[ https://issues.apache.org/jira/browse/HBASE-18511?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16126642#comment-16126642 ]
stack edited comment on HBASE-18511 at 8/15/17 4:17 AM: -------------------------------------------------------- .007 changes hbase.balancer.tablesOnMaster to be boolean. If enabled, master is like any other regionserver. If false -- the default -- then the master carries no tables. A new config, hbase.balancer.tablesOnMaster.systemTablesOnly, when enabled makes it so master carries system tables. Adds a test that tries the three configs along w/ some master killing to make sure the deploy-type prevails past master kill. A bunch of other tests are changed. The old deploy mode of master exclusively carrying system tables was around so long, a bunch of tests came to depend on this layout. Coarse test changes were made so deal w/ the new deploy mode. For a few tests, they don't make sense in the new deploy mode. On why I changed the type of the flag, as is, the implication was that you could specify any set of tables for Master to host -- it did not have to be system tables only. Also you had to hand specify the system table set. On the former, already have a well-developed mechanism for matching tables to server sets, it is called rsgroup. This mechanism on master was an implementation of a subset of rsgroup. HBase is hard enough already w/o two means to same end. Given system tables on master was a config some of us require, a simple flag to enable this exception seemed way to go. Oh, as part of testing I verified that we are indeed doing short-circuit RPC if we find a region colocated. [~zyork] You ok w/ this sir? Anyone up for a review? This is last one before alpha2. Thanks. was (Author: stack): .007 changes hbase.balancer.tablesOnMaster to be boolean. If enabled, master is like any other regionserver. If false -- the default -- then the master carries no tables. A new config, hbase.balancer.tablesOnMaster.systemTablesOnly, when enabled makes it so master carries system tables. Adds a test that tries the three configs along w/ some master killing to make sure the deploy-type prevails past master kill. A bunch of other tests are changed. The old deploy mode of master exclusively carrying system tables was around so long, a bunch of tests came to depend on this layout. Coarse test changes were made so deal w/ the new deploy mode. For a few tests, they don't make sense in the new deploy mode. [~zyork] You ok w/ this sir? Anyone up for a review? This is last one before alpha2. Thanks. > Default no regions on master > ---------------------------- > > Key: HBASE-18511 > URL: https://issues.apache.org/jira/browse/HBASE-18511 > Project: HBase > Issue Type: Task > Components: master > Reporter: stack > Assignee: stack > Priority: Blocker > Fix For: 2.0.0 > > Attachments: HBASE-18511.master.001.patch, > HBASE-18511.master.002.patch, HBASE-18511.master.003.patch, > HBASE-18511.master.004.patch, HBASE-18511.master.005.patch, > HBASE-18511.master.006.patch, HBASE-18511.master.007.patch > > > Let this be umbrella issue for no-regions-on-master as default deploy (as it > was in branch-1). > Also need to make sure we can run WITH regions on master; in particular > system tables with RPC short-circuit as it is now in hbase master. > Background is that master branch carried a change that allowed Master carry > regions. On top of this improvement on branch-1, Master defaulted to carry > system tables only. No release was made with this configuration. Now we are > going to cut the 2.0.0 release, the decision is that hbase-2 should have the > same layout as hbase-1 so this issue implements the undoing of Master > carrying system tables by default (though the capability remains). -- This message was sent by Atlassian JIRA (v6.4.14#64029)