Preventing Data Resurrection with Repair Based Tombstone Garbage Collection

ScyllaDB has introduced a novel repair-based tombstone garbage collection (GC) method as an experimental feature in its Open Source 5.0 version to address the issue of data resurrection. This approach departs from the traditional timeout-based tombstone GC, which relies on the gc_grace_seconds parameter to determine when tombstones should be purged. The repair-based method ensures that tombstones are removed only after a repair is conducted, eliminating the pressure to complete repairs within a specific timeframe and enhancing data consistency by preventing the reappearance of deleted data. This new method also reduces the need for administrators to determine an appropriate gc_grace_seconds value and allows for more flexible repair scheduling, potentially improving read performance and reducing latency impacts on user workloads. Users can activate this feature through new table configuration options, ensuring greater control over tombstone management and overall database safety.