server (Redis)**: Stores external query cache results.
- **Coordination Service (ZooKeeper)**: Cluster coordination and configuration.
- **TSM Controller**: Tableau Services Manager — manages the Tableau Server processes.
In a single-node deployment, all processes run on one machine. In a distributed deployment, processes are distributed across nodes.
The minimum viable HA configuration
Genuine HA requires three nodes for the Coordination Service quorum. ZooKeeper (the Coordination Service) uses a quorum-based consensus model — it requires a majority of nodes to agree before committing configuration changes. With two nodes, one failure leaves no quorum. With three nodes, one failure still leaves a 2-of-3 quorum, maintaining availability.
The minimum viable HA configuration:
**Three-node cluster**:
- Node 1 (initial node): All processes. Repository primary. Coordination Service.
- Node 2 (worker 1): VizQL, Backgrounder, Application Server, Data Server, File Store, Coordination Service, Repository passive.
- Node 3 (worker 2): VizQL, Backgrounder, Application Server, Coordination Service.
This configuration survives the loss of any single node. If Node 1 fails, Nodes 2 and 3 maintain quorum (2 of 3 Coordination Services), and the Repository passive on Node 2 is promoted to primary.
Repository HA: active/passive
The Repository (PostgreSQL) stores all Tableau Server metadata. A Repository failure means Tableau Server cannot authenticate users, load workbooks, or schedule jobs. The Repository must be highly available.
Tableau Server supports Repository HA through active/passive replication:
- One Repository is designated primary (active) — all reads and writes go here
- One Repository is designated passive — it replicates from the primary but does not serve traffic
- If the primary fails, the passive is promoted to primary automatically
The passive Repository must be on a separate node from the primary. For HA to function correctly on Repository failure, the passive is included in the cluster from the start.
File Store redundancy
The File Store holds extract files, workbook .twbx files, and flow outputs. In a distributed deployment, multiple File Store instances are configured and kept in sync via the Coordination Service. Any node running File Store receives a copy of all files. This ensures that if a node with a File Store instance goes down, the files remain available from other nodes.
For very large extract deployments (100GB+ of extract files), File Store sync time becomes a factor — adding a node to the cluster with File Store requires syncing all existing files to the new node, which takes time proportional to the total file store size.
Load balancing
In a multi-node Tableau Server, a load balancer sits in front of all application server nodes and distributes user requests. Tableau Server does not include a built-in load balancer — you need an external load balancer (AWS ALB, Azure Application Gateway, HAProxy, F5, or NGINX) configured to distribute traffic across Application Server nodes.
Configure the load balancer:
- Sticky sessions (session affinity): Not required but reduces unnecessary session re-establishment overhead
- Health checks: The Tableau Server health endpoint (/api/2.4/serverinfo) returns JSON with the server status — use this for load balancer health checks
- SSL termination: Typically handled at the load balancer, with HTTP internally
Tableau Cloud as the HA alternative
Tableau Cloud (SaaS) is Tableau Server's HA already handled by Salesforce/Tableau. Multi-region redundancy, automatic failover, and infrastructure management are all included in the licensing. For organisations evaluating Tableau Server HA, the total cost of ownership calculation often favours Tableau Cloud — not just the licensing, but the elimination of HA infrastructure and operational overhead.
The tradeoff: Tableau Cloud puts infrastructure outside your control. For organisations with strict data residency requirements or on-premises data sources that cannot be reached from the cloud, Tableau Server HA remains necessary.
Operational requirements
A Tableau Server HA deployment requires:
- Monitoring: alerts on process failures, backgrounder queue depth, Repository replication lag
- Patching strategy: Tableau Server upgrades require downtime. In an HA cluster, coordinate upgrades across nodes with minimum downtime maintenance windows
- Backup: Regular Repository backups and File Store backups (the tsm maintenance backup command creates a full backup including repository and file store)
- DR documentation: Runbooks for common failure scenarios — node failure, Repository failure, File Store corruption
For the Tableau Server administration foundations, see tableau server admin guide and tableau server performance tuning guide. Our managed BI services include 24/7 Tableau Server monitoring and HA deployment management — book a free infrastructure review.
A former Microsoft data architect audits your data foundation, identifies your top priorities, and sends you a written plan. Free. No pitch.
Book a Call →