[
https://issues.apache.org/jira/browse/HBASE-6721?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14662670#comment-14662670
]
Guorui Wu commented on HBASE-6721:
----------------------------------
As I've been studying this code, in order to aid with the understanding of
HBASE-6721, I’ve created some UML sequence diagrams of functions within
GroupBasedLoadBalancer.java. This UML diagram only focuses on
GroupBasedLoadBalancer, since GroupBasedLoadBalancer, along with
GroupInfoManager, GroupInfoManagerImpl and GroupInfo as well as some way to
configure groups represent the core functionality needed for this
implementation. This represents the 1,500 core lines implementing the region
server group functionality; this documentation does not cover the extensive
configuration management via CLI stored in Zookeeper and an HBase table.
The goal of {{GroupBasedLoadBalancer}} is to separate servers and regions into
groups, and to balance within groups using another load balancer (referred to
as internal load balancer). The default load balancer used is
{{StochasticLoadBalancer}}. Each group is stored as a {{GroupInfo}} object, and
{{GroupInfoManagerImpl}} holds a collection of {{GroupInfo}} objects with APIs
allowing for the manipulation of this collection. These groups can be created
and configured via the CLI, and are stored within an HBase table as well as
Zookeeper. As {{GroupBasedLoadBalancer}} implements the {{LoadBalancer}}
interface, it has the following functions from the interface:
As GroupBasedLoadBalancer implements the LoadBalancer interface, it has the
following functions from the interface:
• balanceCluster
• roundRobinAssignment
• retainAssignment
• immediateAssignment
• randomAssignment
• initialize
• isStopped
• setGroupInfoManager
Additionally, GroupBasedLoadBalancer has several helper functions:
• offlineRetainAssignment
• onlineRetainAssignment
• generateGroupMaps
• filterOfflineServers
• filterServers
• getMisplacedRegions
• correctAssignments
• isOnline
• getGroupInfoManager
Through reading the code for the load balancer portion of HBASE-6721, I have a
few questions:
1. Within the interface GroupInfoManager.java, I noticed that the function
getGroupOfServer returns a GroupInfo object, but the function getGroupOfTable
returns a String object. Was there a performance consideration or some other
reason for returning a string? (It seems the API would be more consistent to
return a GroupInfo object.)
2. For the function onlineRetainAssignment why are regions assigned to
bogus so it ends up in RIT if a server is not available? (We would like to keep
as few regions in RIT as possible in order to maximize our availability.)
3. On the topic of onlineRetainAssignment, what is the objective for
separating online and offline servers? I noticed that another balancer such as
the StochasticLoadBalancer does not make such a distinction.
Lastly, the UML diagram I created can be edited by downloading the attached XML
file and editing with http://draw.io.
> RegionServer Group based Assignment
> -----------------------------------
>
> Key: HBASE-6721
> URL: https://issues.apache.org/jira/browse/HBASE-6721
> Project: HBase
> Issue Type: New Feature
> Reporter: Francis Liu
> Assignee: Francis Liu
> Attachments: 6721-master-webUI.patch, HBASE-6721-DesigDoc.pdf,
> HBASE-6721-DesigDoc.pdf, HBASE-6721-DesigDoc.pdf, HBASE-6721-DesigDoc.pdf,
> HBASE-6721_10.patch, HBASE-6721_11.patch, HBASE-6721_8.patch,
> HBASE-6721_9.patch, HBASE-6721_9.patch, HBASE-6721_94.patch,
> HBASE-6721_94.patch, HBASE-6721_94_2.patch, HBASE-6721_94_3.patch,
> HBASE-6721_94_3.patch, HBASE-6721_94_4.patch, HBASE-6721_94_5.patch,
> HBASE-6721_94_6.patch, HBASE-6721_94_7.patch, HBASE-6721_trunk.patch,
> HBASE-6721_trunk.patch, HBASE-6721_trunk.patch, HBASE-6721_trunk1.patch,
> HBASE-6721_trunk2.patch
>
>
> In multi-tenant deployments of HBase, it is likely that a RegionServer will
> be serving out regions from a number of different tables owned by various
> client applications. Being able to group a subset of running RegionServers
> and assign specific tables to it, provides a client application a level of
> isolation and resource allocation.
> The proposal essentially is to have an AssignmentManager which is aware of
> RegionServer groups and assigns tables to region servers based on groupings.
> Load balancing will occur on a per group basis as well.
> This is essentially a simplification of the approach taken in HBASE-4120. See
> attached document.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
