Hi David, bq. Too many regions can easily burden the hbase cluster heavily, even if empty region Is that true, HBase-committers?
You'll need to delete the SYSTEM.SEQUENCE, make sure to set the phoenix.sequence.saltBuckets property, and bounce your cluster. Then the next time you try to establish a connection to your cluster, the SYSTEM.SEQUENCE table will be re-created based on that property. The reason we've salted this table in this manner is to spread the load of getting a batch of new sequence values. If it's not pre-split, then one region server would handle all of the allocation of new sequence values which we wanted to avoid. Since the sequence table is not scanned, but rather point lookups are performed to find the sequence row for a given sequence, we salted it with the maximum number of buckets. Thanks, James On Tue, Dec 23, 2014 at 5:48 PM, chenwenhui <[email protected]> wrote: > Hi James, > Too many regions can easily burden the hbase cluster heavily, even if empty > region, and also i think the SYSTEM.SEQUENCE table indeed need not so much > regions. > After modify the phoenix.sequence.saltBuckets, and then restart hbase, the > region number of SYSTEM.SEQUENCE table does not become less in the test > cluster. > Thanks > David chen > >
