Are rolling upgrades supported?
Rolling upgrades work by updating one "node" or "instance" at a time while the others continue to handle live traffic. Here is how this is achieved across the different OCI layers:
If you are renting a cluster of web servers, OCI supports rolling upgrades through Instance Pools and Instance Configurations.
The Process: When you update your "Instance Configuration" (e.g., to use a new OS image or a larger server shape), you can trigger a Rolling Update.
Batching: OCI spins up a new server with the updated configuration, waits for it to pass a Health Check from the Load Balancer, and only then terminates one of the old servers.
Control: You can define the "Batch Size" (e.g., update 10% of the fleet at a time) to ensure you always have enough capacity to handle your users.
In 2026, OKE features Non-destructive worker node updates.
Boot Volume Replacement: OCI can update the underlying OS or Kubernetes version of a worker node without terminating the actual server instance.
Cordon and Drain: OKE automatically "cordons" a node (stops new work from going to it), "drains" the existing pods to other healthy nodes, and then performs the upgrade. This keeps your IPs and OCIDs stable while moving your application forward.
Oracle is the industry leader in database rolling upgrades.
Oracle RAC (Active-Active): Since RAC has multiple nodes accessing the same data, OCI can patch Node 1 while Node 2 handles all transactions. Once Node 1 is finished, the system switches and patches Node 2.
Autonomous Database: This service uses a decoupled architecture. Oracle switches your database to a "pre-patched" compute stack in the background. In 2026, the Online Restart option reduces the "hiccup" during this switch to just a few seconds, which most applications handle via standard connection retries.
Data Guard: For single-instance databases, you can use a "Standby" database. You upgrade the Standby first, verify it, and then perform a Switchover to make it the Primary, resulting in zero data loss and minimal downtime.
| Service | Method | User Impact |
| Instance Pools | Batch Replacement | Zero (if Load Balancer is used) |
| OKE (Kubernetes) | Cordon & Drain | Zero (Pods move to other nodes) |
| Oracle RAC | Node-by-Node | Zero (Connections failover to active node) |
| Autonomous DB | Compute Swapping | Minimal (Brief connection blip) |
Before a rolling upgrade begins, OCI's AI-based system health checks analyze your current traffic. If the system predicts that taking one node offline (even for a rolling update) would cause a performance bottleneck, it will automatically "burst" your capacity or delay the update until a lower-traffic window is detected.
To make a rolling upgrade truly "invisible" to users, always enable Connection Draining on your Load Balancer. This tells the Load Balancer to stop sending new requests to a server scheduled for an upgrade, but allows existing users to finish their sessions before the server is taken offline.