rmdmattingly commented on PR #6593: URL: https://github.com/apache/hbase/pull/6593#issuecomment-2581157980
Hey @Apache9, I see your name in a lot of the StochasticLoadBalancer git blame so I wanted to reach out in case you're interested here. I know this is a pretty big changeset, so I'd be happy to break it up if you'd like, but the individual components are pretty nicely divided between: 1. Some changes to the StochasticLoadBalancer, mostly minor changes or gated behind conditional usage 2. The replica distribution conditional 3. The system table isolation conditional 4. The meta table isolation conditional 5. System table isolation candidate generators 6. Table colocation candidate generators 7. The replica distribution candidate generator 8. And finally, a robust test suite I think this will be a meaningful improvement for hbase's load balancer. We envision this being useful in: 1. Better distributing read replicas, while also keeping the balancer simple to configure based on storefile size. I dig into it above, but the existing stuff doesn't isolate secondaries well and it's difficult to balance replica multipliers well with everything else. In comparison to the replica cost functions, I think this conditionals stuff is quite simple 2. System table isolation without needing to manage wastefully over-provisioned RegionServer groups 3. Likewise, simple/cheap meta table isolation 4. We also have a large project in the early stages which would involve colocating regions based on a prefix in their start/stop keys. This would allow us to flexibly control logical groupings of our customers' data so that we could pull a lever to, for example, increase isolation between customers in an otherwise randomly distributed table so that a single RegionServer outage would only affect some smaller proportion of all customers -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@hbase.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
