HBASE-30073 Test fixes for some flappers and a reproducible error (#8057)

Signed-off by: Charles Connell <cconnell@apache.org>

Author: haridsv

hbase-balancer/src/main/java/org/apache/hadoop/hbase/master/balancer/StochasticLoadBalancer.java

@@ -581,7 +581,6 @@ protected List<RegionPlan> balanceTable(TableName tableName,
       rackManager, regionCacheRatioOnOldServerMap);
 
     long startTime = EnvironmentEdgeManager.currentTime();
-    cluster.setStopRequestedAt(startTime + maxRunningTime);
 
     initCosts(cluster);
     balancerConditionals.loadClusterState(cluster);
@@ -632,6 +631,10 @@ protected List<RegionPlan> balanceTable(TableName tableName,
       currentCost / sumMultiplier, functionCost(), computedMaxSteps);
 
     final String initFunctionTotalCosts = totalCostsPerFunc();
+    long searchStartTime = EnvironmentEdgeManager.currentTime();
+    // Budget maxRunningTime for the stochastic walk only; initialization (cluster costs, etc.)
+    // can be substantial on busy hosts and must not consume the search deadline.
+    cluster.setStopRequestedAt(searchStartTime + maxRunningTime);
     // Perform a stochastic walk to see if we can get a good fit.
     long step;
     boolean planImprovedConditionals = false;

hbase-common/src/test/java/org/apache/hadoop/hbase/HBaseJupiterExtension.java

@@ -84,7 +84,7 @@ public class HBaseJupiterExtension implements InvocationInterceptor, BeforeAllCa
 
   private static final Map<String, Duration> TAG_TO_TIMEOUT =
     ImmutableMap.of(SmallTests.TAG, Duration.ofMinutes(3), MediumTests.TAG, Duration.ofMinutes(6),
-      LargeTests.TAG, Duration.ofMinutes(13), IntegrationTests.TAG, Duration.ZERO);
+      LargeTests.TAG, Duration.ofMinutes(20), IntegrationTests.TAG, Duration.ZERO);
 
   private static final String EXECUTOR = "executor";
 

hbase-compression/pom.xml

@@ -45,6 +45,11 @@
       <artifactId>hbase-resource-bundle</artifactId>
       <optional>true</optional>
     </dependency>
+    <dependency>
+      <groupId>org.junit.jupiter</groupId>
+      <artifactId>junit-jupiter-api</artifactId>
+      <scope>test</scope>
+    </dependency>
   </dependencies>
 
   <build>

hbase-server/src/test/java/org/apache/hadoop/hbase/HBaseTestingUtil.java

@@ -434,7 +434,13 @@ private void createSubDirAndSystemProperty(String propertyName, Path parent, Str
 
     String sysValue = System.getProperty(propertyName);
 
-    if (sysValue != null) {
+    // Check if directory sharing should be disabled for this test.
+    // Tests that run with high parallelism and don't need shared directories can set this
+    // to avoid race conditions where one test's tearDown() deletes directories another test
+    // is still using.
+    boolean disableSharing = conf.getBoolean("hbase.test.disable-directory-sharing", false);
+
+    if (sysValue != null && !disableSharing) {
       // There is already a value set. So we do nothing but hope
       // that there will be no conflicts
       LOG.info("System.getProperty(\"" + propertyName + "\") already set to: " + sysValue
@@ -447,7 +453,7 @@ private void createSubDirAndSystemProperty(String propertyName, Path parent, Str
       }
       conf.set(propertyName, sysValue);
     } else {
-      // Ok, it's not set, so we create it as a subdirectory
+      // Ok, it's not set (or sharing is disabled), so we create it as a subdirectory
       createSubDir(propertyName, parent, subDirName);
       System.setProperty(propertyName, conf.get(propertyName));
     }

hbase-server/src/test/java/org/apache/hadoop/hbase/TestZooKeeper.java

@@ -75,7 +75,6 @@ public static void setUpBeforeClass() throws Exception {
     conf.setInt(HConstants.ZK_SESSION_TIMEOUT, 1000);
     conf.setClass(HConstants.HBASE_MASTER_LOADBALANCER_CLASS, MockLoadBalancer.class,
       LoadBalancer.class);
-    TEST_UTIL.startMiniDFSCluster(2);
   }
 
   @AfterAll

hbase-server/src/test/java/org/apache/hadoop/hbase/client/AbstractTestAsyncTableScan.java

@@ -73,8 +73,19 @@
 public abstract class AbstractTestAsyncTableScan {
 
   protected static final OpenTelemetryClassRule OTEL_CLASS_RULE = OpenTelemetryClassRule.create();
-  protected static final MiniClusterRule MINI_CLUSTER_RULE = MiniClusterRule.newBuilder()
-    .setMiniClusterOption(StartTestingClusterOption.builder().numWorkers(3).build()).build();
+
+  private static Configuration createConfiguration() {
+    Configuration conf = new Configuration();
+    // Disable directory sharing to prevent race conditions when tests run in parallel.
+    // Each test instance gets its own isolated directories to avoid one test's tearDown()
+    // deleting directories another parallel test is still using.
+    conf.setBoolean("hbase.test.disable-directory-sharing", true);
+    return conf;
+  }
+
+  protected static final MiniClusterRule MINI_CLUSTER_RULE =
+    MiniClusterRule.newBuilder().setConfiguration(createConfiguration())
+      .setMiniClusterOption(StartTestingClusterOption.builder().numWorkers(3).build()).build();
 
   protected static final ConnectionRule CONN_RULE =
     ConnectionRule.createAsyncConnectionRule(MINI_CLUSTER_RULE::createAsyncConnection);

hbase-server/src/test/java/org/apache/hadoop/hbase/client/TestConnection.java

@@ -87,6 +87,10 @@ public class TestConnection {
   @BeforeClass
   public static void setUpBeforeClass() throws Exception {
     ResourceLeakDetector.setLevel(Level.PARANOID);
+    // Disable directory sharing to prevent race conditions when tests run in parallel.
+    // Each test instance gets its own isolated directories to avoid one test's tearDown()
+    // deleting directories another parallel test is still using.
+    TEST_UTIL.getConfiguration().setBoolean("hbase.test.disable-directory-sharing", true);
     TEST_UTIL.getConfiguration().setBoolean(HConstants.STATUS_PUBLISHED, true);
     // Up the handlers; this test needs more than usual.
     TEST_UTIL.getConfiguration().setInt(HConstants.REGION_SERVER_HIGH_PRIORITY_HANDLER_COUNT, 10);

hbase-server/src/test/java/org/apache/hadoop/hbase/io/hfile/bucket/TestPrefetchWithBucketCache.java

@@ -22,7 +22,6 @@
 import static org.apache.hadoop.hbase.io.hfile.BlockCacheFactory.BUCKET_CACHE_BUCKETS_KEY;
 import static org.apache.hadoop.hbase.io.hfile.bucket.BucketCache.QUEUE_ADDITION_WAIT_TIME;
 import static org.junit.Assert.assertEquals;
-import static org.junit.Assert.assertNotEquals;
 import static org.junit.Assert.assertNotNull;
 import static org.junit.Assert.assertNull;
 import static org.junit.Assert.assertTrue;
@@ -328,7 +327,8 @@ public void testPrefetchRunTriggersEvictions() throws Exception {
     conf.setLong(QUEUE_ADDITION_WAIT_TIME, 0);
     blockCache = BlockCacheFactory.createBlockCache(conf);
     cacheConf = new CacheConfig(conf, blockCache);
-    Path storeFile = writeStoreFile("testPrefetchInterruptOnCapacity", 10000);
+    // Use 15000 KVs to ensure file reliably exceeds 1MB cache capacity even with size variance
+    Path storeFile = writeStoreFile("testPrefetchRunTriggersEvictions", 15000);
     // Prefetches the file blocks
     createReaderAndWaitForPrefetchInterruption(storeFile);
     Waiter.waitFor(conf, (PrefetchExecutor.getPrefetchDelay() + 1000),
@@ -343,14 +343,16 @@ public void testPrefetchRunTriggersEvictions() throws Exception {
       }
       return true;
     });
-    if (bc.getStats().getFailedInserts() == 0) {
-      // With no wait time configuration, prefetch should trigger evictions once it reaches
-      // cache capacity
-      assertNotEquals(0, bc.getStats().getEvictedCount());
-    } else {
-      LOG.info("We had {} cache insert failures, which may cause cache usage "
-        + "to never reach capacity.", bc.getStats().getFailedInserts());
-    }
+    // With no wait time configuration, prefetch will either trigger evictions when reaching
+    // cache capacity, or have failed inserts when the writer queue fills faster than it drains.
+    // Both outcomes are valid - test should only fail if NEITHER happens, which would indicate
+    // a problem with the capacity management logic.
+    long evictions = bc.getStats().getEvictedCount();
+    long failedInserts = bc.getStats().getFailedInserts();
+    assertTrue(
+      "Expected either evictions or failed inserts to demonstrate capacity management, "
+        + "but got evictions=" + evictions + ", failedInserts=" + failedInserts,
+      evictions > 0 || failedInserts > 0);
   }
 
   @Test

hbase-server/src/test/java/org/apache/hadoop/hbase/master/assignment/TestRollbackSCP.java

@@ -133,6 +133,11 @@ public static void tearDownAfterClass() throws IOException {
   @BeforeEach
   public void setUp() throws IOException {
     UTIL.ensureSomeNonStoppedRegionServersAvailable(2);
+    // Surefire reruns failed tests in the same JVM without re-running @BeforeClass. Reset injection
+    // state so compareAndSet in persistToMeta can succeed again and kill-before-store flags clear.
+    INJECTED.set(false);
+    ProcedureTestingUtility.setKillAndToggleBeforeStoreUpdateInRollback(
+      UTIL.getMiniHBaseCluster().getMaster().getMasterProcedureExecutor(), false);
   }
 
   private ServerCrashProcedure getSCPForServer(ServerName serverName) throws IOException {

hbase-server/src/test/java/org/apache/hadoop/hbase/quotas/TestBlockBytesScannedQuota.java

@@ -260,7 +260,7 @@ public void testBBSMultiGet() throws Exception {
 
   private void testTraffic(Callable<Long> trafficCallable, long expectedSuccess, long marginOfError)
     throws Exception {
-    TEST_UTIL.waitFor(5_000, () -> {
+    TEST_UTIL.waitFor(30_000, () -> {
       long actualSuccess;
       try {
         actualSuccess = trafficCallable.call();