Add On: This procedure also works for the new Juniper MAG appliances. But keep in mind during the upgrade of the second host (and also the first): BE PATIENT!!
A Juniper SA cluster can be configured as active/active or active/standby cluster. An active/active cluster uses an external load balancer or DNS round-robin to enable load-sharing across multiple appliances. Today I had to upgrade an active/standby cluster and found an KB article on the Juniper website (restricted access) about the preferred upgrade method.
Juniper uses the following steps to upgrade a cluster:
- 1. Login directly to a member in the cluster as administrator;
- 2. Disable the member from the cluster;
- 3. Upgrade the service package on the disabled member;
- 4. After the upgrade is completed login back to the IVE and enable the disabled member in the cluster configuration;
The following notes are mentioned by Juniper:
- In active/standby cluster mode, it is recommended to start the upgrade process with the passive members and after completing the upgrade on the passive IVE and moving to the upgrade of the active IVE please note all connections are dropped when the active IVE is disabled. However after disabling the active node the passive IVE becomes active;
- Once the upgraded member is enabled back in the cluster, it shows the other nodes as Unreachable. This is expected behavior as the cluster members are running different versions and hence cannot sync with each other;
- Once the second IVE is being upgraded all user connections are dropped and not migrated due to the mismatch of software versions. This limitation is addressed in 4.0 with the Minimal downtime cluster upgrade available in the licensable Central Manager feature set;
I followed the steps mentioned above and the upgrade of the IVE cluster went smoothly. I disabled the passive node and upgraded the firmware with the new package. After the upgrade (and a reboot) the passive node was reachable in standalone mode. Next I logged in to the active IVE and enabled the passive node back into the cluster. When you hit Enable you receive the warning message that the configuration of the new cluster node will be erased and overwritten with the configuration of the active node. Just choose Yes.
After enabling the passive node, you will loose your web session with the active node. The VIP address is taken over by the new node in the cluster and the “old active” node starts updating automatically. This is a little tricky, because you don’t notice anything from the update process taken place. Just have patience and ping the node to check when it is online again. When the node is back online, login to the IVE and check the Cluster Status. Both IVE are now updated and members of the cluster. You could decide to do a manual Fail-Over IP to the “old active” node so everything is back to the original state before the upgrade.