Configuring when takeover occurs

Reasons for takeover

Takeovers can happen for several different reasons. Some system errors must cause a takeover; for example, when a clustered storage system loses power, it automatically fails over to the other cluster node.

However, for some system errors, a takeover is optional, depending on how you configure your cluster. The following table outlines which system errors can cause a takeover to occur, and whether you can configure the cluster for that error.

System error	Option used to configure	Default value	Notes
A clustered system undergoes a system failure and cannot reboot.	`cf.takeover.on_failure` set to On	On	You should leave this option enabled unless instructed otherwise by technical support.
A clustered system undergoes a software or system failure leading to a panic.	`cf.takeover.on_panic` set to On	Off, unless: FCP or iSCSI is licensed the system is a FAS270	For more information, see Enabling and disabling immediate takeover of a panicked partner.
There is a mismatch between the disks that one clustered system can see and the disks that the other system can see.	`cf.takeover.on_disk_shelf_ miscompare` set to On	Off	For more information, see How disk shelf comparison takeover works.
All the network interface cards (NICs) or VIFS enabled for negotiated failover on a node failed.	`cf.takeover.on_network_ interface_failure` set to On, `cf.takeover.on_network_ interface_failure.policy` set to `all_nics`	By default, takeover on network failure is disabled.	To enable a network interface for negotiated failover, you use the `ifconfig` if_name `nfo` command. For more information, see the Network Management Guide.
One or more of the NICs or VIFs enabled for negotiated failover failed.	`cf.takeover.on_network_ interface_failure` set to On, `cf.takeover.on_network_ interface_failure.policy` set to `any_nic`	By default, takeover on network failure is disabled.	To enable a network interface or VIF for negotiated failover, you use the `ifconfig` if_name `nfo` command. For more information, see the Network Management Guide.
A clustered system fails within 60 seconds of booting up.	`cf.takeover.on_short_uptime` set to On	On	Changing the value of this option on one node automatically updates the option on the partner node.
A clustered system cannot send heartbeat messages to its partner.	n/a		You cannot prevent this condition from causing a takeover.
You halt one of the clustered systems without using the `-f` flag.	n/a		You cannot prevent this condition from causing a takeover.
You initiate a takeover manually using the `cf takeover` command.	n/a		You cannot prevent this condition from causing a takeover.

Specifying the time period before takeover

You can specify how long (in seconds) a partner in a cluster can be unresponsive before the other partner takes over.

To specify the time period before takeover, complete the following step.

Step	Action
1	Enter the following command: options cf.takeover.detection.seconds number_of_seconds The valid values for number_of_seconds are 10 through 180; the default is 15.

Both partners do not need to have the same value for this option. Thus, you can have one partner that takes over more quickly than the other.

Note
If your cluster is failing over only because one of the nodes is too busy to respond to its partner, increase the value of the cf.takeover.detection.seconds option on the partner.

Attention
If the specified time is less than 15 seconds, unnecessary takeovers can occur. Use caution when assigning a takeover time of less than 15 seconds.

How disk shelf comparison takeover works

When communication between clustered nodes is first established through the cluster interconnect adapters, the nodes exchange a list of disk shelves that are visible on the A and B loops of each node. If, later, a system sees that the B loop shelf count on its partner is greater than its local A loop shelf count, the system concludes that it is impaired and prompts its partner to initiate a takeover.

Note
Disk shelf comparison does not function for clusters using software disk ownership, or for Fabric Attached MetroClusters.

Configuring VIFs or interfaces for automatic takeover

After you configure your interfaces or VIFs to allow takeovers and givebacks to be completed successfully, you can also optionally configure them to trigger automatic takeover if any or all of them experience a non-transient failure.

To configure a VIF or interface for automatic takeover, complete the following steps.

Step	Action
1	For every VIF or interface on which you want to enable automatic takeover, enter the following command: *ifconfig interface_name* nfo** For more information about this command, see the na_ifconfig man page or the Network Management Guide. For more information about configuring automatic takeover, see Configuring when takeover occurs.
2	Update the /etc/rc file so that your changes persist across reboots.
3	The default policy is that takeover only occurs if all the NICs or VIFs on a node that are configured for automatic takeover fail. If you want takeover to occur if any NIC or VIF configured for automatic takeover fails, enter the following command: `options cf.takeover.on_network_interface_failure.policy any_nic`
4	Enter the following command to enable the feature: `options cf.takeover.on_network_interface_failure enable`