This is an automated email from the ASF dual-hosted git repository.
Apache9 pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/hbase.git
The following commit(s) were added to refs/heads/master by this push:
new 9af77151403 HBASE-30119 Stop JmxCacheBuster in HBTU (#8145)
9af77151403 is described below
commit 9af771514036da3db37da23381d7aa403f0e0f5f
Author: Duo Zhang <[email protected]>
AuthorDate: Tue Apr 28 09:27:52 2026 +0800
HBASE-30119 Stop JmxCacheBuster in HBTU (#8145)
Signed-off-by: Xiao Liu <[email protected]>
---
.../org/apache/hadoop/hbase/HBaseTestingUtil.java | 19 +++++++++++++++++++
.../procedure/TestCreateTableNoRegionServer.java | 8 --------
2 files changed, 19 insertions(+), 8 deletions(-)
diff --git
a/hbase-server/src/test/java/org/apache/hadoop/hbase/HBaseTestingUtil.java
b/hbase-server/src/test/java/org/apache/hadoop/hbase/HBaseTestingUtil.java
index 365855bbd3c..22e19a31fd7 100644
--- a/hbase-server/src/test/java/org/apache/hadoop/hbase/HBaseTestingUtil.java
+++ b/hbase-server/src/test/java/org/apache/hadoop/hbase/HBaseTestingUtil.java
@@ -153,6 +153,7 @@ import
org.apache.hadoop.hdfs.server.datanode.fsdataset.FsDatasetSpi;
import org.apache.hadoop.hdfs.server.namenode.EditLogFileOutputStream;
import org.apache.hadoop.mapred.JobConf;
import org.apache.hadoop.mapred.MiniMRCluster;
+import org.apache.hadoop.metrics2.impl.JmxCacheBuster;
import org.apache.hadoop.minikdc.MiniKdc;
import org.apache.yetus.audience.InterfaceAudience;
import org.apache.yetus.audience.InterfaceStability;
@@ -214,6 +215,24 @@ public class HBaseTestingUtil extends HBaseZKTestingUtil {
/** This is for unit tests parameterized with a single boolean. */
public static final List<Object[]> MEMSTORETS_TAGS_PARAMETRIZED =
memStoreTSAndTagsCombination();
+ static {
+ // JmxCacheBuster may cause dead lock in test environment. As on master
side, the table/region
+ // related metrics updating will finally lead to a meta access, so if meta
is not online yet, we
+ // will block when updating while holding the metrics lock. But when we
assign meta, there are
+ // bunch of places where we need to register a new metrics thus need to
get the metrics lock,
+ // and then lead to a dead lock and cause the test to hang forever.
+ // The code is in hadoop so there is no easy way for us to fix, so here we
just stop
+ // JmxCacheBuster to stabilize our tests first. See HBASE-30118 for more
details and future
+ // plans.
+ try {
+ JmxCacheBuster.stop();
+ } catch (NullPointerException e) {
+ // Expected, the scheduled job has not been initialized so it will throw
a NPE, but before
+ // throwing we have already set the stopped flag to false. Since there
is no public method to
+ // only set stopped flag to false, we use this try-catch trick here to
avoid using reflection.
+ }
+ }
+
/**
* Checks to see if a specific port is available.
* @param port the port number to check for availability
diff --git
a/hbase-server/src/test/java/org/apache/hadoop/hbase/master/procedure/TestCreateTableNoRegionServer.java
b/hbase-server/src/test/java/org/apache/hadoop/hbase/master/procedure/TestCreateTableNoRegionServer.java
index 5119133e7b0..e1f8d925b43 100644
---
a/hbase-server/src/test/java/org/apache/hadoop/hbase/master/procedure/TestCreateTableNoRegionServer.java
+++
b/hbase-server/src/test/java/org/apache/hadoop/hbase/master/procedure/TestCreateTableNoRegionServer.java
@@ -38,7 +38,6 @@ import org.apache.hadoop.hbase.procedure2.Procedure;
import org.apache.hadoop.hbase.testclassification.MasterTests;
import org.apache.hadoop.hbase.testclassification.MediumTests;
import org.apache.hadoop.hbase.util.Bytes;
-import org.apache.hadoop.metrics2.impl.JmxCacheBuster;
import org.junit.jupiter.api.AfterAll;
import org.junit.jupiter.api.BeforeAll;
import org.junit.jupiter.api.Tag;
@@ -112,13 +111,6 @@ public class TestCreateTableNoRegionServer {
public static void setUp() throws Exception {
UTIL.startMiniCluster(
StartTestingClusterOption.builder().masterClass(HMasterForTest.class).build());
- // this may cause dead lock if there is no live region server and want to
start a new server.
- // In JmxCacheBuster we will reinitialize the metrics system so it will
get some metrics which
- // will need to access meta, since there is no region server, the request
will hang there for a
- // long time while holding the lock of MetricsSystemImpl, but when start a
new region server, we
- // also need to update metrics in handleReportForDutyResponse, since we
are all in the same
- // process and uses the same metrics instance, we hit dead lock.
- JmxCacheBuster.stop();
}
@AfterAll
