gistfile1.txt

***From Riak IRC 5/6/2010***

chids # re: node data migration - if I understood the mailing list discussion correct issuing a "riak-admin leave" will not currently migrate the data away from the node. Instead I should attach to the console as per Grant's email. Correct?

chids # So "riak-admin leave" is currently a no-op?

chids # k, thx About the "Grant approach" is it considered solid?

hemulen # lemme look at that mail again. onesec. that looks correct, yes
you'll have to monitor the data directory to make sure all of the data has been migrated

chids # checK

chids # And of course, migrating data is only needed if my N value doesn't cater for removing one node. In combination with clients usage of R, W, DW

hemulen # chids: leave is useful when you want to permanent remove a node from the cluster
errr, permanently remove a node

chids # Say I was to upgrade a node and change storage backend at the same time - what'd you recommend

hemulen # if your N was high enough you could get away with taking the node down gracefully, doing the change, and restarting it
otherwise you'd want to do the leave, migrate data, update node, and join dance

chids # And there's no interop problems between nodes of different versions? (repl protocol, m/r distribution etc)
"different versions" being a bit vauge there

hemulen # well
i don't think we test with a cluster running different versions of riak not sure what that'd look like, actually
hmmm. one sec

hemulen # i stand slightly corrected after consulting the Basho Brain Trust

hemulen # in general, rolling cluster upgrades from one version of riak to another is fine we try to warn in release notes if there are specific issues to worry about

chids # "we try to warn", hehe. Noted.

chids # The scenario here being me considering an upgrade from v0.9 to v0.10 for a 3 node cluster

hemulen # chids: the bugaboo with that upgrade is the changes in module names to reflect the core/kv split

chids # yeah, would that cause any trouble other than me having to fix the config files?

hemulen # config and possibly m/r jobs if you're using erlang functions in them

chids # Config is cool, don't have any m/r jobs running in this cluster, and no usage of links

hemulen # you _should_ be good, then but i'd do it on a test cluster first :)

chids # "Warranty void if done untested"