[
https://issues.apache.org/jira/browse/HBASE-1909?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12766224#action_12766224
]
Jonathan Gray commented on HBASE-1909:
--------------------------------------
There is a very misleading comment:
{noformat}
int nRegionsToAssign = regionsToAssign.size();
// Now many regions to assign this server.
int nregions = regionsPerServer(nRegionsToAssign, thisServersLoad);
{noformat}
In reality, what's returned by regionsPerServer is the number of regions that
should be assigned to less-loaded servers (javadoc of that method is right)
{noformat}
if (nRegionsToAssign <= nregions) {
// I do not know whats supposed to happen in this case. Assign one.
LOG.debug("Assigning one region only (playing it safe..)");
assignRegions(regionsToAssign, 1, info, returnMsgs);
} else {
nRegionsToAssign -= nregions;
if (nRegionsToAssign > 0 || isMetaAssign) {
{noformat}
HBASE-1868 made it so if nRegionsToAssign <= nregions, we just assign one.
This was to prevent conditions where we didn't assign and a region got left in
the abyss.
Need to study what was happening in HBASE-1868...
> Balancer assigns to most loaded server when assigning a single region
> ---------------------------------------------------------------------
>
> Key: HBASE-1909
> URL: https://issues.apache.org/jira/browse/HBASE-1909
> Project: Hadoop HBase
> Issue Type: Bug
> Affects Versions: 0.20.1
> Reporter: Jonathan Gray
> Attachments: unbalanced.txt
>
>
> fake on irc found a bug where he truncated his tables and they all ended up
> getting assigned to the same regionserver.
> 4 total nodes, root and meta each ended up on one node, another node had 0
> regions, and the final one ended up with all 5 regions of user tables.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
