Re: How do you apply (CQL) schema modification patches across a cluster?
On Mon, Jan 12, 2015 at 5:46 PM, Sotirios Delimanolis sotodel...@yahoo.com wrote: So do we have to guarantee that the schema change will be backwards compatible? Which node should send the schema change query? Should we just make all nodes send it and ignore failures? - Yes is the easiest answer. - Any single node, or an operator. Are you changing schema so frequently that you really need to automate this process? - Concurrent identical schema modification has historically meant asking for trouble, though in theory in the present/future it should be safe. =Rob
How do you apply (CQL) schema modification patches across a cluster?
Hey all, Assuming a cluster with X 1 application nodes backed by Y 1 Cassandra nodes, how do you best apply a schema modification? Typically, such a schema modification is going to be done in parallel with code changes (for querying the table) so all application nodes have to be restarted. However, in our case we can't (or don't want to) restart/turn off all application nodes at the same time. So do we have to guarantee that the schema change will be backwards compatible? Which node should send the schema change query? Should we just make all nodes send it and ignore failures? It's unclear to me how data model schema changes are done in a clustered environment. What are some good practices for handling this issue? Thanks,Sotirios
Re: How do you apply (CQL) schema modification patches across a cluster?
Are you changing schema so frequently that you really need to automate this process? I guess not. Though, if such a (consistent) process existed, I'd love to use it. The single node solution will have to do. Because of the source code change, it seems I still have to make sure that the patch script and the code deploy happens on that first single node before the code goes to all the other nodes. Does that sound right? Soto On Monday, January 12, 2015 6:10 PM, Robert Coli rc...@eventbrite.com wrote: On Mon, Jan 12, 2015 at 5:46 PM, Sotirios Delimanolis sotodel...@yahoo.com wrote: So do we have to guarantee that the schema change will be backwards compatible? Which node should send the schema change query? Should we just make all nodes send it and ignore failures? - Yes is the easiest answer.- Any single node, or an operator. Are you changing schema so frequently that you really need to automate this process?- Concurrent identical schema modification has historically meant asking for trouble, though in theory in the present/future it should be safe. =Rob