[
https://issues.apache.org/jira/browse/HBASE-24134?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17077538#comment-17077538
]
Michael Stack commented on HBASE-24134:
---------------------------------------
Running the change locally first, I ran into TestMultiRespectsLimits. It kept
failing. I hooked it up to mission control and could see that unfettered, it
kept going to 2.5G at least. I made below change and now its well behaved well
under 1G.
{code}
diff --git
a/hbase-server/src/test/java/org/apache/hadoop/hbase/client/TestMultiRespectsLimits.java
b/hbase-server/src/test/java/org/apache/hadoop/hbase/client/TestMultiRespectsLimits.java
index a460fee10f..d8baa8d1cb 100644
---
a/hbase-server/src/test/java/org/apache/hadoop/hbase/client/TestMultiRespectsLimits.java
+++
b/hbase-server/src/test/java/org/apache/hadoop/hbase/client/TestMultiRespectsLimits.java
@@ -64,7 +64,7 @@ public class TestMultiRespectsLimits {
private static final MetricsAssertHelper METRICS_ASSERT =
CompatibilityFactory.getInstance(MetricsAssertHelper.class);
private final static byte[] FAMILY = Bytes.toBytes("D");
- public static final int MAX_SIZE = 500;
+ public static final int MAX_SIZE = 100;
@Rule
public TestName name = new TestName();
{code}
> Down forked JVM heap size from 2800m to 2200m for jdk8 and jdk11
> ----------------------------------------------------------------
>
> Key: HBASE-24134
> URL: https://issues.apache.org/jira/browse/HBASE-24134
> Project: HBase
> Issue Type: Sub-task
> Reporter: Michael Stack
> Priority: Major
>
> Docker container we run tests in is hardcoded 20G. Our fork count is 0.25C
> currently. The jenkins machines are 16CPUs. 0.25C means 4CPUs so 4 forks
> running at any one time. Our current forked JVM heap size is 2800m if jdk8 or
> 3200m if jdk11. If we want to move our fork count to 0.5C, then that means 8
> forks possibly running at once: i.e. 8 * 2800 = 22.4G... i.e. too big for the
> container. This issue is about running w/ smaller forked jvm heap size. If
> 2200m then, its 19.6G which is cutting it close but might allow us sqeeze
> through (mvn takes 4G!). Lets see.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
