Github user aweisberg commented on a diff in the pull request:
https://github.com/apache/cassandra/pull/224#discussion_r188782956
--- Diff:
src/java/org/apache/cassandra/cql3/statements/AlterKeyspaceStatement.java ---
@@ -96,7 +98,35 @@ private void warnIfIncreasingRF(KeyspaceMetadata ksm,
KeyspaceParams params)
StorageService.instance.getTokenMetadata(),
DatabaseDescriptor.getEndpointSnitch(),
params.replication.options);
- if (newStrategy.getReplicationFactor() >
oldStrategy.getReplicationFactor())
+
+ validateTransientReplication(oldStrategy, newStrategy);
+ warnIfIncreasingRF(oldStrategy, newStrategy);
+ }
+
+ private void validateTransientReplication(AbstractReplicationStrategy
oldStrategy, AbstractReplicationStrategy newStrategy)
+ {
+ if (oldStrategy.getReplicationFactor().trans == 0 &&
newStrategy.getReplicationFactor().trans > 0)
+ {
+ Keyspace ks = Keyspace.open(keyspace());
+ for (ColumnFamilyStore cfs: ks.getColumnFamilyStores())
+ {
+ if (cfs.viewManager.hasViews())
+ {
+ throw new ConfigurationException("Cannot use transient
replication on keyspaces using materialized views");
+ }
+
+ if (cfs.indexManager.hasIndexes())
+ {
+ throw new ConfigurationException("Cannot use transient
replication on keyspaces using secondary indexes");
+ }
+ }
+
+ }
+ }
--- End diff --
I've never actually thought about the procedure for changing RF. I think I
get the principle behind one at a time changes.
What happens when you do this and write at CL.ALL and read at CL.ONE? What
happens for ranges where the node that is now a full replica but doesn't have
the full data is read from?
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
