Manufacturing company · Jun–Aug 2026
Modernizing the BI platform: Tableau on a highly available Linux cluster
Moved the company's BI platform off single Windows servers onto a highly available Linux cluster — without reworking hundreds of existing data sources.
Business impactDashboards the business runs on no longer go dark when one server does, and the move cost no report rework.
Role: Owned the platform decision, migration plan, and delivery
- failover after losing a node, down from ~5
- ~1 min
- cluster recovery after losing a worker
- ~80 s
- data sources reworked in the migration
- 0
The problem
Tableau ran on aging single Windows servers: one reboot meant no dashboards. Moving to Linux promised high availability, but hundreds of data sources depended on Windows-style authentication, and certificates, licensing, and backups were all manual.
How it works
- 01
Highly available cluster
A multi-node Linux cluster laid out per the vendor's HA reference, with a replicated repository and tuned failover timing.
- 02
Existing data sources, unchanged
Delegated service-account authentication lets existing extracts and live connections refresh from Linux without editing a single one.
- 03
Separate embedded analytics
Customer-facing embedded dashboards run on their own server, isolated from internal load.
- 04
Automated care
Nightly backups with off-site copies, unattended certificate rotation that halts on any anomaly, and a frequent license-health check.
What I built
- The cluster design, migration, and failover tuning.
- Authentication for existing Windows-auth data sources from Linux — including live connections, which the published guidance says won't work.
- Backup automation with off-site cloud storage, certificate automation, and license monitoring.
Results
- Failover after a node loss cut from ~5 minutes to ~1–1.5 minutes.
- Surviving nodes recover in ~80 seconds when a worker goes down.
- Certificate renewal became an unattended ~6–7 minute job per environment.
- License monitoring now catches a failure mode that previously looked healthy from the outside.
What I learned
- The first plan — running Tableau in containers — was abandoned because the vendor doesn't support it. Knowing when to stop is part of the job.
- “Healthy” isn't the same as “working.” Monitor the thing users need, not the port.
Stack
- Tableau Server
- Linux
- PostgreSQL replication
- Cloud object storage
- Bash