[jira] Resolved: (HBASE-2327) [EC2] Allocate elastic IP addresses for ZK and master nodes

["Andrew Purtell (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/HBASE-2327?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Andrew Purtell resolved HBASE-2327.
-----------------------------------

    Resolution: Fixed

Tested a few times. Works ok. I'm a bit concerned about what would happen if the elastic IP association takes too long. Committed to trunk and 0.20 branch. We can revisit if there are problems in practice. 

> [EC2] Allocate elastic IP addresses for ZK and master nodes
> -----------------------------------------------------------
>
>                 Key: HBASE-2327
>                 URL: https://issues.apache.org/jira/browse/HBASE-2327
>             Project: Hadoop HBase
>          Issue Type: Sub-task
>            Reporter: Andrew Purtell
>            Assignee: Andrew Purtell
>             Fix For: 0.20.4, 0.21.0
>
>         Attachments: HBASE-2327.patch
>
>
> Amazon EC2 supports Elastic IP Addresses to implement the effect of having a static IP address for public servers running on EC2. Up on hbase-users@ there was some recent discussion, confirmed, that when an EC2 instance queries the external DNS name of an elastic IP address, EC2 DNS returns the internal IP address of the instance to which the elastic IP address is bound, so it is safe to use elastic IPs for the ZK and master nodes. We gain the ability to do transparent replacement of one instance, e.g. failed, with another without incurring any additional cost. 
> Update {{launch-hbase-zookeeper}} and {{launch-hbase-master}} to allocate elastic IPs:
> {noformat}
>   $ ec2-allocate-address 
>   ADDRESS 1.1.1.1
> {noformat}
> and then assign the elastic IP address to the appropriate instance(s):
> {noformat}
> $ ec2-associate-address -i i-11111111 1.1.1.1
> ADDRESS 1.1.1.1  i-11111111
> {noformat}
> and then get the external DNS name to use when performing substitutions on master and slave configs:
> {noformat}
> $ ec2-describe-instances i-11111111 | egrep ^INSTANCE | cut -f4
> ec2-1-1-1-1.compute-1.amazonaws.com
> {noformat}
> When shutting down the cluster, just release the elastic IPs after terminating the instances:
> {noformat}
> ec2-release-address 1.1.1.1
> ...
> {noformat}
> NOTE: AWS accounts default to a limit of 5 Elastic IP addresses but most will run with 1 master and 3 or 1 ZK instances. And, the ZK ensemble can be shared. 
> A follow up issue can address making scripts to launch replacements for failed instances transparently. 

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.