HDDS-14768. Fix lock leak during snapshot cache cleanup and handle eviction race appropriately.

hadoop-ozone/ozone-manager/src/main/java/org/apache/hadoop/ozone/om/snapshot/SnapshotCache.java

@@ -25,6 +25,7 @@
 import com.google.common.annotations.VisibleForTesting;
 import com.google.common.cache.CacheLoader;
 import java.io.IOException;
+import java.util.Collections;
 import java.util.Set;
 import java.util.UUID;
 import java.util.concurrent.ConcurrentHashMap;
@@ -135,6 +136,11 @@ ConcurrentHashMap<UUID, ReferenceCounted<OmSnapshot>> getDbMap() {
     return dbMap;
   }
 
+  @VisibleForTesting
+  Set<UUID> getPendingEvictionQueue() {
+    return Collections.unmodifiableSet(pendingEvictionQueue);
+  }
+
   /**
    * @return number of DB instances currently held in cache.
    */
@@ -158,6 +164,7 @@ public void invalidate(UUID key) {
         }
         omMetrics.decNumSnapshotCacheSize();
       }
+      pendingEvictionQueue.remove(k);
       return null;
     });
   }
@@ -323,8 +330,17 @@ private UncheckedAutoCloseableSupplier<OMLockDetails> lock(Supplier<OMLockDetail
 
     AtomicReference<OMLockDetails> lockDetails = new AtomicReference<>(emptyLockFunction.get());
     if (lockDetails.get().isLockAcquired()) {
-      if (!cleanupFunction.get()) {
-        lockDetails.set(emptyUnlockFunction.get());
+      try {
+        if (!cleanupFunction.get()) {
+          throw new IllegalStateException("Failed to acquire lock as cleanup did not drain the cache.");
+        }
+      } catch (Throwable t) {
+        try {
+          lockDetails.set(emptyUnlockFunction.get());
+        } catch (Throwable unlockThrowable) {
+          t.addSuppressed(unlockThrowable);
+        }
+        throw t;
       }
     }
 
@@ -377,26 +393,25 @@ private synchronized Void cleanup(UUID evictionKey, boolean expectKeyToBePresent
     }
 
     dbMap.compute(evictionKey, (k, v) -> {
-      pendingEvictionQueue.remove(k);
+      ReferenceCounted<OmSnapshot> result = null;
       if (v == null) {
-        throw new IllegalStateException("SnapshotId '" + k + "' does not exist in cache. The RocksDB " +
-            "instance of the Snapshot may not be closed properly.");
-      }
-
-      if (v.getTotalRefCount() > 0) {
+        LOG.info("SnapshotId '{}' does not exist in cache during cleanup; "
+            + "it may have already been invalidated, closed, and removed.", k);
+      } else if (v.getTotalRefCount() > 0) {
         LOG.debug("SnapshotId {} is still being referenced ({}), skipping its clean up.", k, v.getTotalRefCount());
-        return v;
+        result = v;
       } else {
         LOG.debug("Closing SnapshotId {}. It is not being referenced anymore.", k);
         // Close the instance, which also closes its DB handle.
         try {
           v.get().close();
         } catch (IOException ex) {
-          throw new IllegalStateException("Error while closing snapshot DB.", ex);
+          throw new IllegalStateException("Error while closing snapshot DB for snapshotId " + k, ex);
         }
         omMetrics.decNumSnapshotCacheSize();
-        return null;
       }
+      pendingEvictionQueue.remove(k);
+      return result;
     });
     return null;
   }

hadoop-ozone/ozone-manager/src/test/java/org/apache/hadoop/ozone/om/snapshot/TestSnapshotCache.java

@@ -19,6 +19,7 @@
 
 import static org.apache.hadoop.ozone.om.lock.DAGLeveledResource.SNAPSHOT_DB_LOCK;
 import static org.apache.hadoop.ozone.om.lock.OzoneManagerLock.LeveledResource.VOLUME_LOCK;
+import static org.junit.jupiter.api.Assertions.assertDoesNotThrow;
 import static org.junit.jupiter.api.Assertions.assertEquals;
 import static org.junit.jupiter.api.Assertions.assertFalse;
 import static org.junit.jupiter.api.Assertions.assertInstanceOf;
@@ -46,6 +47,7 @@
 import java.util.concurrent.TimeoutException;
 import java.util.concurrent.atomic.AtomicBoolean;
 import org.apache.hadoop.hdds.conf.OzoneConfiguration;
+import org.apache.hadoop.hdds.utils.db.DBStore;
 import org.apache.hadoop.hdds.utils.db.Table;
 import org.apache.hadoop.ozone.om.OMMetadataManager;
 import org.apache.hadoop.ozone.om.OMMetrics;
@@ -511,4 +513,137 @@ void testSnapshotOperationsNotBlockedDuringCompaction() throws IOException, Inte
     verify(store1, times(1)).compactTable("table2");
     verify(store1, times(0)).compactTable("keyTable");
   }
+
+  private static IOzoneManagerLock newAcquiringLock() {
+    IOzoneManagerLock acquiringLock = mock(IOzoneManagerLock.class);
+    when(acquiringLock.acquireReadLock(eq(SNAPSHOT_DB_LOCK), any(String[].class)))
+        .thenReturn(OMLockDetails.EMPTY_DETAILS_LOCK_ACQUIRED);
+    when(acquiringLock.releaseReadLock(eq(SNAPSHOT_DB_LOCK), any(String[].class)))
+        .thenReturn(OMLockDetails.EMPTY_DETAILS_LOCK_NOT_ACQUIRED);
+    when(acquiringLock.acquireResourceWriteLock(eq(SNAPSHOT_DB_LOCK)))
+        .thenReturn(OMLockDetails.EMPTY_DETAILS_LOCK_ACQUIRED);
+    when(acquiringLock.releaseResourceWriteLock(eq(SNAPSHOT_DB_LOCK)))
+        .thenReturn(OMLockDetails.EMPTY_DETAILS_LOCK_NOT_ACQUIRED);
+    when(acquiringLock.acquireWriteLock(eq(SNAPSHOT_DB_LOCK), any(String[].class)))
+        .thenReturn(OMLockDetails.EMPTY_DETAILS_LOCK_ACQUIRED);
+    when(acquiringLock.releaseWriteLock(eq(SNAPSHOT_DB_LOCK), any(String[].class)))
+        .thenReturn(OMLockDetails.EMPTY_DETAILS_LOCK_NOT_ACQUIRED);
+    return acquiringLock;
+  }
+
+  private OmSnapshot mockSnapshot(UUID snapshotId) {
+    final OmSnapshot omSnapshot = mock(OmSnapshot.class);
+    when(omSnapshot.getSnapshotTableKey()).thenReturn(snapshotId.toString());
+    when(omSnapshot.getSnapshotID()).thenReturn(snapshotId);
+
+    return omSnapshot;
+  }
+
+  @Test
+  @DisplayName("Stale eviction key (invalidate + late close) is cleaned up without throwing")
+  void testStaleEvictionKeyDuringCleanup() throws IOException {
+    snapshotCache = new SnapshotCache(cacheLoader, CACHE_SIZE_LIMIT, omMetrics, 0, true, newAcquiringLock());
+    final UUID snapshotId = UUID.randomUUID();
+
+    // Acquire a snapshot handle so it is ref-counted in the cache.
+    UncheckedAutoCloseableSupplier<OmSnapshot> handle = snapshotCache.get(snapshotId);
+    assertEquals(1, snapshotCache.size());
+
+    // Invalidate removes the dbMap entry. The handle still exists and will later hit refcount=0.
+    snapshotCache.invalidate(snapshotId);
+    assertEquals(0, snapshotCache.size());
+
+    // Late close triggers ReferenceCounted callback which can re-add snapshotId to pendingEvictionQueue.
+    handle.close();
+    assertTrue(snapshotCache.getPendingEvictionQueue().contains(snapshotId));
+
+    // cleanup(true) is invoked by lock(); it should remove the stale key and not throw.
+    assertDoesNotThrow(() -> {
+      try (UncheckedAutoCloseableSupplier<OMLockDetails> lockDetails = snapshotCache.lock()) {
+        assertTrue(lockDetails.get().isLockAcquired());
+      }
+    });
+    assertFalse(snapshotCache.getPendingEvictionQueue().contains(snapshotId));
+  }
+
+  @Test
+  @DisplayName("Close failure keeps snapshot in eviction queue for retry")
+  void testCloseFailureRetriesSnapshot() throws Exception {
+
+    IOzoneManagerLock acquiringLock = newAcquiringLock();
+    snapshotCache = new SnapshotCache(cacheLoader, CACHE_SIZE_LIMIT, omMetrics, 0, true, acquiringLock);
+    final UUID snapshotId = UUID.randomUUID();
+
+    final AtomicBoolean failCloseOnce = new AtomicBoolean(true);
+    final OmSnapshot failingSnapshot = mockSnapshot(snapshotId);
+
+    OMMetadataManager metadataManager = mock(OMMetadataManager.class);
+    DBStore store = mock(DBStore.class);
+    when(failingSnapshot.getMetadataManager()).thenReturn(metadataManager);
+    when(metadataManager.getStore()).thenReturn(store);
+    when(store.listTables()).thenReturn(new ArrayList<>());
+
+    doAnswer(invocation -> {
+      if (failCloseOnce.getAndSet(false)) {
+        throw new IOException("close failed");
+      }
+      return null;
+    }).when(failingSnapshot).close();
+
+    when(cacheLoader.load(eq(snapshotId))).thenReturn(failingSnapshot);
+
+    // Load + close handle so refcount transitions to 0 and snapshotId is queued for eviction.
+    try (UncheckedAutoCloseableSupplier<OmSnapshot> ignored = snapshotCache.get(snapshotId)) {
+      assertEquals(1, snapshotCache.size());
+      assertEquals(1, omMetrics.getNumSnapshotCacheSize());
+    }
+    assertEquals(0L, snapshotCache.getDbMap().get(snapshotId).getTotalRefCount());
+    assertTrue(snapshotCache.getPendingEvictionQueue().contains(snapshotId));
+
+    // First cleanup attempt fails to close; entry should remain in dbMap and key should stay queued for retry.
+    assertThrows(IllegalStateException.class, () -> snapshotCache.lock());
+    verify(acquiringLock, times(1)).acquireResourceWriteLock(eq(SNAPSHOT_DB_LOCK));
+    verify(acquiringLock, times(1)).releaseResourceWriteLock(eq(SNAPSHOT_DB_LOCK));
+    assertTrue(snapshotCache.getDbMap().containsKey(snapshotId));
+    assertTrue(snapshotCache.getPendingEvictionQueue().contains(snapshotId));
+    assertEquals(1, omMetrics.getNumSnapshotCacheSize());
+
+    // Second cleanup attempt should succeed (close no longer throws), removing entry and eviction key.
+    try (UncheckedAutoCloseableSupplier<OMLockDetails> lockDetails = snapshotCache.lock()) {
+      assertTrue(lockDetails.get().isLockAcquired());
+    }
+    assertFalse(snapshotCache.getDbMap().containsKey(snapshotId));
+    assertFalse(snapshotCache.getPendingEvictionQueue().contains(snapshotId));
+    assertEquals(0, omMetrics.getNumSnapshotCacheSize());
+  }
+
+  @Test
+  @DisplayName("lock supplier releases write lock if cleanup throws an exception")
+  void testLockSupplierReleasesWriteLockOnCleanupException() throws Exception {
+    IOzoneManagerLock acquiringLock = newAcquiringLock();
+    snapshotCache = new SnapshotCache(cacheLoader, CACHE_SIZE_LIMIT, omMetrics, 0, true, acquiringLock);
+
+    final UUID snapshotId = UUID.randomUUID();
+    final OmSnapshot failingSnapshot = mockSnapshot(snapshotId);
+
+    OMMetadataManager metadataManager = mock(OMMetadataManager.class);
+    DBStore store = mock(DBStore.class);
+    when(failingSnapshot.getMetadataManager()).thenReturn(metadataManager);
+    when(metadataManager.getStore()).thenReturn(store);
+    // Trigger an unchecked exception during compaction, which is not caught by cleanup().
+    when(store.listTables()).thenThrow(new RuntimeException("listTables failed"));
+
+    when(cacheLoader.load(eq(snapshotId))).thenReturn(failingSnapshot);
+
+    // Load the snapshot and close so it is enqueued for eviction (refcount reaches 0).
+    try (UncheckedAutoCloseableSupplier<OmSnapshot> ignored = snapshotCache.get(snapshotId)) {
+      assertEquals(1, snapshotCache.size());
+    }
+    assertTrue(snapshotCache.getPendingEvictionQueue().contains(snapshotId));
+
+    // cleanup(true) will throw -> lock() should release the resource write lock before rethrowing.
+    assertThrows(RuntimeException.class, () -> snapshotCache.lock());
+    verify(acquiringLock, times(1)).acquireResourceWriteLock(eq(SNAPSHOT_DB_LOCK));
+    verify(acquiringLock, times(1)).releaseResourceWriteLock(eq(SNAPSHOT_DB_LOCK));
+  }
 }