Be really cautious here, this can be deceptive There are races in some versions of cassandra that can leave you with different combinations of cfid
The cfid is on disk for schema It’s in memory for schema It’s used for the table path on disk Those three have to match for things to work properly and various races can make them mismatch If you have one ID for the path and a different in the schema table on disk and you bounce, the database throws away the old directory and you start fresh with a new empty directory So resetlocalschema May “fix” this but it may fix it by throwing away one copy of data If only one host is out of sync you may want to pretend it died and replace it. Alternatively a ton of inspection can tell you which variation of mismatch you have and you can correct it properly but it’s more work than I’m prepared to type today Sorry for the unpleasant reality. > On Apr 5, 2020, at 3:38 PM, Erick Ramirez <erick.rami...@datastax.com> wrote: > > >> Another suggestion before resetlocalschema. Try rolling restart all the >> nodes in the cluster and see if it fix the problem. After the restart all >> the nodes will use the same schema for the table. > > That's a little bit heavy-handed. 🙂 Resetting a node's schema is a simple, > online operation that doesn't involve a cluster-wide restart. Cheers!
