Abstract: Techniques for address mirroring of a file system journal are presented. A transactional file system update can correspond to a write operation to be performed at a first participant node of a data storage system. A journal transfer component can transfer the transactional file system update from an initiator node of the data storage system to a first journal associated with the first participant node and recovery information relating to the update to a second journal associated with a second participant node of the data storage system. The recovery information can be determined based on the update. To facilitate the update, a journal descriptor block can be generated and sent to the second participant node, such block comprising the recovery information, which can include transaction state, block addresses involved in the transaction, error correction-related information, or a mirror block address set for the block addresses where desired.

Abstract: Systems and methods facilitating fault tolerance for transaction mirroring are described herein. A method as described herein can include receiving a commit command for a data transaction from an initiator node of the system, wherein the data transaction is associated with a first failure domain, and wherein the commit command is directed to a primary participant node and a secondary participant node of the system; determining whether a response to the commit command has been received at the primary participant node from the secondary participant node in response to the receiving; and, in response to determining that the response to the commit command was not received at the primary participant node, indicating that the secondary participant node is invalid in a data store associated with a second failure domain that is distinct from the first failure domain.

Abstract: Techniques for address mirroring of a file system journal are presented. A transactional file system update can correspond to a write operation to be performed at a first participant node of a data storage system. A journal transfer component can transfer the transactional file system update from an initiator node of the data storage system to a first journal associated with the first participant node and recovery information relating to the update to a second journal associated with a second participant node of the data storage system. The recovery information can be determined based on the update. To facilitate the update, a journal descriptor block can be generated and sent to the second participant node, such block comprising the recovery information, which can include transaction state, block addresses involved in the transaction, error correction-related information, or a mirror block address set for the block addresses where desired.

Abstract: Systems and methods facilitating fault tolerance for transaction mirroring are described herein. A method as described herein can include receiving a commit command for a data transaction from an initiator node of the system, wherein the data transaction is associated with a first failure domain, and wherein the commit command is directed to a primary participant node and a secondary participant node of the system; determining whether a response to the commit command has been received at the primary participant node from the secondary participant node in response to the receiving; and, in response to determining that the response to the commit command was not received at the primary participant node, indicating that the secondary participant node is invalid in a data store associated with a second failure domain that is distinct from the first failure domain.

Abstract: Implementations are provided herein for using a Node State Block (“NSB”) to track the validity of various copies of the journal. The NSB can contain states local to the node and be implemented independent of the cluster state, and be readable at node boot time. The NSB can be mirrored to every storage drive within the node. The NSB can contain a generation number and a validity bit. Every time the NSB is updated, the generation number can be incremented. An update must be written to a quorum of drives to be considered committed. At mount time, the NSB is read from all the drives, and the copy with the highest generation number is used. If a quorum of drives cannot be read, the mount fails. Read quorum, or the quorum required at mount when reading the NSB across the set of storage drives requires a strict majority of the storage drives to have valid copies of the NSB. Write quorum, or the quorum required to update the NSB across all storage drives, requires a simple majority of drives.

Abstract: Implementations are provided herein for a node among a cluster of nodes to maintain copies of its journal both locally and on a buddy node. Each copy of the journal can be associated with a validity bit and a clean bit that can be provided as inputs to a journal mirror state machine that can be used to maintain consistency between local copies of the journal and mirrored copies of the journal. The node can operate in a read-only mode or a read-write mode, whereby operating in read-only mode prevents changes to the local journal. The status of the mirror copy of the journal can be established in a link status depending on its availability. The journal can then transition between various states of the state machine upon triggering events that change the link status or the mode of the node. It can be appreciated that the transitioning among states of the state machine can provide crash-consistency for the filesystem during operation.

Abstract: Implementations are provided herein for an efficient reconstruction of distributed lock states during a group change within a distributed file system. When a node leaves or joins the cluster, a consistent hash can be used in reassigning the coordinator node to all the distributed locks throughout the file system. Through the use of the consistent hash, wherever possible, nodes that exist both prior to and after the group change can maintain their coordinator status for the locks they manage. If the node is aware they are maintaining coordinator status for the locks it manages, it can retain the locking status previously associated with the locks it was coordinating prior to the group change, thus reducing necessary backplane communication from a coordinator change, and increasing the efficiency for the group change process.

Abstract: Implementations are provided herein for sub-block ranged locking on a journal block. A file system locking scheme can be adjusted to provide ranged locks on sub-blocks of a journal block. By locking sub-block ranges of a journal block, two independent operations on the same block that do not overlap on a sub-block range can operate independent of each other and be processed concurrently without requiring an exclusive lock over the entire block. Delta write locks, delta read locks, and shared locks can be established on a sub-block range level.