Just as a theoretical interest, do we support online merging of Regions in any way? Are there ways to merge Regions while still supporting reads and writes to them?
If not, can we do them in the following manner:
At each major delete/explicit order from user to merge Regions, we can create a new empty memstore which will take any reads for the Region in discussion. Then, we can build a leftist tree of the Region and store it in the memory itself. We so not need to write it into disk yet. This shall also allow us to serve reads using the Region itself without any extra additional copies.
We do it for all the Regions being merged and then merge the leftist trees made ordered by the range of keys served by the Regions. We then traverse the final all merged leftist tree, write its data to a HFile and write the new memstore being used for write's data into the HFile as well.
It's just a thought.Please let me know your feedback and comments on it.
It is in 0.95/0.96/0.98 releases. On Sat, Mar 1, 2014 at 12:01 PM, Atri Sharma <[EMAIL PROTECTED]> wrote: // Jonathan Hsieh (shay) // HBase Tech Lead, Software Engineer, Cloudera // [EMAIL PROTECTED] // @jmhsieh
You are very welcome to contribute. The best place to start is JIRA. Search for open JIRAs where no-one has yet contributed, and look at it. There is some easier than others. Just take one which you feel comfortable with.
JM 2014-03-01 22:00 GMT-05:00 Atri Sharma <[EMAIL PROTECTED]>: