HDDS-659. Implement pagination in GET bucket (object list) endpoint. Contributed by Bharat Viswanadham.

hadoop-ozone/s3gateway/src/main/java/org/apache/hadoop/ozone/s3/endpoint/BucketEndpoint.java

@@ -36,7 +36,6 @@
 import java.io.InputStream;
 import java.time.Instant;
 import java.util.Iterator;
-import javax.ws.rs.core.Response.ResponseBuilder;
 
 import org.apache.hadoop.ozone.client.OzoneBucket;
 import org.apache.hadoop.ozone.client.OzoneKey;
@@ -48,10 +47,13 @@
 import org.apache.hadoop.ozone.s3.exception.S3ErrorTable;
 
 import org.apache.commons.lang3.StringUtils;
+import org.apache.hadoop.ozone.s3.util.S3utils;
 import org.apache.http.HttpStatus;
 import org.slf4j.Logger;
 import org.slf4j.LoggerFactory;
 
+import static org.apache.hadoop.ozone.s3.util.S3Consts.ENCODING_TYPE;
+
 /**
  * Bucket level rest endpoints.
  */
@@ -76,6 +78,8 @@ public Response list(
       @DefaultValue("1000") @QueryParam("max-keys") int maxKeys,
       @QueryParam("prefix") String prefix,
       @QueryParam("browser") String browser,
+      @QueryParam("continuation-token") String continueToken,
+      @QueryParam("start-after") String startAfter,
       @Context HttpHeaders hh) throws OS3Exception, IOException {
 
     if (browser != null) {
@@ -87,60 +91,91 @@ public Response list(
       }
     }
 
-    if (delimiter == null) {
-      delimiter = "/";
-    }
     if (prefix == null) {
       prefix = "";
     }
 
     OzoneBucket bucket = getBucket(bucketName);
 
-    Iterator<? extends OzoneKey> ozoneKeyIterator = bucket.listKeys(prefix);
+    Iterator<? extends OzoneKey> ozoneKeyIterator;
+
+    String decodedToken = S3utils.decodeContinueToken(continueToken);
+
+    if (startAfter != null && continueToken != null) {
+      // If continuation token and start after both are provided, then we
+      // ignore start After
+      ozoneKeyIterator = bucket.listKeys(prefix, decodedToken);
+    } else if (startAfter != null && continueToken == null) {
+      ozoneKeyIterator = bucket.listKeys(prefix, startAfter);
+    } else if (startAfter == null && continueToken != null){
+      ozoneKeyIterator = bucket.listKeys(prefix, decodedToken);
+    } else {
+      ozoneKeyIterator = bucket.listKeys(prefix);
+    }
+
 
     ListObjectResponse response = new ListObjectResponse();
     response.setDelimiter(delimiter);
     response.setName(bucketName);
     response.setPrefix(prefix);
     response.setMarker("");
-    response.setMaxKeys(1000);
+    response.setMaxKeys(maxKeys);
-    response.setEncodingType("url");
+    response.setEncodingType(ENCODING_TYPE);
     response.setTruncated(false);
+    response.setContinueToken(continueToken);
 
     String prevDir = null;
+    String lastKey = null;
+    int count = 0;
     while (ozoneKeyIterator.hasNext()) {
       OzoneKey next = ozoneKeyIterator.next();
       String relativeKeyName = next.getName().substring(prefix.length());
 
-      int depth =
+      int depth = StringUtils.countMatches(relativeKeyName, delimiter);
-          StringUtils.countMatches(relativeKeyName, delimiter);
+      if (delimiter != null) {
-
-      if (prefix.length() > 0 && !prefix.endsWith(delimiter)
-          && relativeKeyName.length() > 0) {
-        response.addPrefix(prefix + "/");
-        break;
-      }
         if (depth > 0) {
-        String dirName = relativeKeyName
+          // means key has multiple delimiters in its value.
-            .substring(0, relativeKeyName.indexOf(delimiter));
+          // ex: dir/dir1/dir2, where delimiter is "/" and prefix is dir/
+          String dirName = relativeKeyName.substring(0, relativeKeyName
+              .indexOf(delimiter));
           if (!dirName.equals(prevDir)) {
-          response.addPrefix(
+            response.addPrefix(prefix + dirName + delimiter);
-              prefix + dirName + delimiter);
             prevDir = dirName;
+            count++;
           }
         } else if (relativeKeyName.endsWith(delimiter)) {
+          // means or key is same as prefix with delimiter at end and ends with
+          // delimiter. ex: dir/, where prefix is dir and delimiter is /
           response.addPrefix(relativeKeyName);
-      } else if (relativeKeyName.length() > 0) {
+          count++;
-        KeyMetadata keyMetadata = new KeyMetadata();
+        } else {
-        keyMetadata.setKey(next.getName());
+          // means our key is matched with prefix if prefix is given and it
-        keyMetadata.setSize(next.getDataSize());
+          // does not have any common prefix.
-        keyMetadata.setETag("" + next.getModificationTime());
+          addKey(response, next);
-        keyMetadata.setStorageClass("STANDARD");
+          count++;
-        keyMetadata
+        }
-            .setLastModified(Instant.ofEpochMilli(next.getModificationTime()));
+      } else {
-        response.addKey(keyMetadata);
+        addKey(response, next);
+        count++;
+      }
+
+      if (count == maxKeys) {
+        lastKey = next.getName();
+        break;
       }
     }
+
+    response.setKeyCount(count);
+
+    if (count < maxKeys) {
+      response.setTruncated(false);
+    } else if(ozoneKeyIterator.hasNext()) {
+      response.setTruncated(true);
+      response.setNextToken(S3utils.generateContinueToken(lastKey));
+    } else {
+      response.setTruncated(false);
+    }
+
     response.setKeyCount(
         response.getCommonPrefixes().size() + response.getContents().size());
     return Response.ok(response).build();
@@ -253,4 +288,14 @@ public MultiDeleteResponse multiDelete(@PathParam("bucket") String bucketName,
     }
     return result;
   }
+
+  private void addKey(ListObjectResponse response, OzoneKey next) {
+    KeyMetadata keyMetadata = new KeyMetadata();
+    keyMetadata.setKey(next.getName());
+    keyMetadata.setSize(next.getDataSize());
+    keyMetadata.setETag("" + next.getModificationTime());
+    keyMetadata.setStorageClass("STANDARD");
+    keyMetadata.setLastModified(Instant.ofEpochMilli(next.getModificationTime()));
+    response.addKey(keyMetadata);
+  }
 }

hadoop-ozone/s3gateway/src/main/java/org/apache/hadoop/ozone/s3/endpoint/ListObjectResponse.java

@@ -60,6 +60,12 @@ public class ListObjectResponse {
   @XmlElement(name = "IsTruncated")
   private boolean isTruncated;
 
+  @XmlElement(name = "NextContinuationToken")
+  private String nextToken;
+
+  @XmlElement(name = "continueToken")
+  private String continueToken;
+
   @XmlElement(name = "Contents")
   private List<KeyMetadata> contents = new ArrayList<>();
 
@@ -148,6 +154,22 @@ public void addPrefix(String relativeKeyName) {
     commonPrefixes.add(new CommonPrefix(relativeKeyName));
   }
 
+  public String getNextToken() {
+    return nextToken;
+  }
+
+  public void setNextToken(String nextToken) {
+    this.nextToken = nextToken;
+  }
+
+  public String getContinueToken() {
+    return continueToken;
+  }
+
+  public void setContinueToken(String continueToken) {
+    this.continueToken = continueToken;
+  }
+
   public int getKeyCount() {
     return keyCount;
   }

hadoop-ozone/s3gateway/src/main/java/org/apache/hadoop/ozone/s3/util/S3Consts.java

@@ -15,5 +15,6 @@ private S3Consts() {
 
   public static final String COPY_SOURCE_HEADER = "x-amz-copy-source";
   public static final String STORAGE_CLASS_HEADER = "x-amz-storage-class";
+  public static final String ENCODING_TYPE = "url";
 
 }

hadoop-ozone/s3gateway/src/main/java/org/apache/hadoop/ozone/s3/util/S3utils.java

@@ -0,0 +1,73 @@
+package org.apache.hadoop.ozone.s3.util;
+
+import org.apache.commons.codec.DecoderException;
+import org.apache.commons.codec.binary.Hex;
+import org.apache.commons.codec.digest.DigestUtils;
+import org.apache.hadoop.ozone.s3.exception.OS3Exception;
+import org.apache.hadoop.ozone.s3.exception.S3ErrorTable;
+
+import java.nio.charset.StandardCharsets;
+
+/**
+ * Utility class for S3.
+ */
+public final class S3utils {
+
+  private S3utils() {
+
+  }
+  private static final String CONTINUE_TOKEN_SEPERATOR = "-";
+
+  /**
+   * Generate a continuation token which is used in get Bucket.
+   * @param key
+   * @return if key is not null return continuation token, else returns null.
+   */
+  public static String generateContinueToken(String key) {
+    if (key != null) {
+      byte[] byteData = key.getBytes(StandardCharsets.UTF_8);
+      String hex = Hex.encodeHexString(byteData);
+      String digest = DigestUtils.sha256Hex(key);
+      return hex + CONTINUE_TOKEN_SEPERATOR + digest;
+    } else {
+      return null;
+    }
+  }
+
+  /**
+   * Decode a continuation token which is used in get Bucket.
+   * @param key
+   * @return if key is not null return decoded token, otherwise returns null.
+   * @throws OS3Exception
+   */
+  public static String decodeContinueToken(String key) throws OS3Exception {
+    if (key != null) {
+      int indexSeparator = key.indexOf(CONTINUE_TOKEN_SEPERATOR);
+      if (indexSeparator == -1) {
+        throw S3ErrorTable.newError(S3ErrorTable.INVALID_ARGUMENT, key);
+      }
+      String hex = key.substring(0, indexSeparator);
+      String digest = key.substring(indexSeparator + 1);
+      try {
+        byte[] actualKeyBytes = Hex.decodeHex(hex);
+        String digestActualKey = DigestUtils.sha256Hex(actualKeyBytes);
+        if (digest.equals(digestActualKey)) {
+          return new String(actualKeyBytes);
+        } else {
+          OS3Exception ex = S3ErrorTable.newError(S3ErrorTable
+              .INVALID_ARGUMENT, key);
+          ex.setErrorMessage("The continuation token provided is incorrect");
+          throw ex;
+        }
+      } catch (DecoderException ex) {
+        OS3Exception os3Exception = S3ErrorTable.newError(S3ErrorTable
+            .INVALID_ARGUMENT, key);
+        os3Exception.setErrorMessage("The continuation token provided is " +
+            "incorrect");
+        throw os3Exception;
+      }
+    } else {
+      return null;
+    }
+  }
+}

hadoop-ozone/s3gateway/src/test/java/org/apache/hadoop/ozone/client/OzoneBucketStub.java

@@ -27,6 +27,7 @@
 import java.util.Iterator;
 import java.util.List;
 import java.util.Map;
+import java.util.TreeMap;
 import java.util.stream.Collectors;
 
 import org.apache.hadoop.hdds.client.ReplicationFactor;
@@ -116,7 +117,8 @@ public OzoneKeyDetails getKey(String key) throws IOException {
 
   @Override
   public Iterator<? extends OzoneKey> listKeys(String keyPrefix) {
-    return keyDetails.values()
+    Map<String, OzoneKey> sortedKey = new TreeMap<String, OzoneKey>(keyDetails);
+    return sortedKey.values()
         .stream()
         .filter(key -> key.getName().startsWith(keyPrefix))
         .collect(Collectors.toList())
@@ -126,7 +128,8 @@ public Iterator<? extends OzoneKey> listKeys(String keyPrefix) {
   @Override
   public Iterator<? extends OzoneKey> listKeys(String keyPrefix,
       String prevKey) {
-    return keyDetails.values()
+    Map<String, OzoneKey> sortedKey = new TreeMap<String, OzoneKey>(keyDetails);
+    return sortedKey.values()
         .stream()
         .filter(key -> key.getName().compareTo(prevKey) > 0)
         .filter(key -> key.getName().startsWith(keyPrefix))

hadoop-ozone/s3gateway/src/test/java/org/apache/hadoop/ozone/s3/endpoint/TestBucketGet.java

@@ -29,6 +29,8 @@
 import org.junit.Assert;
 import org.junit.Test;
 
+import static org.junit.Assert.fail;
+
 /**
  * Testing basic object list browsing.
  */
@@ -45,7 +47,7 @@ public void listRoot() throws OS3Exception, IOException {
 
     ListObjectResponse getBucketResponse =
         (ListObjectResponse) getBucket
-            .list("b1", "/", null, null, 100, "", null, null)
+            .list("b1", "/", null, null, 100, "", null, null, null, null)
             .getEntity();
 
     Assert.assertEquals(1, getBucketResponse.getCommonPrefixes().size());
@@ -68,8 +70,8 @@ public void listDir() throws OS3Exception, IOException {
     getBucket.setClient(client);
 
     ListObjectResponse getBucketResponse =
-        (ListObjectResponse) getBucket
+        (ListObjectResponse) getBucket.list("b1", "/", null, null, 100,
-            .list("b1", "/", null, null, 100, "dir1", null, null).getEntity();
+            "dir1", null, null, null, null).getEntity();
 
     Assert.assertEquals(1, getBucketResponse.getCommonPrefixes().size());
     Assert.assertEquals("dir1/",
@@ -85,13 +87,16 @@ public void listSubDir() throws OS3Exception, IOException {
     BucketEndpoint getBucket = new BucketEndpoint();
 
     OzoneClient ozoneClient =
-        createClientWithKeys("dir1/file2", "dir1/dir2/file2");
+        createClientWithKeys("dir1/file2", "dir1/dir2/file2", "dir1bh/file",
+            "dir1bha/file2");
 
     getBucket.setClient(ozoneClient);
 
     ListObjectResponse getBucketResponse =
         (ListObjectResponse) getBucket
-            .list("b1", "/", null, null, 100, "dir1/", null, null).getEntity();
+            .list("b1", "/", null, null, 100, "dir1/", null, null,
+                null, null)
+            .getEntity();
 
     Assert.assertEquals(1, getBucketResponse.getCommonPrefixes().size());
     Assert.assertEquals("dir1/dir2/",
@@ -103,6 +108,218 @@ public void listSubDir() throws OS3Exception, IOException {
 
   }
 
+
+  @Test
+  public void listWithPrefixAndDelimiter() throws OS3Exception, IOException {
+
+    BucketEndpoint getBucket = new BucketEndpoint();
+
+    OzoneClient ozoneClient =
+        createClientWithKeys("dir1/file2", "dir1/dir2/file2", "dir1bh/file",
+            "dir1bha/file2", "file2");
+
+    getBucket.setClient(ozoneClient);
+
+    ListObjectResponse getBucketResponse =
+        (ListObjectResponse) getBucket.list("b1", "/", null, null, 100,
+            "dir1", null, null, null, null).getEntity();
+
+    Assert.assertEquals(3, getBucketResponse.getCommonPrefixes().size());
+
+  }
+
+  @Test
+  public void listWithPrefixAndDelimiter1() throws OS3Exception, IOException {
+
+    BucketEndpoint getBucket = new BucketEndpoint();
+
+    OzoneClient ozoneClient =
+        createClientWithKeys("dir1/file2", "dir1/dir2/file2", "dir1bh/file",
+            "dir1bha/file2", "file2");
+
+    getBucket.setClient(ozoneClient);
+
+    ListObjectResponse getBucketResponse =
+        (ListObjectResponse) getBucket.list("b1", "/", null, null, 100,
+            "", null, null, null, null).getEntity();
+
+    Assert.assertEquals(3, getBucketResponse.getCommonPrefixes().size());
+    Assert.assertEquals("file2", getBucketResponse.getContents().get(0)
+        .getKey());
+
+  }
+
+  @Test
+  public void listWithPrefixAndDelimiter2() throws OS3Exception, IOException {
+
+    BucketEndpoint getBucket = new BucketEndpoint();
+
+    OzoneClient ozoneClient =
+        createClientWithKeys("dir1/file2", "dir1/dir2/file2", "dir1bh/file",
+            "dir1bha/file2", "file2");
+
+    getBucket.setClient(ozoneClient);
+
+    ListObjectResponse getBucketResponse =
+        (ListObjectResponse) getBucket.list("b1", "/", null, null, 100,
+            "dir1bh", null, null, "dir1/dir2/file2", null).getEntity();
+
+    Assert.assertEquals(2, getBucketResponse.getCommonPrefixes().size());
+
+  }
+
+  @Test
+  public void listWithContinuationToken() throws OS3Exception, IOException {
+
+    BucketEndpoint getBucket = new BucketEndpoint();
+
+    OzoneClient ozoneClient =
+        createClientWithKeys("dir1/file2", "dir1/dir2/file2", "dir1bh/file",
+            "dir1bha/file2", "file2");
+
+    getBucket.setClient(ozoneClient);
+
+    int maxKeys = 2;
+    // As we have 5 keys, with max keys 2 we should call list 3 times.
+
+    // First time
+    ListObjectResponse getBucketResponse =
+        (ListObjectResponse) getBucket.list("b1", null, null, null, maxKeys,
+            "", null, null, null, null).getEntity();
+
+    Assert.assertTrue(getBucketResponse.isTruncated());
+    Assert.assertTrue(getBucketResponse.getContents().size() == 2);
+
+    // 2nd time
+    String continueToken = getBucketResponse.getNextToken();
+    getBucketResponse =
+        (ListObjectResponse) getBucket.list("b1", null, null, null, maxKeys,
+            "", null, continueToken, null, null).getEntity();
+    Assert.assertTrue(getBucketResponse.isTruncated());
+    Assert.assertTrue(getBucketResponse.getContents().size() == 2);
+
+
+    continueToken = getBucketResponse.getNextToken();
+
+    //3rd time
+    getBucketResponse =
+        (ListObjectResponse) getBucket.list("b1", null, null, null, maxKeys,
+            "", null, continueToken, null, null).getEntity();
+
+    Assert.assertFalse(getBucketResponse.isTruncated());
+    Assert.assertTrue(getBucketResponse.getContents().size() == 1);
+
+  }
+
+  @Test
+  /**
+   * This test is with prefix and delimiter and verify continuation-token
+   * behavior.
+   */
+  public void listWithContinuationToken1() throws OS3Exception, IOException {
+
+    BucketEndpoint getBucket = new BucketEndpoint();
+
+    OzoneClient ozoneClient =
+        createClientWithKeys("dir1/file1", "dir1bh/file1",
+            "dir1bha/file1", "dir0/file1", "dir2/file1");
+
+    getBucket.setClient(ozoneClient);
+
+    int maxKeys = 2;
+    // As we have 5 keys, with max keys 2 we should call list 3 times.
+
+    // First time
+    ListObjectResponse getBucketResponse =
+        (ListObjectResponse) getBucket.list("b1", "/", null, null, maxKeys,
+            "dir", null, null, null, null).getEntity();
+
+    Assert.assertTrue(getBucketResponse.isTruncated());
+    Assert.assertTrue(getBucketResponse.getCommonPrefixes().size() == 2);
+
+
+    // 2nd time
+    String continueToken = getBucketResponse.getNextToken();
+    getBucketResponse =
+        (ListObjectResponse) getBucket.list("b1", "/", null, null, maxKeys,
+            "dir", null, continueToken, null, null).getEntity();
+    Assert.assertTrue(getBucketResponse.isTruncated());
+    Assert.assertTrue(getBucketResponse.getCommonPrefixes().size() == 2);
+
+
+    //3rd time
+    continueToken = getBucketResponse.getNextToken();
+    getBucketResponse =
+        (ListObjectResponse) getBucket.list("b1", "/", null, null, maxKeys,
+            "dir", null, continueToken, null, null).getEntity();
+
+    Assert.assertFalse(getBucketResponse.isTruncated());
+    Assert.assertTrue(getBucketResponse.getCommonPrefixes().size() == 1);
+
+  }
+
+  @Test
+  public void listWithContinuationTokenFail() throws OS3Exception, IOException {
+
+    BucketEndpoint getBucket = new BucketEndpoint();
+
+    OzoneClient ozoneClient =
+        createClientWithKeys("dir1/file2", "dir1/dir2/file2", "dir1bh/file",
+            "dir1bha/file2", "dir1", "dir2", "dir3");
+
+    getBucket.setClient(ozoneClient);
+
+    try {
+      ListObjectResponse getBucketResponse =
+          (ListObjectResponse) getBucket.list("b1", "/", null, null, 2,
+              "dir", null, "random", null, null).getEntity();
+      fail("listWithContinuationTokenFail");
+    } catch (OS3Exception ex) {
+      Assert.assertEquals("random", ex.getResource());
+      Assert.assertEquals("Invalid Argument", ex.getErrorMessage());
+    }
+
+  }
+
+
+  @Test
+  public void testStartAfter() throws IOException, OS3Exception {
+    BucketEndpoint getBucket = new BucketEndpoint();
+
+    OzoneClient ozoneClient =
+        createClientWithKeys("dir1/file1", "dir1bh/file1",
+            "dir1bha/file1", "dir0/file1", "dir2/file1");
+
+    getBucket.setClient(ozoneClient);
+
+    ListObjectResponse getBucketResponse =
+        (ListObjectResponse) getBucket.list("b1", null, null, null, 1000,
+            null, null, null, null, null).getEntity();
+
+    Assert.assertFalse(getBucketResponse.isTruncated());
+    Assert.assertTrue(getBucketResponse.getContents().size() == 5);
+
+    //As our list output is sorted, after seeking to startAfter, we shall
+    // have 4 keys.
+    String startAfter = "dir0/file1";
+
+    getBucketResponse =
+        (ListObjectResponse) getBucket.list("b1", null, null, null,
+            1000, null, null, null, startAfter, null).getEntity();
+
+    Assert.assertFalse(getBucketResponse.isTruncated());
+    Assert.assertTrue(getBucketResponse.getContents().size() == 4);
+
+    getBucketResponse =
+        (ListObjectResponse) getBucket.list("b1", null, null, null,
+            1000, null, null, null, "random", null).getEntity();
+
+    Assert.assertFalse(getBucketResponse.isTruncated());
+    Assert.assertTrue(getBucketResponse.getContents().size() == 0);
+
+
+  }
+
   private OzoneClient createClientWithKeys(String... keys) throws IOException {
     OzoneClient client = new OzoneClientStub();