Ning Li wrote:
Or, hbase allows specifying the class that implements
regionserver -- could facade current regionserver with an indexing
regionserver.
Could you elaborate? Is this the coupled approach? Meaning
each region maintains a corresponding index - on the same
node and almost-synchronized?
The coupled approach would be modifying the current regionserver
implemenation to add keeping-up an index.
The suggestion above is less tightly-coupled. It suggests that you
subclass regionserver to add the new facility. Your subclass would be
called, say, IndexingRegionServer. To have hbase use this regionserver
implementation instead of the current
org.apache.hadoop.hbase.regionserver.HRegionServer, you'd set the
hbase-default.xml property hbase.regionserver.class to be your
implementation instead.
I was just pointing out this as an option.
Tightly-coupled with the feature disabled by default with a simple
switch to turn on the facility is what would make most happy I'm guessing.
Thanks,
St.Ack
