HBASE-26320 Implement a separate thread pool for the LogCleaner

This avoids starvation when the archive directory is large and takes a long time
to iterate through.

Author: z-york

hbase-server/src/main/java/org/apache/hadoop/hbase/master/HMaster.java

@@ -352,7 +352,8 @@ public class HMaster extends HRegionServer implements MasterServices {
 
   private HbckChore hbckChore;
   CatalogJanitor catalogJanitorChore;
-  private DirScanPool cleanerPool;
+  private DirScanPool archiveCleanerPool;
+  private DirScanPool logCleanerPool;
   private LogCleaner logCleaner;
   private HFileCleaner hfileCleaner;
   private ReplicationBarrierCleaner replicationBarrierCleaner;
@@ -1068,7 +1069,8 @@ private void finishActiveMasterInitialization(MonitoredTask status) throws IOExc
        (EnvironmentEdgeManager.currentTime() - masterActiveTime) / 1000.0f));
     this.masterFinishedInitializationTime = EnvironmentEdgeManager.currentTime();
     configurationManager.registerObserver(this.balancer);
-    configurationManager.registerObserver(this.cleanerPool);
+    configurationManager.registerObserver(this.archiveCleanerPool);
+    configurationManager.registerObserver(this.logCleanerPool);
     configurationManager.registerObserver(this.hfileCleaner);
     configurationManager.registerObserver(this.logCleaner);
     configurationManager.registerObserver(this.regionsRecoveryConfigManager);
@@ -1430,21 +1432,23 @@ private void startServiceThreads() throws IOException {
         ExecutorType.MASTER_TABLE_OPERATIONS).setCorePoolSize(1));
     startProcedureExecutor();
 
-    // Create cleaner thread pool
-    cleanerPool = new DirScanPool(conf);
+    // Create log cleaner thread pool
+    logCleanerPool = DirScanPool.getOldLogsScanner(conf);
     Map<String, Object> params = new HashMap<>();
     params.put(MASTER, this);
     // Start log cleaner thread
     int cleanerInterval =
       conf.getInt(HBASE_MASTER_CLEANER_INTERVAL, DEFAULT_HBASE_MASTER_CLEANER_INTERVAL);
     this.logCleaner = new LogCleaner(cleanerInterval, this, conf,
-      getMasterWalManager().getFileSystem(), getMasterWalManager().getOldLogDir(), cleanerPool, params);
+      getMasterWalManager().getFileSystem(), getMasterWalManager().getOldLogDir(), logCleanerPool, params);
     getChoreService().scheduleChore(logCleaner);
 
     // start the hfile archive cleaner thread
     Path archiveDir = HFileArchiveUtil.getArchivePath(conf);
+    // Create archive cleaner thread pool
+    archiveCleanerPool = DirScanPool.getArchiveScanner(conf);
     this.hfileCleaner = new HFileCleaner(cleanerInterval, this, conf,
-      getMasterFileSystem().getFileSystem(), archiveDir, cleanerPool, params);
+      getMasterFileSystem().getFileSystem(), archiveDir, archiveCleanerPool, params);
     getChoreService().scheduleChore(hfileCleaner);
 
     // Regions Reopen based on very high storeFileRefCount is considered enabled
@@ -1496,9 +1500,13 @@ protected void stopServiceThreads() {
     stopChores();
 
     super.stopServiceThreads();
-    if (cleanerPool != null) {
-      cleanerPool.shutdownNow();
-      cleanerPool = null;
+    if (archiveCleanerPool != null) {
+      archiveCleanerPool.shutdownNow();
+      archiveCleanerPool = null;
+    }
+    if (logCleanerPool != null) {
+      logCleanerPool.shutdownNow();
+      logCleanerPool = null;
     }
 
     LOG.debug("Stopping service threads");

hbase-server/src/main/java/org/apache/hadoop/hbase/master/cleaner/CleanerChore.java

@@ -64,6 +64,8 @@ public abstract class CleanerChore<T extends FileCleanerDelegate> extends Schedu
    */
   public static final String CHORE_POOL_SIZE = "hbase.cleaner.scan.dir.concurrent.size";
   static final String DEFAULT_CHORE_POOL_SIZE = "0.25";
+  public static final String LOG_CLEANER_CHORE_SIZE = "hbase.log.cleaner.scan.dir.concurrent.size";
+  static final String DEFAULT_LOG_CLEANER_CHORE_POOL_SIZE = "1";
 
   private final DirScanPool pool;
 

hbase-server/src/main/java/org/apache/hadoop/hbase/master/cleaner/DirScanPool.java

@@ -39,21 +39,28 @@ public class DirScanPool implements ConfigurationObserver {
   private final ThreadPoolExecutor pool;
   private int cleanerLatch;
   private boolean reconfigNotification;
+  private final String cleanerPoolConfig;
+  private final String cleanerPoolConfigDefault;
+  private final String name;
 
-  public DirScanPool(Configuration conf) {
-    String poolSize = conf.get(CleanerChore.CHORE_POOL_SIZE, CleanerChore.DEFAULT_CHORE_POOL_SIZE);
+  private DirScanPool(Configuration conf, String cleanerPoolConfig, String cleanerPoolConfigDefault,
+      String name) {
+    this.cleanerPoolConfig = cleanerPoolConfig;
+    this.cleanerPoolConfigDefault = cleanerPoolConfigDefault;
+    this.name = name;
+    String poolSize = conf.get(cleanerPoolConfig, cleanerPoolConfigDefault);
     size = CleanerChore.calculatePoolSize(poolSize);
     // poolSize may be 0 or 0.0 from a careless configuration,
     // double check to make sure.
-    size = size == 0 ? CleanerChore.calculatePoolSize(CleanerChore.DEFAULT_CHORE_POOL_SIZE) : size;
-    pool = initializePool(size);
-    LOG.info("Cleaner pool size is {}", size);
+    size = size == 0 ? CleanerChore.calculatePoolSize(cleanerPoolConfigDefault) : size;
+    pool = initializePool(size, name);
+    LOG.info("{} Cleaner pool size is {}", name, size);
     cleanerLatch = 0;
   }
 
-  private static ThreadPoolExecutor initializePool(int size) {
+  private static ThreadPoolExecutor initializePool(int size, String name) {
     return Threads.getBoundedCachedThreadPool(size, 1, TimeUnit.MINUTES,
-      new ThreadFactoryBuilder().setNameFormat("dir-scan-pool-%d").setDaemon(true)
+      new ThreadFactoryBuilder().setNameFormat(name + "dir-scan-pool-%d").setDaemon(true)
         .setUncaughtExceptionHandler(Threads.LOGGING_EXCEPTION_HANDLER).build());
   }
 
@@ -64,9 +71,10 @@ private static ThreadPoolExecutor initializePool(int size) {
   @Override
   public synchronized void onConfigurationChange(Configuration conf) {
     int newSize = CleanerChore.calculatePoolSize(
-      conf.get(CleanerChore.CHORE_POOL_SIZE, CleanerChore.DEFAULT_CHORE_POOL_SIZE));
+      conf.get(cleanerPoolConfig, cleanerPoolConfigDefault));
     if (newSize == size) {
-      LOG.trace("Size from configuration is same as previous={}, no need to update.", newSize);
+      LOG.trace("{} Cleaner Size from configuration is same as previous={}, no need to update.",
+          name, newSize);
       return;
     }
     size = newSize;
@@ -109,11 +117,21 @@ synchronized void tryUpdatePoolSize(long timeout) {
         break;
       }
     }
-    LOG.info("Update chore's pool size from {} to {}", pool.getPoolSize(), size);
+    LOG.info("Update {} chore's pool size from {} to {}", name, pool.getPoolSize(), size);
     pool.setCorePoolSize(size);
   }
 
   public int getSize() {
     return size;
   }
+
+  public static DirScanPool getArchiveScanner(Configuration conf) {
+    return new DirScanPool(conf, CleanerChore.CHORE_POOL_SIZE,
+        CleanerChore.DEFAULT_CHORE_POOL_SIZE, "Archive");
+  }
+
+  public static DirScanPool getOldLogsScanner(Configuration conf) {
+    return new DirScanPool(conf, CleanerChore.LOG_CLEANER_CHORE_SIZE,
+        CleanerChore.DEFAULT_LOG_CLEANER_CHORE_POOL_SIZE, "Log");
+  }
 }

hbase-server/src/test/java/org/apache/hadoop/hbase/backup/TestHFileArchiving.java

@@ -111,7 +111,7 @@ public static void setupCluster() throws Exception {
     // We don't want the cleaner to remove files. The tests do that.
     UTIL.getMiniHBaseCluster().getMaster().getHFileCleaner().cancel(true);
 
-    POOL = new DirScanPool(UTIL.getConfiguration());
+    POOL = DirScanPool.getArchiveScanner(UTIL.getConfiguration());
   }
 
   private static void setupConf(Configuration conf) {

hbase-server/src/test/java/org/apache/hadoop/hbase/backup/example/TestZooKeeperTableArchiveClient.java

@@ -119,7 +119,7 @@ public static void setupCluster() throws Exception {
     String archivingZNode = ZKTableArchiveClient.getArchiveZNode(UTIL.getConfiguration(), watcher);
     ZKUtil.createWithParents(watcher, archivingZNode);
     rss = mock(RegionServerServices.class);
-    POOL = new DirScanPool(UTIL.getConfiguration());
+    POOL= DirScanPool.getArchiveScanner(UTIL.getConfiguration());
   }
 
   private static void setupConf(Configuration conf) {

hbase-server/src/test/java/org/apache/hadoop/hbase/master/cleaner/TestCleanerChore.java

@@ -61,7 +61,7 @@ public class TestCleanerChore {
 
   @BeforeClass
   public static void setup() {
-    POOL = new DirScanPool(UTIL.getConfiguration());
+    POOL = DirScanPool.getArchiveScanner(UTIL.getConfiguration());
   }
 
   @AfterClass
@@ -469,6 +469,57 @@ public void testOnConfigurationChange() throws Exception {
     t.join();
   }
 
+  @Test
+  public void testOnConfigurationChangeLogCleaner() throws Exception {
+    int availableProcessorNum = Runtime.getRuntime().availableProcessors();
+    if (availableProcessorNum == 1) { // no need to run this test
+      return;
+    }
+
+    DirScanPool pool = DirScanPool.getOldLogsScanner(UTIL.getConfiguration());
+
+    // have at least 2 available processors/cores
+    int initPoolSize = availableProcessorNum / 2;
+    int changedPoolSize = availableProcessorNum;
+
+    Stoppable stop = new StoppableImplementation();
+    Configuration conf = UTIL.getConfiguration();
+    Path testDir = UTIL.getDataTestDir();
+    FileSystem fs = UTIL.getTestFileSystem();
+    String confKey = "hbase.test.cleaner.delegates";
+    conf.set(confKey, AlwaysDelete.class.getName());
+    conf.set(CleanerChore.LOG_CLEANER_CHORE_SIZE, String.valueOf(initPoolSize));
+    final AllValidPaths chore =
+        new AllValidPaths("test-file-cleaner", stop, conf, fs, testDir, confKey, pool);
+    chore.setEnabled(true);
+    // Create subdirs under testDir
+    int dirNums = 6;
+    Path[] subdirs = new Path[dirNums];
+    for (int i = 0; i < dirNums; i++) {
+      subdirs[i] = new Path(testDir, "subdir-" + i);
+      fs.mkdirs(subdirs[i]);
+    }
+    // Under each subdirs create 6 files
+    for (Path subdir : subdirs) {
+      createFiles(fs, subdir, 6);
+    }
+    // Start chore
+    Thread t = new Thread(new Runnable() {
+      @Override
+      public void run() {
+        chore.chore();
+      }
+    });
+    t.setDaemon(true);
+    t.start();
+    // Change size of chore's pool
+    conf.set(CleanerChore.LOG_CLEANER_CHORE_SIZE, String.valueOf(changedPoolSize));
+    pool.onConfigurationChange(conf);
+    assertEquals(changedPoolSize, chore.getChorePoolSize());
+    // Stop chore
+    t.join();
+  }
+
   @Test
   public void testMinimumNumberOfThreads() throws Exception {
     Configuration conf = UTIL.getConfiguration();

hbase-server/src/test/java/org/apache/hadoop/hbase/master/cleaner/TestHFileCleaner.java

@@ -71,7 +71,7 @@ public class TestHFileCleaner {
   public static void setupCluster() throws Exception {
     // have to use a minidfs cluster because the localfs doesn't modify file times correctly
     UTIL.startMiniDFSCluster(1);
-    POOL = new DirScanPool(UTIL.getConfiguration());
+    POOL = DirScanPool.getArchiveScanner(UTIL.getConfiguration());
   }
 
   @AfterClass

hbase-server/src/test/java/org/apache/hadoop/hbase/master/cleaner/TestHFileLinkCleaner.java

@@ -67,7 +67,7 @@ public class TestHFileLinkCleaner {
 
   @BeforeClass
   public static void setUp() {
-    POOL = new DirScanPool(TEST_UTIL.getConfiguration());
+    POOL = DirScanPool.getArchiveScanner(TEST_UTIL.getConfiguration());
   }
 
   @AfterClass

hbase-server/src/test/java/org/apache/hadoop/hbase/master/cleaner/TestLogsCleaner.java

@@ -94,7 +94,7 @@ public class TestLogsCleaner {
   public static void setUpBeforeClass() throws Exception {
     TEST_UTIL.startMiniZKCluster();
     TEST_UTIL.startMiniDFSCluster(1);
-    POOL = new DirScanPool(TEST_UTIL.getConfiguration());
+    POOL = DirScanPool.getOldLogsScanner(TEST_UTIL.getConfiguration());
   }
 
   @AfterClass

hbase-server/src/test/java/org/apache/hadoop/hbase/master/region/MasterRegionTestBase.java

@@ -48,7 +48,9 @@ public class MasterRegionTestBase {
 
   protected ChoreService choreService;
 
-  protected DirScanPool cleanerPool;
+  protected DirScanPool archiveCleanerPool;
+
+  protected DirScanPool logCleanerPool;
 
   protected static byte[] CF1 = Bytes.toBytes("f1");
 
@@ -80,7 +82,8 @@ public void setUp() throws IOException {
     htu.getConfiguration().setBoolean(CommonFSUtils.UNSAFE_STREAM_CAPABILITY_ENFORCE, false);
     configure(htu.getConfiguration());
     choreService = new ChoreService(getClass().getSimpleName());
-    cleanerPool = new DirScanPool(htu.getConfiguration());
+    archiveCleanerPool = DirScanPool.getArchiveScanner(htu.getConfiguration());
+    logCleanerPool = DirScanPool.getOldLogsScanner(htu.getConfiguration());
     Server server = mock(Server.class);
     when(server.getConfiguration()).thenReturn(htu.getConfiguration());
     when(server.getServerName())
@@ -103,7 +106,8 @@ public void setUp() throws IOException {
   @After
   public void tearDown() throws IOException {
     region.close(true);
-    cleanerPool.shutdownNow();
+    archiveCleanerPool.shutdownNow();
+    logCleanerPool.shutdownNow();
     choreService.shutdown();
     htu.cleanupTestDir();
   }