HBase is a is a consistent store CP (in the parlor of the CAP theorem). Availability is explicitly traded for consistency. If availability is more important to you than consistency you should look at one of the many eventually consistent (CA) stores
out there (Riak, Cassandra, etc) or wait for HBASE-10070, which will allow some eventual consistency in HBase as well. In those cases you have to be well aware and prepared for temporary inconsistencies. -- Lars ________________________________ From: Jeff Storey <[email protected]> To: [email protected] Sent: Subject: Re: Region server failover Thanks for the quick reply. It looks like this feature is in the works. What are people doing in the absence of this feature? Simply allowing the downtime? On Thu, Mar 13, 2014 at 4:51 PM, Ted Yu <[email protected]> wrote: > Take a look at HBASE-10070: > HBase read high-availability using eventually consistent region replicas > > Cheers > > > On Thu, Mar 13, 2014 at 1:49 PM, Jeff Storey <[email protected]> > wrote: > > > I'm still somewhat new to HBase and I'm trying to understand > possibilities > > for region server failover. If I have a region server that goes down, it > > looks like I won't have any access to that region's data until a backup > > server comes online. I read that I can reduce the zookeeper timeout to > > minimize that time down, but in a system that is meant to be highly > > available, it seems like this could be cause for concern since a portion > of > > the data will be entirely inaccessible. > > > > Is there a way to have a secondary read-only region server that becomes > the > > new primary region server in the event of a failure? This way that data > > will always be accessible unless all of the region server backups fail? > Or > > are there other ways to replicate region data to avoid loss of access for > > possibly several minutes to that data if the server goes down? > > > > Thanks, > > Jeff > > >
