Continuous Availability (CA) is a feature in OneFS 8.0 that contributes to a transparent failover during a node or NIC failure. Dell recommends using CA enabled shares to minimize video loss during node or NIC failure operations.
CA describes when a node becomes inaccessible for any reason, such as administrative, failure, or infrastructure, then another node can be chosen to take its place and work can resume. CA is also known as "SMB Transparent Failover."
There are a couple of key features for this feature to work:
- Transparent Failover
- The "Transparent Failover" capability of SMB3 allows the connection to the shares to be maintained despite the transition the node's role (planned, unplanned). This capability allows more seamless access to the video files during most failure scenarios. Because the failover activity is not instant, to cover the short period when the client is reconnecting and reopening its files on another node, sufficient or insufficient buffering determines the video frame loss.
- Witness service
- Witness is service running on a second node that acts as the SMB connection witness monitoring the availability of the CA file shares. If there is a failover, the witness node has the responsibility to notify the SMB3 client that it should move its connection to a new node without waiting for TCP timeouts or DNS queries.
The Windows client starts the Witness connection. When the client makes the SMB connection to a node, the client also sends a Witness call (RPC) requesting a list of the cluster's IP addresses. The client picks the first address in the list marked as available for Witness registration. The client makes Witness calls to register for notification. There is no load-balancing algorithm because, at the moment, Witness is only providing status change notification.
When the CA-enabled share is available cluster-wide, the SMB connection moves from one node to another node seamlessly via a reconnection.
After a connection to a node fails with non-CA enabled shares, the operating system (OS) will quickly connect to a new node. Although the OS connection to the new node exists very quickly the SMB sessions are not available immediately. SMB maintains several timeout timers that must expire before the SMB session for the failed connection is made available.
To set up continuous availability, mark the CA box when creating the share. This selection causes a bit to be set in the supported configuration mask. If the client understands and supports the option, it just happens. It is important to remember you cannot add that setting after creating a share. There are tools to support re-creating the share with the CA setting enabled but they require disconnecting all the active clients during the process.