Tombstone Accumulation Slowing Read Path
warningFrequent deletes create tombstones that must be scanned during reads. When tombstone counts exceed 1000 per read, query latency degrades dramatically despite healthy system resources, often masking as a generic slow query problem.
Watch for cassandra_client_request_read_time_50p and _99p increasing while CPU and memory remain healthy. Correlate with rising cassandra_compaction_tasks_pending and check system logs for 'tombstone scanned' warnings exceeding 1000. Use nodetool tablestats to verify average tombstones per slice trending upward.
Set or reduce TTL (Time-To-Live) for appropriate tables to expire data naturally instead of explicit deletes. Avoid large batch deletes. Run nodetool compact on affected tables to purge expired tombstones. Review gc_grace_seconds settings and ensure repair operations run regularly to safely remove tombstones.