Detail
Customers often see regular CPU spikes every 24 hours and contact Support to understand the cause. In most cases, the cause is the tomb-raider, a special background mechanism which removes tombstones that are no longer needed.Answer
By default, the tomb-raider runs every 24 hours. When tomb-raider runs, it spawns a thread per device in which the storage device is read from beginning to end, one large block at a time. Between each block the thread sleeps for the configured tomb-raider-sleep period, if any. When the end of the device is reached, return the threads and wait for the tomb-raider-period to elapse (default of 24 hours),before starting the process again.The above workflow accounts for increased large block reads on the storage device. If not correctly sized, this could introduce latency into the cluster, which could impact transactional operations. To reduce the impact, you can set serialize-tomb-raider = true, which will prevent tomb-raider from scanning multiple namespaces concurrently.