On Wed, Aug 21, 2013 at 10:40 PM, lars hofhansl <la...@apache.org> wrote:
> Could one say that 0.98 is to 0.96 what 0.94 is to 0.92? > Are we planning the same compatibility guarantees - i.e. no downtime > during an upgrade from any 0.96.x to any 0.98.x release? Sort of. A 0.96 client should be able to talk to a 0.98 server, and a mixed 0.96 and 0.98 server environment should be possible as long as the configuration doesn't change. I f we can introduce HBASE-6721 into 0.96 at some point, then you could set up placement groups first, and then deploy 0.98 to one placement group, online migrate the table(s) placed there, stay in this configuration for days or weeks, and the rest of the cluster and all of the clients would be fine with that. A new feature should be implemented within a well defined space for that purpose: as a coprocessor, plugin, or as a feature of or change to HFile V3. It could be necessary to do the placement group trick or otherwise complete the rolling upgrade first before turning something on (including HFile V3). The difference is Jon has suggested API cleanups. We could have a result where 0.96 clients will run just fine against 0.98 servers but when you want to upgrade HBase jars for an application then a recompilation would be needed. I think that's reasonable until the 1.0 release. -- Best regards, - Andy Problems worthy of attack prove their worth by hitting back. - Piet Hein (via Tom White)