0

We have two tables with replication factor 3. We write in batches to both tables, the same data. We write with LOCAL_QUORUM.

When we read with LOCAL_ONE we get different results for data written months ago. If we read with LOCAL_QUORUM it's fine.

From what I understand eventually consistent means that the data wrote in the last seconds might not be consistent .. but not the data older than months.

Dan Priscornic
  • 11
  • 2
  • Hey Dan, can you share your schemas and queries to give us a better idea of what's going on? While Cassandra is an eventually consistent system, inconsistencies usually mean a larger problem. Cassandra replicates in microseconds - if it takes seconds for all replicas to get the data, something is very wrong. – MarcintheCloud May 18 '16 at 15:11
  • do you ever run `nodetool repair`? It's still a recommended practice afaik. also are your nodes synced up wrt time? the times should be the same within a few millis. otherwise writes can get messed up and older data can appear to be newer because of a bad timestamp. – LHWizard May 18 '16 at 18:06

1 Answers1

0

UPDATE: Of course, the issues was caused by not running regular repairs.

Dan Priscornic
  • 11
  • 2