Repairing raid systems at per-stripe granularity

A technique for managing RAID (Redundant Array of Independent Disks) storage includes maintaining active-stripe metadata that stores, for each of multiple stripes that have been written by a RAID system, an entry that identifies the stripe and identifies a respective configuration plan in effect a last time the stripe was written. In response to a disk drive failure, the technique further includes generating a set of new plans that specify disk drive elements to be used in place of damaged ones and performing a selective repair operation. The selective repair operation iterates over a set of entries in the active-stripe metadata, performs a comparison between the identified plan for each stripe and a current plan implemented by the RAID system for writing to that stripe, and repairs the stripe when the identified plan and the current plan are different.