[
https://issues.apache.org/jira/browse/HBASE-16375?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15446333#comment-15446333
]
Andrew Purtell commented on HBASE-16375:
----------------------------------------
The change seems ok to me, no less magic than other YARN / MR config wrangling.
bq. I'd say we need a test added in hbase to ensure we keep doing the correct
thing
Agreed, as long as the test doesn't cause more trouble than it's worth. We
compile HBase 0.98 by default with Hadoop 2.2.0. Meanwhile trunk requires
Hadoop 2.7.1. Can a test cover correct MR job submission through all of those
different versions?
> Mapreduce mini cluster using HBaseTestingUtility not setting correct
> resourcemanager and jobhistory webapp address of MapReduceTestingShim
> --------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-16375
> URL: https://issues.apache.org/jira/browse/HBASE-16375
> Project: HBase
> Issue Type: Bug
> Affects Versions: 1.2.0
> Reporter: Loknath Priyatham Teja Singamsetty
> Assignee: Loknath Priyatham Teja Singamsetty
> Priority: Minor
> Fix For: 2.0.0, 1.2.0, 1.3.0, 1.4.0, 1.2.2, 1.1.6, 1.3.1, 1.2.3,
> 0.98.22, 1.2.4
>
> Attachments: HBASE-16375_0.98_and_above.patch
>
>
> Starting mapreduce mini cluster using HBaseTestingUtility is not setting
> "yarn.resourcemanager.webapp.address" and
> "mapreduce.jobhistory.webapp.address" which are required for getting the
> submitted yarn apps using mapreduce webapp. These properties are not being
> copied from jobConf of MapReduceTestingShim resulting in default values.
> {quote}
> HBaseTestingUtility.java
> // Allow the user to override FS URI for this map-reduce cluster to use.
> mrCluster = new MiniMRCluster(servers,
> FS_URI != null ? FS_URI : FileSystem.get(conf).getUri().toString(), 1,
> null, null, new JobConf(this.conf));
> JobConf jobConf = MapreduceTestingShim.getJobConf(mrCluster);
> if (jobConf == null) {
> jobConf = mrCluster.createJobConf();
> }
> jobConf.set("mapreduce.cluster.local.dir",
> conf.get("mapreduce.cluster.local.dir")); //Hadoop MiniMR overwrites
> this while it should not
> LOG.info("Mini mapreduce cluster started");
> // In hadoop2, YARN/MR2 starts a mini cluster with its own conf instance
> and updates settings.
> // Our HBase MR jobs need several of these settings in order to properly
> run. So we copy the
> // necessary config properties here. YARN-129 required adding a few
> properties.
> conf.set("mapreduce.jobtracker.address",
> jobConf.get("mapreduce.jobtracker.address"));
> // this for mrv2 support; mr1 ignores this
> conf.set("mapreduce.framework.name", "yarn");
> conf.setBoolean("yarn.is.minicluster", true);
> String rmAddress = jobConf.get("yarn.resourcemanager.address");
> if (rmAddress != null) {
> conf.set("yarn.resourcemanager.address", rmAddress);
> }
> String historyAddress = jobConf.get("mapreduce.jobhistory.address");
> if (historyAddress != null) {
> conf.set("mapreduce.jobhistory.address", historyAddress);
> }
> String schedulerAddress =
> jobConf.get("yarn.resourcemanager.scheduler.address");
> if (schedulerAddress != null) {
> conf.set("yarn.resourcemanager.scheduler.address", schedulerAddress);
> }
> {quote}
> As a immediate fix for phoenix e2e test to succeed, need the below lines to
> be added as well
> {quote}
> String rmWebappAddress =
> jobConf.get("yarn.resourcemanager.webapp.address");
> if (rmWebappAddress != null) {
> conf.set("yarn.resourcemanager.webapp.address", rmWebappAddress);
> }
> String historyWebappAddress =
> jobConf.get("mapreduce.jobhistory.webapp.address");
> if (historyWebappAddress != null) {
> conf.set("mapreduce.jobhistory.webapp.address", historyWebappAddress);
> }
> {quote}
> Eventually, we should also see if we can copy over all the jobConf properties
> to HBaseTestingUtility conf object.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
