While LifeKeeper for Linux supports both resource fencing and node fencing, its primary fencing mechanism is storage fencing through SCSI reservations. This fence, which provides the highest level of data protection for shared storage, allows for maximum flexibility and maximum security providing very granular locking to the LUN level. The underlying shared resource (LUN) is the primary quorum device in this architecture. Quorum can be defined as exclusive access to shared storage, meaning this shared storage can only be accessed by one server at a time. The server who has quorum (exclusive access) owns the role of “primary.” The establishment of quorum (who gets this exclusive access) is determined by the “quorum device.”
As stated above, with reservations enabled, the quorum device is the shared resource. The shared resource establishes quorum by determining who owns the reservation on it. This allows a cluster to continue to operate down to a single server as long as that single server can access the LUN.
SCSI reservations protect the shared user data so that only the system designated by LifeKeeper can modify the data. No other system in the cluster or outside the cluster is allowed to modify that data. SCSI reservations also allow the application being protected by LifeKeeper to safely access the shared user data when there are multiple server failures in the cluster. A majority quorum of servers is not required; the only requirement is establishing ownership of the shared data.
Adding quorum/witness capabilities provides for the establishment of quorum membership. Without this membership, split-brain situations could result in multiple servers, even all servers, killing each other. Watchdog added to configurations with reservations enabled provides a mechanism to recover from partially hung servers. In cases where a hung server goes undetected by LifeKeeper, watchdog will begin recovery. Also, in the case where a server is hung and not able to detect that the reservation has been stolen, watchdog can reboot the server to begin its recovery.
In addition to resource fencing using SCSI reservations, LifeKeeper for Linux also supports disabling reservations. Regardless of whether reservations are enabled or disabled, there are two issues to be aware of:
Access to the storage must be controlled by LifeKeeper.
Great care must be taken to ensure that the storage is not accessed unintentionally such as by mounting file systems manually, fsck manually, etc.
If these two rules are followed and reservations are enabled, LifeKeeper will prevent most errors from occurring. With reservations disabled (alone), there is no protection. Therefore, other options must be explored in order to provide this protection. The following sections discuss these different fencing options and alternatives that help LifeKeeper provide a reliable configuration even without reservations.
© 2016 SIOS Technology Corp., the industry's leading provider of business continuity solutions, data replication for continuous data protection.
Open topic with navigation