Thanks Ted.
I will check if I could get some resources to contribute.
Yong
On Sat, Jan 3, 2015 at 2:13 PM, Ted Yu yuzhih...@gmail.com wrote:
YARN-3003 was logged.
But it is not known when it will be implemented.
Yong:
Please consider using zookeeper for recoding master nodes. This makes
Thanks Ted.
Basically what we are doing is to ask HPC workload and BigData workload
share the same infrastructure by Slider. The current way to fit Hadoop into
HPC community is either on demand provisioning Hadoop cluster by HPC
scheduler (for example HOD or myHadoop) or enhancing Hadoop as HPC
Jon, Ted
Happy new year.
After read more, I understand knowing the host list or single host of
master role in Slider script or template is not supported by either Slider
or YARN.
So what I can do is
a) pre-define a group of hosts with label of master
b) specify label of master for master role
Hi John
I do expect the hosts allocated to master role of my slider application are
labeled as appname_rolename, but the question is who will add such label
and how? and how to get the hosts with the label?
Thanks,
Yong
On Wed, Dec 31, 2014 at 3:57 PM, Jon Maron jma...@hortonworks.com wrote:
Hi Team,
The template of regionserver of HBase application uses following scripts to
configure configuraiton file of regionserver of HBase. I did not find where
the rs_hosts is defined in any file in Slider HBase package. How a template
is designed is mentioned a little at the end of
bq. host list of all the instances of a component
Are you using node labels feature of YARN ?
If not, the list would comprise of all the nodes in your YARN cluster.
Cheers
On Tue, Dec 30, 2014 at 4:43 PM, Yong Feng fengyong...@gmail.com wrote:
Hi Team,
The template of regionserver of HBase
ah, rs_hosts means all the nodes managed by YARN.
I know a little about labels feature of YARN which adds a tag to host and
allow yarn API client to filter hosts according to the tag. I don't think
it helps my case.
Taking a typical master/slave cluster as an example. To support
active-passive
What's the timeframe for this migration ?
If time permits, you should consider rewriting master failover logic in
your application.
Using zookeeper is a common technique. HBase, e.g., registers active master
under a well known zookeeper znode.
This is much more flexible than the way you