HDDS-13009. Background snapshot defrag service (Draft v1)

hadoop-hdds/common/src/main/java/org/apache/hadoop/ozone/OzoneConfigKeys.java

@@ -277,6 +277,15 @@ public final class OzoneConfigKeys {
       OZONE_SNAPSHOT_SST_FILTERING_SERVICE_TIMEOUT_DEFAULT = "300s";
       // 300s for default
 
+  public static final String OZONE_SNAPSHOT_DEFRAG_SERVICE_TIMEOUT =
+      "ozone.snapshot.defrag.service.timeout";
+  public static final String
+      OZONE_SNAPSHOT_DEFRAG_SERVICE_TIMEOUT_DEFAULT = "300s";
+  // TODO: Adjust timeout as needed.
+  //  One concern would be that snapdiff can take a long time.
+  //  If snapdiff wait time is included in the timeout it can make it indeterministic.
+  //  -- So don't wait? Trigger and check later?
+
   public static final String OZONE_SNAPSHOT_DELETING_SERVICE_INTERVAL =
       "ozone.snapshot.deleting.service.interval";
   public static final String

hadoop-hdds/common/src/main/java/org/apache/hadoop/ozone/OzoneConsts.java

@@ -520,6 +520,7 @@ public final class OzoneConsts {
   public static final String OM_SNAPSHOT_DIR = "db.snapshots";
   public static final String OM_SNAPSHOT_CHECKPOINT_DIR = OM_SNAPSHOT_DIR
       + OM_KEY_PREFIX + "checkpointState";
+  public static final String OM_SNAPSHOT_CHECKPOINT_DEFRAGGED_DIR = "checkpointStateDefragged";
   public static final String OM_SNAPSHOT_DIFF_DIR = OM_SNAPSHOT_DIR
       + OM_KEY_PREFIX + "diffState";
 

hadoop-hdds/common/src/main/resources/ozone-default.xml

@@ -3758,13 +3758,28 @@
       Snapshot Deleting Service per run.
     </description>
   </property>
+  <property>
+    <name>ozone.snapshot.defrag.limit.per.task</name>
+    <value>1</value>
+    <tag>OZONE, PERFORMANCE, OM</tag>
+    <description>The maximum number of snapshots that would be defragmented in
+      each task run of snapshot defragmentation service.
+    </description>
+  </property>
   <property>
     <name>ozone.snapshot.filtering.service.interval</name>
     <value>1m</value>
     <tag>OZONE, PERFORMANCE, OM</tag>
     <description>Time interval of the SST File filtering service from Snapshot.
     </description>
   </property>
+  <property>
+    <name>ozone.snapshot.defrag.service.interval</name>
+    <value>-1</value>
+    <tag>OZONE, PERFORMANCE, OM</tag>
+    <description>Task interval of snapshot defragmentation service.
+    </description>
+  </property>
   <property>
     <name>ozone.om.snapshot.checkpoint.dir.creation.poll.timeout</name>
     <value>20s</value>

hadoop-hdds/framework/src/main/java/org/apache/hadoop/hdds/utils/db/RDBSstFileWriter.java

@@ -28,15 +28,15 @@
 /**
  * DumpFileWriter using rocksdb sst files.
  */
-class RDBSstFileWriter implements Closeable {
+public class RDBSstFileWriter implements Closeable {
 
   private ManagedSstFileWriter sstFileWriter;
   private File sstFile;
   private AtomicLong keyCounter;
   private ManagedOptions emptyOption = new ManagedOptions();
   private final ManagedEnvOptions emptyEnvOptions = new ManagedEnvOptions();
 
-  RDBSstFileWriter(File externalFile) throws RocksDatabaseException {
+  public RDBSstFileWriter(File externalFile) throws RocksDatabaseException {
     this.sstFileWriter = new ManagedSstFileWriter(emptyEnvOptions, emptyOption);
     this.keyCounter = new AtomicLong(0);
     this.sstFile = externalFile;
@@ -60,6 +60,17 @@ public void put(byte[] key, byte[] value) throws RocksDatabaseException {
     }
   }
 
+  public void delete(byte[] key) throws RocksDatabaseException {
+    try {
+      sstFileWriter.delete(key);
+      keyCounter.incrementAndGet();
+    } catch (RocksDBException e) {
+      closeOnFailure();
+      throw new RocksDatabaseException("Failed to delete key (length=" + key.length
+          + "), sstFile=" + sstFile.getAbsolutePath(), e);
+    }
+  }
+
   @Override
   public void close() throws RocksDatabaseException {
     if (sstFileWriter != null) {

hadoop-hdds/framework/src/main/java/org/apache/hadoop/hdds/utils/db/RocksDatabase.java

@@ -237,7 +237,7 @@ public boolean isClosed() {
    *
    * @see ManagedCheckpoint
    */
-  final class RocksCheckpoint implements Closeable {
+  public final class RocksCheckpoint implements Closeable {
     private final ManagedCheckpoint checkpoint;
 
     private RocksCheckpoint() {
@@ -609,7 +609,7 @@ public List<LiveFileMetaData> getLiveFilesMetaData() throws RocksDatabaseExcepti
     }
   }
 
-  RocksCheckpoint createCheckpoint() {
+  public RocksCheckpoint createCheckpoint() {
     return new RocksCheckpoint();
   }
 
@@ -660,7 +660,7 @@ public Collection<ColumnFamily> getExtraColumnFamilies() {
     return Collections.unmodifiableCollection(columnFamilies.values());
   }
 
-  byte[] get(ColumnFamily family, byte[] key) throws RocksDatabaseException {
+  public byte[] get(ColumnFamily family, byte[] key) throws RocksDatabaseException {
     try (UncheckedAutoCloseable ignored = acquire()) {
       return db.get().get(family.getHandle(), key);
     } catch (RocksDBException e) {

hadoop-ozone/common/src/main/java/org/apache/hadoop/ozone/om/OMConfigKeys.java

@@ -428,11 +428,22 @@ public final class OMConfigKeys {
       "ozone.snapshot.deleting.limit.per.task";
   public static final int SNAPSHOT_DELETING_LIMIT_PER_TASK_DEFAULT = 10;
 
+  // Snapshot defragmentation service configuration
+  public static final String SNAPSHOT_DEFRAG_LIMIT_PER_TASK =
+      "ozone.snapshot.defrag.limit.per.task";
+  public static final int SNAPSHOT_DEFRAG_LIMIT_PER_TASK_DEFAULT = 1;
+
   public static final String OZONE_SNAPSHOT_SST_FILTERING_SERVICE_INTERVAL =
       "ozone.snapshot.filtering.service.interval";
   public static final String
       OZONE_SNAPSHOT_SST_FILTERING_SERVICE_INTERVAL_DEFAULT = "60s";
 
+  public static final String OZONE_SNAPSHOT_DEFRAG_SERVICE_INTERVAL =
+      "ozone.snapshot.defrag.service.interval";
+  public static final String
+      OZONE_SNAPSHOT_DEFRAG_SERVICE_INTERVAL_DEFAULT = "-1";
+  // TODO: Disabled by default. Do not enable by default until upgrade handling is complete.
+
   public static final String
       OZONE_SNAPSHOT_CHECKPOINT_DIR_CREATION_POLL_TIMEOUT =
       "ozone.om.snapshot.checkpoint.dir.creation.poll.timeout";