Tombstones

• In scylla you don’t really delete the data you just mark that row for deletion
• created when data is deleted, tombstones disappear after gc_grace_seconds

Cemeteries

• performance can be impacted by presence of tombstones
• make sure tombstones are expiring fast enough
• make sure the cluster has enough resources to compact often

Scylla Manager

• Cluster Operations Automation Tool [COAT]
• Centrally Managed
• persistent, highly available, stateless
• sctool status is sample health check query

Node Repair vs Manager Repair

• local level | global level
• Ad hoc operations | Recurrent/ Ad hoc
• Fire and Forget | Start/stop Follow progress from CLI and Dashboard
• no records | keeps track history
• shard ignorent | shard aware
• no retries | custom retrys

Scylla Repair Intensity and Speed

• When scheduling repair, you may specify --intensity param
  • intensity =1, will repair one segment in each repair command on all shards in parallel
  • value >1 specify the number of segments repaired by scylla (higher speed and cluster load)
• value < 1 shows % of node's shards repaired in parallel
• intensity =0 will use limits specified in scylla manager configuration