Data consistency at cluster failure

Tibor_Mozes

Hello,

What happens with data in a k-safe cluster (k=1) when a transaction is complete but the data is still in the WOS and not yet in the ROS,  and the entire cluster shuts down ungracefully (not just 1 node). Say due to power outage.

Will all data in previously only in WOS be lost?

Thank you,
Tibor

Adrian_Oprea

Well when that happens you will lose all that data if the transaction was not commited(written to disk). If one node would go down the data could be recoverd from the other nodes. So if the moveout doesn't happen before the crush data will be lost. To manage this you need to make sure in case of large loads to use DIRECT so data goes to ROS and no data is lost when the instance crash. To monitor this you need to keep your eyes on the stale checkpoint events and make sure LGE is not behind one any of your nodes.