3 ms·
Ever have an LSMT quit writing a run half way through? Ever need to try and restore Cassandra where mutation storage order isn't guaranteed, with the last muta
by nyrikki 2y ago
Ever have an LSMT quit writing a run half way through? Ever need to try and restore Cassandra where mutation storage order isn't guaranteed, with the last mutation always winning?
It is absolutely not "atomic" if you have ever had to do a recovery from SSTable snapshots.
- bjornsing 2y ago> Ever have an LSMT quit writing a run half way through? That should be very easy to “solve” by just writing the length of the SSTable in a header. > Ever need to try and restore Cassandra where mutation storage order isn't guaranteed, with the last mutation always winning? This is more of a distributed systems challenge as I understand it: Cassandra is “multi-master” so can end up in a situation where transaction order is hard to sort out.