Ignition redundancy runs two copies of the Gateway: a master that holds the configuration and a backup that receives it and takes over when it can no longer reach the master. This article follows Inductive Automation's Ignition 8.3 user manual on what is copied, which settings decide a failover and what clients, history and alarms do, and notes where the 8.1 manual differs. Comparison with FactoryTalk View SE and AVEVA System Platform: SCADA server redundancy.
1. Master, backup and how they talk
Each Gateway has a redundancy Mode: Independent (no redundancy), Master or Backup, with one master and one backup per pair. The Designer connects only to the master and can never be opened on the backup, even while the master is offline. In 8.3 the status and the settings are both under Platform > System > Redundancy; 8.1 puts the settings under Config and the status under Status.
All traffic goes from the backup to the master over the Gateway Network, by default on TCP port 8088 without SSL or 8060 with SSL, so the master must accept connections on that port through any firewall between them. In 8.3 each side approves the other's Gateway Network certificate and the master approves the connection; the backup then downloads a system backup from the master and restarts. Certificates and the rest of Gateway security are covered in Ignition Gateway security.
An Edge Gateway can only be paired with another Edge Gateway. If the two machines run different operating systems, the Windows machine should be the master, or Force Failover will not work.
2. What the backup receives and what it does not
| Item | How it reaches the backup | Notes in the manual |
|---|---|---|
| Configuration: projects, Gateway settings, modules the backup lacks | Queued on the master and pulled by the backup; a full system backup if the backup is too far out of date | Both data-only and full transfers soft-reboot the backup |
| Runtime state, such as current alarm states | Sent as differences; a full state transfer on first connection or when out of sync | No restart |
| Alarm events and shelved alarms | Two queues on the active node, sent as they change | A full queue triggers a full transfer; each queue defaults to 2,000,000 entries (8.3) |
| Redundancy settings | Not shared | Each node keeps its own values |
| Backup versions (8.3 only) | Saved in backupConfig.json, used only on the backup | Ignored on Gateway Settings, Edge Settings and Security General Settings |
When the nodes reconnect after a separation, the backup drops data that conflicts with the master's. With Use Active Uptime to Resolve Conflicts on, the node that has been active longer wins, so the master can overwrite its own data with the backup's.
The 8.1 manual has no Backup versions: every connection must resolve from both nodes, so "localhost" in a database connection will not work. Since 8.3.0, a resource that needs another value on the backup, such as an OPC UA server address, gets a Backup version.
On the master, the Redundancy page shows Pending Updates, which keep growing while the backup is disconnected; the backup's page offers Force Re-Sync, which restarts the backup.
3. Settings that decide failover and recovery
Each node has its own redundancy settings. These shape a failover:
| Setting | Node | What it does (8.3 manual) |
|---|---|---|
| Standby Activity Level | both | Cold: connects to OPC servers without subscribing to tag values, so no extra device load, but failover takes slightly longer. Warm: runs as if active, except that it does not log data or write to devices |
| Startup Connection Allowance | both | How long a starting node waits for its peer before deciding to be active; default 30,000 ms. At 0 or a low value the master goes active first and manual recovery is not possible |
| Sync Timeout | both | Limit for a sync operation; default 60 s |
| Recovery Mode | master | Automatic: the returning master takes over. Manual: the backup stays active until it fails or a user requests the switch |
| Ping Rate, Ping Timeout, Ping Max Missed | backup | How often the backup pings the master, how long it waits, and how many missed pings mark the connection as faulted |
| History Mode | backup | Full or Partial (section 5) |
The manual gives no single failover time. The 8.1 table shows a Failover Timeout (default 10,000 ms) removed in 8.1.37; the 8.3 table has none. Set too low for the network, the failover time causes socket read timeouts and repeated reconnects.
4. Clients during a failover
Vision Clients connect to the active node; when it fails, they move to the other node by themselves and tell the user. The Vision Client Launcher stores the backup's address after connecting to the master, and when the master does not answer it tries the backup, alternating until one responds.
Perspective sessions also use the active node. When the connection drops, a session retries its own Gateway and watches the backup; if the backup is active first, the browser opens the same project and page there. If sessions do not return when the master takes control again, the manual suggests turning off Auto Detect HTTP Address and entering an address both Gateways can reach. Perspective Workstation tries the backup when the master is unavailable at start-up.
Failover works between nodes on different versions, so clients stay connected during an upgrade. For a status screen, the System tags in the Gateway's Redundancy folder include Role, IsMaster, IsActive and ActivityLevel (undecided, cold, warm or active).
5. History, alarm journal and the database
The SQL database sits outside the pair. The simplest layout is one shared database server on another machine, with one connection that the master copies to the backup. The manual lists what depends on the database:
- History. All history goes through Store and Forward, so it is cached while the database is unreachable, but clients cannot view it until it is stored.
- Alarms. Alarm status is not kept in the database and keeps working. The Alarm Journal goes through Store and Forward like history.
- Tags and screens. Query tags and screen queries fail while the database is down; screens that use only tags keep working.
The backup's History Mode deals with a limit the manual states openly: the backup cannot tell whether the master is down or only unreachable. In Full mode the backup logs straight to the database, so history is available at once but may be duplicated. In Partial mode it caches history until it reaches the master again, and forwards only data from the time the master was really down.
For replication that copies only from the primary database to the secondary, the manual pairs a standard failover connection with Partial history; for a cluster that replicates writes both ways, a sticky failover connection with Full history. Transaction groups run only on the active node. The historian itself is covered in what is a process historian.
6. Licensing the backup
Redundancy works on Ignition, Edge and Cloud Edition, with standard (six-character) or leased (eight-character) licences. The backup's licence holds the same modules as the master's: if the master stores history with SQL Bridge, the backup needs it too, or data is lost after a failover.
Two standard licences also let both Gateways run independently if redundancy is turned off. A backup-only licence forces its Gateway into Backup mode; changing the mode means unactivating it and applying another licence, or running the two-hour trial. Restoring a .gwbk onto an existing Gateway leaves its modules, licence and redundancy settings unchanged.
7. Upgrading a redundant pair
The 8.3 manual gives one order for minor and major upgrades, and the 8.1 manual the same for 8.1.17 and later: set the master's Recovery Mode to Manual, upgrade the backup and fail over to it, upgrade the master, then select Assume Control. Older 8.1 releases have their own procedure. The 8.3 steps are in upgrading Ignition 8.1 to 8.3.
What we check first
This is TDL's practice, not an Inductive Automation requirement.
- Both licences, module by module, against what the master uses for history, alarming and transaction groups.
- The version and operating system of each node.
- The Gateway Network port through every firewall between the nodes.
- The database layout, and whether History Mode and the failover connection match it.
- Addresses that differ on the backup, and the launchers and Workstation on operator PCs.
How we test a pair
Also our own practice: each result is compared with what the manual describes.
- Baseline. Both nodes connected, Role and ActivityLevel as expected, Pending Updates not growing.
- Requested failover. Request Failover on the Redundancy page; Vision Clients report the transfer, Perspective sessions open the same page on the other node.
- Master lost. Power off the master, then pull only its network cable; check history for duplicates.
- Recovery. Bring the master back; Automatic or Manual recovery behaves as set.
- Database lost. Stop the database; alarms keep working and history arrives once it returns.
We record the measured failover times; the manual gives none for a whole system.
How TDL approaches it
TDL Industrial Automation is an Ignition Certified Integrator. We design a pair together with its database and licences, and test the failover before handover. More: Ignition SCADA integrator in Vietnam, SCADA and historian and What is Ignition.
Frequently asked questions
Does the Ignition backup Gateway need the same modules as the master?
Yes. The backup's licence holds the same modules as the master's; if the master stores history with SQL Bridge and the backup lacks it, data is lost after a failover. A backup-only licence locks its Gateway into Backup mode.
Can the Designer be opened on the backup Gateway?
No. The Designer connects only to the master and cannot be opened on the backup even when the master is offline. Configuration is made on the master and copied to the backup.
What do Vision Clients and Perspective sessions do on failover?
Vision Clients move to the other node by themselves and tell the user they were transferred. Perspective sessions retry their Gateway while watching the backup, and open the same project and page on the backup if it becomes active first.
Can a failover duplicate history in Ignition?
Yes, in Full history mode, because the backup cannot tell whether the master is down or only unreachable and logs straight to the database. Partial mode caches history and forwards only data from the time the master was really down.
How long does an Ignition failover take?
The manual gives no single figure. The Failover Timeout setting was removed in 8.1.37, the backup's link to the master is timed by its ping settings, and a Cold standby takes slightly longer. We measure the time on the real system during testing.
Sources
- Inductive Automation, Ignition Redundancy (8.3) and Ignition Redundancy (8.1)
- Inductive Automation, Setting Up Redundancy (8.3) and Setting Up Redundancy (8.1)
- Inductive Automation, Redundant Licensing and Database Considerations, Ignition 8.3
- Inductive Automation, Installing and Upgrading Ignition (8.3) and Installing and Upgrading Ignition (8.1)
- Inductive Automation, Vision Client Launcher and Perspective Workstation, Ignition 8.3
- Inductive Automation, System Tags, Gateway and Gateway Network Parameters and New in this Version, Ignition 8.3
- Inductive Automation, Gateway Backup and Restore and Understanding Transaction Groups, Ignition 8.3
Read 15 November 2026. This article paraphrases the sources and does not quote them; read the original documents before designing or working on a system.