Abstract: Techniques are provided for implementing a persistent key-value store for caching client data, journaling, and/or crash recovery. The persistent key-value store may be hosted as a primary cache that provides read and write access to key-value record pairs stored within the persistent key-value store. The key-value record pairs are stored within multiple chains in the persistent key-value store. Journaling is provided for the persistent key-value store such that incoming key-value record pairs are stored within active chains, and data within frozen chains is written in a distributed manner across distributed storage of a distributed cluster of nodes. If there is a failure within the distributed cluster of nodes, then the persistent key-value store may be reconstructed and used for crash recovery.

Abstract: Techniques are provided for a recovery process with selective ordering and concurrent operations in order to recover from a failure. Representations of active log structures are rebuilt within memory according to ordering values assigned to I/O operations logged within the active log structures. Representation of certain active log structures may be concurrently rebuilt based upon the active log structures comprising I/O operations that are non-overlapping within a distributed file system, have no dependencies, relate to different services, and/or target independent files. Representation of stale log structures are concurrently rebuilt within memory. While rebuilding the log structures and executing the I/O operations, a key value map is concurrently rebuilt within the memory for locating data of the I/O operations. Concurrent operations during the recovery process reduces the time to complete the recovery process, and thus reduces client downtime during the recovery process.

Abstract: Systems and methods are for improving latency and throughput for metadata-heavy workloads and/or workloads including metadata bursts by decoupling data journal records and metadata-only journal records are provided. According to one embodiment, expedited and independent space reclamation is facilitated by differentiating between various types of journal records chains of which should be retained until different conditions are met. For example, data journal records may be added to data journal record chains within a persistent KV store and metadata-only journal records may be added to metadata-only journal record chains within the persistent KV store. Reclamation of spaced utilized by a data journal record chain may be reclaimed after both remote node data flush has been completed and the completion of a local CP for all records in the chain, whereas records of a metadata-only journal chain may be freed independently upon completion of a local CP for all records.

Abstract: Techniques are provided for implementing a journal using a block storage device for a plurality of clients. A journal may be hosted as a primary cache for a node, where I/O operations of a plurality of clients are logged within the journal. The node may be part of a distributed cluster of nodes hosted within a container orchestration platform. The journal may be stored in a storage device comprising a block storage device and a cache. Adaptive caching may be implemented to store some journal data of the journal in the cache. For example, a first set of journal data may be stored in the block storage device without storing the first set of journal data in the cache. A second set of journal data may be stored in the block storage device and the cache.

Abstract: Techniques are provided for implementing a persistent key-value store for caching client data, journaling, and/or crash recovery. The persistent key-value store may be hosted as a primary cache that provides read and write access to key-value record pairs stored within the persistent key-value store. The key-value record pairs are stored within multiple chains in the persistent key-value store. Journaling is provided for the persistent key-value store such that incoming key-value record pairs are stored within active chains, and data within frozen chains is written in a distributed manner across distributed storage of a distributed cluster of nodes. If there is a failure within the distributed cluster of nodes, then the persistent key-value store may be reconstructed and used for crash recovery.