hbase-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Tianying Chang (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (HBASE-7060) region load balancing by table does not handle the case where a table's region count is less than the number of the RS in the cluter
Date Mon, 29 Oct 2012 23:04:12 GMT

     [ https://issues.apache.org/jira/browse/HBASE-7060?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Tianying Chang updated HBASE-7060:
----------------------------------

     Description: 
When the table's region count is less than the count of region servers, the region balance
algorithm will not move the region. For example, the cluster has 100 RS, the table has 50
regions sitting on one RS, they will not be moved to any of the other 99 RS.

This is because the algorithm did not calculate the under-loaded RS correctly. This is how
the algorithm works with the above example:

avg-regions-per-RS=0.5
min-RS-per-RS=0
max-RS-per-RS=1

when they calculate the under loaded RS, the code is as below. Since regionCount=0, which
is always >=min, so it will always skip, therefore, no underloaded RS are found.

Map<ServerName, Integer> underloadedServers = new HashMap<ServerName, Integer>();
for (Map.Entry<ServerAndLoad, List<HRegionInfo>> server:
serversByLoad.entrySet()) {
int regionCount = server.getKey().getLoad();
if (regionCount >= min) { break; }
underloadedServers.put(server.getKey().getServerName(), min - regionCount);
}

Later the function returns since underloaded RS size is 0

if (serverUnerloaded ==0) return regionsToReturn;


  was:

When the table's region count is less than the count of region servers, the region balance
algorithm will not move the region. For example, the cluster has 100 RS, the table has 50
regions sitting on one RS, they will not be moved to any of the other 99 RS.

This is because the algorithm did not calculate the under-loaded RS correctly. This is how
the algorithm works with the above example:

avg-regions-per-RS=0.5
min-RS-per-RS=0
max-RS-per-RS=1

when they calculate the under loaded RS, the code is as below. Since regionCount=0, which
is always >=min, so it will always skip, therefore, no underloaded RS are found.

Map<ServerName, Integer> underloadedServers = new HashMap<ServerName, Integer>();
for (Map.Entry<ServerAndLoad, List<HRegionInfo>> server:
serversByLoad.entrySet()) {
int regionCount = server.getKey().getLoad();
if (regionCount >= min) { break; }
underloadedServers.put(server.getKey().getServerName(), min - regionCount);
}

Later the function returns since underloaded RS size is 0

if (serverUnerloaded ==0) return regionsToReturn;


    Release Note: 
@ted 
>  The above would produce unnecessarily high number of underloaded servers, right ?

I think it shouldn't. Because when the code reach here, it means this server is an under-loaded
RS. Therefore, it is not procuding un-necessary underloaded server.  For the special case
when min=1, max=1, we want to shed the extra regions to those RS what has regionCount of 0.
And the regionToTake for those RS should be 1. 
    
> region load balancing by table does not handle the case where a table's region count
is less than the number of the RS in the cluter
> ------------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: HBASE-7060
>                 URL: https://issues.apache.org/jira/browse/HBASE-7060
>             Project: HBase
>          Issue Type: Bug
>          Components: master
>    Affects Versions: 0.92.0
>            Reporter: Tianying Chang
>            Assignee: Tianying Chang
>         Attachments: HBASE-7060.patch
>
>
> When the table's region count is less than the count of region servers, the region balance
algorithm will not move the region. For example, the cluster has 100 RS, the table has 50
regions sitting on one RS, they will not be moved to any of the other 99 RS.
> This is because the algorithm did not calculate the under-loaded RS correctly. This is
how the algorithm works with the above example:
> avg-regions-per-RS=0.5
> min-RS-per-RS=0
> max-RS-per-RS=1
> when they calculate the under loaded RS, the code is as below. Since regionCount=0, which
is always >=min, so it will always skip, therefore, no underloaded RS are found.
> Map<ServerName, Integer> underloadedServers = new HashMap<ServerName, Integer>();
> for (Map.Entry<ServerAndLoad, List<HRegionInfo>> server:
> serversByLoad.entrySet()) {
> int regionCount = server.getKey().getLoad();
> if (regionCount >= min) { break; }
> underloadedServers.put(server.getKey().getServerName(), min - regionCount);
> }
> Later the function returns since underloaded RS size is 0
> if (serverUnerloaded ==0) return regionsToReturn;

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message