Re: Durability of transactions in Cluster

[Sandeep Jangra]

Thanks for the response.

In InnoDB all completed transactions are written to bin log files before they are committed to the client. So at any point in time all committed transactions are either in the InnoDB tables or in bin log files (which is disk).

But in the case of cluster, there can be a time slice when the committed transactions are in the RAM of nodes. I know the time window for this is relatively small. Worst case if all the nodes in this HA group go down, I can see some probability of losing data. If MySQL cluster was writing to bin logs (like InnoDB) before every commit, I would feel less chances of data loss. I guess cluster is by-design kept to keep data in memory for performance reasons.