HBASE-27054 TestStochasticLoadBalancerRegionReplicaLargeCluster.testR…

…egionReplicasOnLargeCluster is flaky (#4454)

Signed-off-by: Andrew Purtell <[email protected]>

Conflicts:
	hbase-server/src/test/java/org/apache/hadoop/hbase/master/balancer/TestStochasticLoadBalancerRegionReplicaLargeCluster.java

hbase-server/src/test/java/org/apache/hadoop/hbase/master/balancer/TestStochasticLoadBalancerRegionReplicaLargeCluster.java

@@ -25,14 +25,20 @@
 import org.junit.experimental.categories.Category;
 
 @Category({ MasterTests.class, LargeTests.class })
-public class TestStochasticLoadBalancerRegionReplicaLargeCluster extends BalancerTestBase2 {
+public class TestStochasticLoadBalancerRegionReplicaLargeCluster extends BalancerTestBase {
 
   @ClassRule
   public static final HBaseClassTestRule CLASS_RULE =
     HBaseClassTestRule.forClass(TestStochasticLoadBalancerRegionReplicaLargeCluster.class);
 
   @Test
   public void testRegionReplicasOnLargeCluster() {
+    // With default values for moveCost and tableSkewCost, the balancer makes much slower progress.
+    // Since we're only looking for balance in region counts and no colocated replicas, we can
+    // ignore these two cost functions to allow us to make any move that helps other functions.
+    conf.setFloat("hbase.master.balancer.stochastic.moveCost", 0f);
+    conf.setFloat("hbase.master.balancer.stochastic.tableSkewCost", 0f);
+    loadBalancer.onConfigurationChange(conf);
     int numNodes = 1000;
     int numRegions = 20 * numNodes; // 20 * replication regions per RS
     int numRegionsPerServer = 19; // all servers except one