[
https://issues.apache.org/jira/browse/HBASE-919?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12641066#action_12641066
]
Jean-Daniel Cryans commented on HBASE-919:
------------------------------------------
I think it's related to the fact that HBase should have a "safe mode". Jim's
first patch added this:
{code}
public HServerAddress findRootRegion() {
- return regionManager.getRootRegionLocation();
+ if (regionManager.allRegionsAssigned()) {
+ rootServer = regionManager.getRootRegionLocation();
+ }
+ return rootServer;
{code}
This is a kind of safe mode, but it has implications. To be very precise about
your comment stack, I would say that anyone who wants to use an HTable right
after HBase is started (and I do mean the second after) must wait a moment
before creating a HTable in the same process (or JVM, not sure) or it will
deadlock. Anyone outside is OK because they don't go inside the same
synchronized block in HCM.
> Master and Region Server need to provide root region location if they are
> using HTable
> --------------------------------------------------------------------------------------
>
> Key: HBASE-919
> URL: https://issues.apache.org/jira/browse/HBASE-919
> Project: Hadoop HBase
> Issue Type: Sub-task
> Components: master, regionserver
> Affects Versions: 0.18.0
> Reporter: Jim Kellerman
> Assignee: Jim Kellerman
> Priority: Blocker
> Fix For: 0.19.0
>
> Attachments: 919.patch, 919.patch
>
>
> There needs to be a mechanism by which master and region servers can provide
> the HServerAddress for the root region to HConnectionManager so that they can
> use a HTable.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
