Concurrent access to data during replay of a transaction log

摘要

A dataset is concurrently recovered after a server crash while the dataset is actively used for servicing client requests. In response to a reboot, records of not-yet-completed transactions in a transaction log are parsed to create a graph of dependencies between the transactions. Once this graph has been created, clients may access a specified block of the dataset after on-demand recovery of the specified block. The on-demand recovery is concurrent with a background recovery task that replays the not-yet-completed transactions in time order. The on-demand recovery uses the dependency graph to replay any and all transactions that support recovery of the specified block, so that recovery of the specified block includes update of other blocks to be consistent with the recovered block in accordance with the not-yet-completed transactions.

主权项

1. A method of recovery of a dataset in response to reboot of a data processor of a data storage system, the data storage system having data storage storing the dataset and a log of records of transactions upon the dataset, said method comprising the data processor executing computer instructions stored on a non-transitory computer readable storage medium to perform the steps of:
(a) parsing records in the log of transactions not-yet-completed by the time of the re-boot in order to create a dependency graph of dependencies between the not-yet-completed transactions; and then (b) performing a background task of replay of the not-yet-completed transactions in a time order sequence, and concurrent with the background task of replay of the not-yet-completed transactions in the time order sequence, responding to a request from a client for access to a specified block of data in the dataset by performing on-demand recovery of the specified block and then performing client access to the recovered specified block, and the on-demand recovery of the specified block accessing the dependency graph in order to replay the not-yet-completed transactions that support recovery of the specified block.