HDDS-10374. Make container scanner generate merkle trees during the scan

hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/checksum/ContainerChecksumTreeManager.java

@@ -83,15 +83,10 @@ public void stop() {
    * The data merkle tree within the file is replaced with the {@code tree} parameter, but all other content of the
    * file remains unchanged.
    * Concurrent writes to the same file are coordinated internally.
-   * This method also updates the container's data checksum in the {@code data} parameter, which will be seen by SCM
-   * on container reports.
    */
   public ContainerProtos.ContainerChecksumInfo writeContainerDataTree(ContainerData data,
       ContainerMerkleTreeWriter tree) throws IOException {
     long containerID = data.getContainerID();
-    // If there is an error generating the tree and we cannot obtain a final checksum, use 0 to indicate a metadata
-    // failure.
-    long dataChecksum = 0;
     ContainerProtos.ContainerChecksumInfo checksumInfo = null;
     Lock writeLock = getLock(containerID);
     writeLock.lock();
@@ -113,13 +108,9 @@ public ContainerProtos.ContainerChecksumInfo writeContainerDataTree(ContainerDat
           .setContainerMerkleTree(treeProto);
       checksumInfo = checksumInfoBuilder.build();
       write(data, checksumInfo);
-      // If write succeeds, update the checksum in memory. Otherwise 0 will be used to indicate the metadata failure.
-      dataChecksum = treeProto.getDataChecksum();
-      LOG.debug("Merkle tree for container {} updated with container data checksum {}", containerID,
-          checksumToString(dataChecksum));
+      LOG.debug("Data merkle tree for container {} updated with container checksum {}", containerID,
+          checksumToString(treeProto.getDataChecksum()));
     } finally {
-      // Even if persisting the tree fails, we should still update the data checksum in memory to report back to SCM.
-      data.setDataChecksum(dataChecksum);
       writeLock.unlock();
     }
     return checksumInfo;

hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/common/helpers/ContainerUtils.java

@@ -34,9 +34,6 @@
 import java.nio.file.Paths;
 import java.security.MessageDigest;
 import java.security.NoSuchAlgorithmException;
-import java.time.Duration;
-import java.time.Instant;
-import java.util.Optional;
 import java.util.UUID;
 import java.util.regex.Matcher;
 import java.util.regex.Pattern;
@@ -54,7 +51,6 @@
 import org.apache.hadoop.ozone.container.common.impl.ContainerData;
 import org.apache.hadoop.ozone.container.common.impl.ContainerDataYaml;
 import org.apache.hadoop.ozone.container.common.impl.ContainerSet;
-import org.apache.hadoop.ozone.container.common.interfaces.Container;
 import org.apache.hadoop.ozone.container.keyvalue.KeyValueContainerData;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
@@ -237,29 +233,6 @@ public static String getContainerFileChecksum(String containerDataYamlStr)
     }
   }
 
-  public static boolean recentlyScanned(Container<?> container,
-      long minScanGap, Logger log) {
-    Optional<Instant> lastScanTime =
-        container.getContainerData().lastDataScanTime();
-    Instant now = Instant.now();
-    // Container is considered recently scanned if it was scanned within the
-    // configured time frame. If the optional is empty, the container was
-    // never scanned.
-    boolean recentlyScanned = lastScanTime.map(scanInstant ->
-        Duration.between(now, scanInstant).abs()
-            .compareTo(Duration.ofMillis(minScanGap)) < 0)
-        .orElse(false);
-
-    if (recentlyScanned && log.isDebugEnabled()) {
-      log.debug("Skipping scan for container {} which was last " +
-              "scanned at {}. Current time is {}.",
-          container.getContainerData().getContainerID(), lastScanTime.get(),
-          now);
-    }
-
-    return recentlyScanned;
-  }
-
   /**
    * Get the .container file from the containerBaseDir.
    * @param containerBaseDir container base directory. The name of this

hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/common/interfaces/Handler.java

@@ -31,6 +31,7 @@
 import org.apache.hadoop.hdds.protocol.datanode.proto.ContainerProtos.ContainerType;
 import org.apache.hadoop.hdds.scm.container.common.helpers.StorageContainerException;
 import org.apache.hadoop.ozone.container.checksum.ContainerChecksumTreeManager;
+import org.apache.hadoop.ozone.container.checksum.ContainerMerkleTreeWriter;
 import org.apache.hadoop.ozone.container.checksum.DNContainerOperationClient;
 import org.apache.hadoop.ozone.container.common.helpers.BlockData;
 import org.apache.hadoop.ozone.container.common.helpers.ContainerMetrics;
@@ -154,6 +155,17 @@ public abstract void exportContainer(
   public abstract void markContainerForClose(Container container)
       throws IOException;
 
+  /**
+   * Updates the container checksum information on disk and in memory.
+   *
+   * @param container The container to update
+   * @param treeWriter The container merkle tree with the updated information about the container
+   * @throws IOException For errors sending an ICR or updating the container checksum on disk. If the disk update
+   * fails, the checksum in memory will not be updated.
+   */
+  public abstract void updateContainerChecksum(Container container, ContainerMerkleTreeWriter treeWriter)
+      throws IOException;
+
   /**
    * Marks the container Unhealthy. Moves the container to UNHEALTHY state.
    *

hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/common/utils/ContainerLogger.java

@@ -148,6 +148,18 @@ public static void logRecovered(ContainerData containerData) {
     LOG.info(getMessage(containerData));
   }
 
+  /**
+   * Logged when a container's checksum is updated.
+   *
+   * @param containerData The container which has the updated data checksum.
+   * @param oldDataChecksum The old data checksum.
+   */
+  public static void logChecksumUpdated(ContainerData containerData, long oldDataChecksum) {
+    LOG.warn(getMessage(containerData,
+        "Container data checksum updated from " + checksumToString(oldDataChecksum) + " to "
+            + checksumToString(containerData.getDataChecksum())));
+  }
+
   /**
    * Logged when a container is reconciled.
    *

hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/keyvalue/KeyValueContainerCheck.java

@@ -189,7 +189,6 @@ public DataScanResult fullCheck(DataTransferThrottler throttler, Canceler cancel
 
     LOG.debug("Running data checks for container {}", containerID);
     try {
-      // TODO HDDS-10374 this tree will get updated with the container's contents as it is scanned.
       ContainerMerkleTreeWriter dataTree = new ContainerMerkleTreeWriter();
       List<ContainerScanError> dataErrors = scanData(dataTree, throttler, canceler);
       if (containerIsDeleted()) {
@@ -376,12 +375,15 @@ private List<ContainerScanError> scanBlock(DBHandle db, File dbFile, BlockData b
           // So, we need to make sure, chunk length > 0, before declaring
           // the missing chunk file.
           if (!block.getChunks().isEmpty() && block.getChunks().get(0).getLen() > 0) {
-            ContainerScanError error = new ContainerScanError(FailureType.MISSING_CHUNK_FILE,
+            ContainerScanError error = new ContainerScanError(FailureType.MISSING_DATA_FILE,
                 new File(containerDataFromDisk.getChunksPath()), new IOException("Missing chunk file " +
                 chunkFile.getAbsolutePath()));
             blockErrors.add(error);
           }
         } else if (chunk.getChecksumData().getType() != ContainerProtos.ChecksumType.NONE) {
+          // Before adding chunks, add a block entry to the tree to represent cases where the block exists but has no
+          // chunks.
+          currentTree.addBlock(block.getBlockID().getLocalID());
           int bytesPerChecksum = chunk.getChecksumData().getBytesPerChecksum();
           ByteBuffer buffer = BUFFER_POOL.getBuffer(bytesPerChecksum);
           // Keep scanning the block even if there are errors with individual chunks.
@@ -419,6 +421,14 @@ private static List<ContainerScanError> verifyChecksum(BlockData block,
 
     List<ContainerScanError> scanErrors = new ArrayList<>();
 
+    // Information used to populate the merkle tree. Chunk metadata will be the same, but we must fill in the
+    // checksums with what we actually observe.
+    ContainerProtos.ChunkInfo.Builder observedChunkBuilder = chunk.toBuilder();
+    ContainerProtos.ChecksumData.Builder observedChecksumData = chunk.getChecksumData().toBuilder();
+    observedChecksumData.clearChecksums();
+    boolean chunkHealthy = true;
+    boolean chunkMissing = false;
+
     ChecksumData checksumData =
         ChecksumData.getFromProtoBuf(chunk.getChecksumData());
     int checksumCount = checksumData.getChecksums().size();
@@ -431,10 +441,7 @@ private static List<ContainerScanError> verifyChecksum(BlockData block,
       if (layout == ContainerLayoutVersion.FILE_PER_BLOCK) {
         channel.position(chunk.getOffset());
       }
-      // Only report one error per chunk. Reporting corruption at every "bytes per checksum" interval will lead to a
-      // large amount of errors when a full chunk is corrupted.
-      boolean chunkHealthy = true;
-      for (int i = 0; i < checksumCount && chunkHealthy; i++) {
+      for (int i = 0; i < checksumCount; i++) {
         // limit last read for FILE_PER_BLOCK, to avoid reading next chunk
         if (layout == ContainerLayoutVersion.FILE_PER_BLOCK &&
             i == checksumCount - 1 &&
@@ -454,7 +461,11 @@ private static List<ContainerScanError> verifyChecksum(BlockData block,
         ByteString expected = checksumData.getChecksums().get(i);
         ByteString actual = cal.computeChecksum(buffer)
             .getChecksums().get(0);
-        if (!expected.equals(actual)) {
+        observedChecksumData.addChecksums(actual);
+        // Only report one error per chunk. Reporting corruption at every "bytes per checksum" interval will lead to a
+        // large amount of errors when a full chunk is corrupted.
+        // Continue scanning the chunk even after the first error so the full merkle tree can be built.
+        if (chunkHealthy && !expected.equals(actual)) {
           String message = String
               .format("Inconsistent read for chunk=%s" +
                   " checksum item %d" +
@@ -466,26 +477,46 @@ private static List<ContainerScanError> verifyChecksum(BlockData block,
                   StringUtils.bytes2Hex(expected.asReadOnlyByteBuffer()),
                   StringUtils.bytes2Hex(actual.asReadOnlyByteBuffer()),
                   block.getBlockID());
+          chunkHealthy = false;
           scanErrors.add(new ContainerScanError(FailureType.CORRUPT_CHUNK, chunkFile,
               new OzoneChecksumException(message)));
-          chunkHealthy = false;
         }
       }
-      // If all the checksums match, also check that the length stored in the metadata matches the number of bytes
-      // seen on the disk.
+
+      observedChunkBuilder.setLen(bytesRead);
+      // If we haven't seen any errors after scanning the whole chunk, verify that the length stored in the metadata
+      // matches the number of bytes seen on the disk.
       if (chunkHealthy && bytesRead != chunk.getLen()) {
-        String message = String
-            .format("Inconsistent read for chunk=%s expected length=%d"
-                    + " actual length=%d for block %s",
-                chunk.getChunkName(),
-                chunk.getLen(), bytesRead, block.getBlockID());
-        scanErrors.add(new ContainerScanError(FailureType.INCONSISTENT_CHUNK_LENGTH, chunkFile,
-            new IOException(message)));
+        if (bytesRead == 0) {
+          // If we could not find any data for the chunk, report it as missing.
+          chunkMissing = true;
+          chunkHealthy = false;
+          String message = String.format("Missing chunk=%s with expected length=%d for block %s",
+                  chunk.getChunkName(), chunk.getLen(), block.getBlockID());
+          scanErrors.add(new ContainerScanError(FailureType.MISSING_CHUNK, chunkFile, new IOException(message)));
+        } else {
+          // We found data for the chunk, but it was shorter than expected.
+          String message = String
+              .format("Inconsistent read for chunk=%s expected length=%d"
+                      + " actual length=%d for block %s",
+                  chunk.getChunkName(),
+                  chunk.getLen(), bytesRead, block.getBlockID());
+          chunkHealthy = false;
+          scanErrors.add(new ContainerScanError(FailureType.INCONSISTENT_CHUNK_LENGTH, chunkFile,
+              new IOException(message)));
+        }
       }
     } catch (IOException ex) {
-      scanErrors.add(new ContainerScanError(FailureType.MISSING_CHUNK_FILE, chunkFile, ex));
+      // An unknown error occurred trying to access the chunk. Report it as corrupted.
+      chunkHealthy = false;
+      scanErrors.add(new ContainerScanError(FailureType.CORRUPT_CHUNK, chunkFile, ex));
     }
 
+    // Missing chunks should not be added to the merkle tree.
+    if (!chunkMissing) {
+      observedChunkBuilder.setChecksumData(observedChecksumData);
+      currentTree.addChunks(block.getBlockID().getLocalID(), chunkHealthy, observedChunkBuilder.build());
+    }
     return scanErrors;
   }