Stitched blobs support different sized component blobs #1175

dharju · 2019-05-23T11:28:07Z

Changed stitched blobs to work with blobs with different data content sizes.

Note: Still need to add serializeMetadataContentV3 into BlobIdTransformer, but regular stitched blob operation changes are ready to be looked at

… sizes

codecov-io · 2019-05-23T11:42:28Z

Codecov Report

Merging #1175 into master will increase coverage by 13.55%.
The diff coverage is n/a.

@@              Coverage Diff              @@
##             master    #1175       +/-   ##
=============================================
+ Coverage     70.06%   83.62%   +13.55%     
+ Complexity     5378       57     -5321     
=============================================
  Files           428        6      -422     
  Lines         32791      348    -32443     
  Branches       4136       38     -4098     
=============================================
- Hits          22975      291    -22684     
+ Misses         8691       45     -8646     
+ Partials       1125       12     -1113

Impacted Files	Coverage Δ	Complexity Δ
...ava/com.github.ambry/config/PerformanceConfig.java
...main/java/com.github.ambry.cloud/CloudReplica.java
...m.github.ambry.replication/ReplicationMetrics.java
...ava/com.github.ambry/frontend/SecurityService.java
...b.ambry.cloud/CloudBlobCryptoAgentFactoryImpl.java
...ain/java/com.github.ambry/frontend/Operations.java
...n/java/com.github.ambry.clustermap/Datacenter.java
...github.ambry.messageformat/TtlUpdateSubRecord.java
...i/src/main/java/com.github.ambry/network/Port.java
...n/java/com.github.ambry.network/CompositeSend.java
... and 405 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 09f0d30...870c22e. Read the comment docs.

jsjtzyy

Some minor comments. Will take a look at tests soon.

jsjtzyy · 2019-05-26T05:18:43Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/MessageFormatRecord.java

+     */
+    public static int getMetadataContentSize(int keySize, int numberOfKeys) {
+      return Version_Field_Size_In_Bytes + TOTAL_SIZE_FIELD_SIZE_IN_BYTES + NUM_OF_KEYS_FIELD_SIZE_IN_BYTES + (
+          numberOfKeys * keySize * SIZE_OF_BLOB_FIELD_IN_BYTES);


Correct me if I am wrong, I think here should be numberOfKeys * ( keySize + SIZE_OF_BLOB_FIELD_IN_BYTES )

You're right, adding unit test to catch this

jsjtzyy · 2019-05-26T05:19:49Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/MessageFormatRecord.java

+     * Serialize a metadata content record.
+     * @param outputBuffer
+     * @param totalSize
+     * @param keysAndContentSizes


Could you complete the java doc here?

jsjtzyy · 2019-05-26T05:34:49Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/CompositeBlobInfo.java

+
+  /**
+   * Construct a {@link CompositeBlobInfo} object.
+   * @param keysAndContentSizes


jsjtzyy · 2019-05-26T05:34:57Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/CompositeBlobInfo.java

+   * blob
+   * @param start inclusive starting byte index
+   * @param end inclusive ending byte index
+   * @return


jsjtzyy · 2019-05-26T05:36:44Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/CompositeBlobInfo.java

+   */
+  public List<StoreKeyAndSizeAndOffset> getStoreKeysInByteRange(long start, long end) {
+    if (end < start || start < 0L || end >= totalSize) {
+      throw new IllegalArgumentException("Bad input parameters, start="+start+" end="+end+" totalSize="+totalSize);


format this file please

jsjtzyy · 2019-05-26T05:54:22Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/CompositeBlobInfo.java

+    int idx = Collections.binarySearch(offsets, start);
+    //binarySearch returns -(insertion point) - 1 if not an exact match, which points to the index of
+    //the first element greater than the key, or list.size() if all elements in the list are
+    // less than the specified key.


I know this description comes from Collections.binarySearch java doc, however, in current context, we'd better use target or start instead of key here.

sure, changed to 'start'

jsjtzyy · 2019-05-26T05:57:59Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/CompositeBlobInfo.java

+    //the first element greater than the key, or list.size() if all elements in the list are
+    // less than the specified key.
+    if (idx < 0) {
+      idx = -idx-2; //points to element with offset closest to but less than 'start'


Although it seems that idx won't be -1, it is worthwhile to add comment here to explain it.

The first offset will always be '0', so idx will never be -1 (it would only be -1 if 'start' was less than the first offset)

jsjtzyy · 2019-05-26T06:11:21Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/MessageFormatRecord.java

+    public static CompositeBlobInfo deserializeMetadataContentRecord(DataInputStream stream,
+        StoreKeyFactory storeKeyFactory) throws IOException {
+      List<Pair<StoreKey, Long>> keysAndContentSizes = new ArrayList<>();
+      stream.readLong(); //total size


I suggest passing the total size into CompositeBlobInfo constructor and compare it with the calculated total size (which is last in your code). If they don't match, throw an exception.

I'm not for adding the total size as a param to the CompositeBlobInfo ctor to just perform a check, but I will add the check here, and in serializeMetadataContentRecord

jsjtzyy

A few more comments. Feel free to drop some if I misunderstood the code logic.

jsjtzyy · 2019-05-26T16:53:14Z

ambry-router/src/main/java/com.github.ambry.router/PutOperation.java

-        List<StoreKey> orderedChunkIdList = new ArrayList<>(indexToChunkIds.values());
-        buf = MetadataContentSerDe.serializeMetadataContent(intermediateChunkSize, getBlobSize(), orderedChunkIdList);
+        List<Pair<StoreKey, Long>> orderedChunkIdList = indexToChunkIds.values().stream().collect(
+            Collectors.toList());


Format this file. Also looks like we can still use new ArrayList<>(indexToChunkIds.values()), any specific reason to use indexToChunkIds.values().stream().collect(Collectors.toList()) here?

I don't think there's a specific reason, replaced

jsjtzyy · 2019-05-26T17:06:53Z

ambry-router/src/main/java/com.github.ambry.router/PutOperation.java

@@ -1619,7 +1623,8 @@ private void finalizeMetadataChunk() {
     * @return a list of all of the successfully put chunk ids associated with this blob
     */
    List<StoreKey> getSuccessfullyPutChunkIds() {
-      return new ArrayList<>(indexToChunkIds.values());
+      return indexToChunkIds.values().stream().map(b -> b.getFirst()).collect(
+          Collectors.toList());


Please format here. And maybe use method reference to simplify:
indexToChunkIds.values().stream().map(Pair::getFirst).collect(Collectors.toList())

jsjtzyy · 2019-05-26T18:27:03Z

ambry-router/src/main/java/com.github.ambry.router/GetBlobOperation.java

@@ -937,9 +943,9 @@ protected ByteBuffer filterChunkToRange(ByteBuffer buf) {
        buf.position(0);
        buf.limit(0);
      } else {
-        long startOffsetInThisChunk = chunkIndex == 0 ? resolvedByteRange.getStartOffset() % chunkSize : 0;


I wonder previously if chunkIndex > 0, why the startOffsetInThisChunk is 0 here?

because it's relative to the chunk itself. The chunkIndexes only refer to chunks that are within a byte range. If the chunkIndex refers to a chunk beyond the first chunk, it will have to start at 0.

Sorry about the confusion around chunkIndex. Since we now have the offset variable, you could consider changing this to long startOffsetInThisChunk = Math.max(resolvedByteRange.getStartOffset() - offset, 0) if you prefer.

jsjtzyy · 2019-05-26T18:27:42Z

ambry-router/src/main/java/com.github.ambry.router/GetBlobOperation.java

        long endOffsetInThisChunk =
-            chunkIndex == (numChunksTotal - 1) ? (resolvedByteRange.getEndOffset() % chunkSize) + 1 : chunkSize;


One more minor question: is endOffsetInThisChunk exclusive or inclusive?
Looks like it is exclusive, right?

exclusive since it's used in Buffer.limit(). I can rename it endOffsetInThisChunkExclusive

If you feel that it is cleaner, you could also express this as long endOffsetInThisChunkExclusive = Math.min(resolvedByteRange.getEndOffset() - offset + 1, chunkSize)

jsjtzyy · 2019-05-26T18:44:13Z

ambry-router/src/main/java/com.github.ambry.router/GetBlobOperation.java

    private BlobProperties serverBlobProperties;

    /**
     * Construct a FirstGetChunk and initialize it with the {@link BlobId} of the overall operation.
     */
    FirstGetChunk() {
-      super(-1, blobId);
+      super(-1, new CompositeBlobInfo.StoreKeyAndSizeAndOffset(blobId, 0L, -1L));


It seems the chunkSize is initialized to -1L here and I don't see it gets populated again because you removed several codes regarding setting chunkSize. Correct me if I am wrong.

That's right, but that's ok since chunkSize is only used when calculating the end of data chunk for a byte range operation if that data chunk is not the last data chunk. For simple data blobs, the FirstChunk will always be the last data chunk so it's unused, and for metadata blobs the size of the FirstChunk is irrelevant.

jsjtzyy · 2019-05-26T18:47:38Z

ambry-router/src/main/java/com.github.ambry.router/GetBlobOperation.java

@@ -1051,7 +1057,6 @@ protected void maybeProcessCallbacks() {
                  decryptCallbackResultInfo.result.getDecryptedUserMetadata().array());
            }
            totalSize = decryptCallbackResultInfo.result.getDecryptedBlobContent().remaining();
-            chunkSize = totalSize;


Similar to previous comment. Where the chunkSize is populated? I notice that getKeysAndSizesAndOffsets contains chunk size info, however, note that filterChunkToRange method directly uses variable chunkSize, which might not be initialized. Please double check code there.

Also, we can discuss offline about the chunk size of a simple blob.

It's ok since chunkSize is only used when calculating the end of data chunk for a byte range operation if that data chunk is not the last data chunk. For simple data blobs, the FirstChunk will always be the last data chunk so it's unused, and for metadata blobs the size of the FirstChunk is irrelevant.

If you choose to use min/max in filterChunkToRange, I think it will be required to set chunkSize appropriately on simple blobs

cgtz

Will continue to review

cgtz · 2019-05-23T15:43:44Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/CompositeBlobInfo.java

+    }
+  }
+
+
  private final int chunkSize;


Are chunkSize/keys still needed after introducing StoreKeyAndSizeAndOffset?

Chunk size would still be used for deserializing V2 metadata content and keys is used so that getKeys() call doesn't have to parse the StoreKeyAndSizeAndOffset object every time it is called

cgtz · 2019-05-28T17:52:45Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/CompositeBlobInfo.java

+   * POJO class for holding a store key, the size of the data content it refers to,
+   * and the offset of the data in the larger composite blob
+   */
+  public static class StoreKeyAndSizeAndOffset {


Is there a more concise name that we could choose for this class? This name might be too specific (i.e. it will become outdated if we add another field in the future)

Maybe ChunkMetadata?

cgtz · 2019-05-28T17:54:49Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/CompositeBlobInfo.java

+    if (idx < 0) {
+      idx = -idx-2; //points to element with offset closest to but less than 'start'
+    }
+    List<StoreKeyAndSizeAndOffset> ans = new ArrayList<>();


could we use List.subList() here to avoid extra copies?

We can, and another cool thing about using subList() is that I can do another binary search for the end index replacing the linearly scaling while loop and make this method a O(log(n)) operation (since subList() itself is a constant operation for ArrayList). Great suggestion!

cgtz · 2019-05-28T18:01:12Z

ambry-router/src/main/java/com.github.ambry.router/PutOperation.java

+        List<Pair<StoreKey, Long>> orderedChunkIdList = indexToChunkIds.values().stream().collect(
+            Collectors.toList());
+
+        buf = MetadataContentSerDe.serializeMetadataContentV3(getBlobSize(), orderedChunkIdList);


We don't want to start writing in v3 before all frontends have been deployed with read support for metadata content v3.

Perhaps we should add a router config for the metadata content version like we have for blob id version and then pass the version number into the MetadataContentSerDe.serializeMetadataContent() method

added config

dharju · 2019-05-28T18:23:07Z

Will be adding more unit tests (specifically unit test for MessageFormatRecord) and incorporation into BlobIdTransformer

…d to be V2 specific, fixed V3 metadata content record size calculation

dharju · 2019-05-30T17:49:27Z

Fixing failing check, also changing PutManagerTest, GetManagerTest, and NonBlockingRouterTest to test against both old and new versions of metadata content

…t, fixed tests to test both versions of metadata content

dharju · 2019-05-30T21:40:36Z

Looking at GetBlobOperationTest failure, appears to be triggered by random chunk size, indicates that a kind of corner case is missing from the test, will fix

dharju · 2019-05-30T21:52:28Z

Source of error was test with a very low probability of generating an incorrect test input. Fixing

jsjtzyy · 2019-05-31T21:40:48Z

ambry-api/src/main/java/com.github.ambry/config/RouterConfig.java

+  /**
+   * The version to use for new metadata blobs.
+   */
+  @Config("router.metadata.content.version")


Is it better to call this router.metadata.blob.content.version ?

I like router.metadata.content.version because it matches the terminology used in the Metadata_Content_Format_* classes

jsjtzyy · 2019-06-03T16:43:58Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/CompositeBlobInfo.java

+
+  /**
+   * Return the version of the metadata content record that this object was deserialized from
+   * @return


please complete this java doc (move line 140 to line 141)

jsjtzyy · 2019-06-04T00:45:37Z

ambry-messageformat/src/test/java/com.github.ambry.messageformat/MessageFormatRecordTest.java

+  }
+
+  private CompositeBlobInfo deserializeMetadataContentV3(ByteBuffer metadataContent, StoreKeyFactory storeKeyFactory)
+      throws MessageFormatException, IOException {


no MessageFormatException thrown from this method

jsjtzyy · 2019-06-04T00:55:06Z

ambry-messageformat/src/test/java/com.github.ambry.messageformat/MessageFormatRecordTest.java

+        List<Pair<StoreKey, Long>> keyAndContentSizes = new ArrayList<>();
+        keyAndContentSizes.add(new Pair<>(keys.get(0), (long) chunkSizes[n][0]));
+        keyAndContentSizes.add(new Pair<>(keys.get(1), (long) chunkSizes[n][1]));
+        MetadataContentSerDe.serializeMetadataContentV3(totalSizes[n], keyAndContentSizes);
        fail("Should have failed to serialize");


You may also test that keys with different size would throw exception here.

jsjtzyy · 2019-06-04T01:07:44Z

ambry-replication/src/test/java/com.github.ambry.replication/BlobIdTransformerTest.java

-    TransformationOutput output = transformer.transform(inputAndExpected.getInput());
-    assertNull("output exception should be null", output.getException());
-    verifyOutput(output.getMsg(), inputAndExpected.getExpected());
+    for (int i = 0; i < 2; i++) {


How about moving this piece of code to BlobIdTransformerTest ctor and pass in parameters to test both V2 and V3 metadataContent in this class ?

agreed, changed

cgtz · 2019-06-04T20:43:06Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/CompositeBlobInfo.java

+    if (endIdx < 0) {
+      endIdx = -endIdx - 2; //points to element with offset closest to but less than 'end'
+    }
+    endIdx++;//List.subList expects an exclusive index


add a space before comment

cgtz · 2019-06-04T21:15:38Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/CompositeBlobInfo.java

@@ -62,4 +126,49 @@ public int getChunkSize() {
  public List<StoreKey> getKeys() {


An alternative to materializing a list of StoreKeys for every CompositeBlobInfo is to return a "view" of chunkMetadataList:

public List<StoreKey> getKeys() { return new AbstractList<StoreKey>() { @Override public StoreKey get(int index) { return chunkMetadataList.get(index).getStoreKey(); } @Override public int size() { return chunkMetadataList.size(); } }; }

It seems like this would suffice for the two use cases that need the getKeys() method, the getChunkIdsOnly option in the router and BlobIdTransformer, as both of them are just interested in iterating through all of the keys and not mutating the list.

Cool idea, I'll use it

cgtz · 2019-06-04T21:23:08Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/MessageFormatRecord.java

+   *  - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
+   * |         |              |            |            |          |            |            |          |
+   * | version |  total size  | # of keys  |  size of   |   key1   |  size of   |    key2    |  ......  |
+   * |(2 bytes)|  (8 bytes)   | (4 bytes)  | key1 blob  |          | key2 blob  |            |  ......  |


Is the totalSize field meant for sanity checking that sum(chunksizes) == totalsize or are there other places where it is used?

Yeah it's only used for checking

cgtz · 2019-06-04T21:51:35Z

ambry-messageformat/src/main/java/com.github.ambry.messageformat/CompositeBlobInfo.java

+   * @return the list of {@link ChunkMetadata} that lie upon the inclusive range
+   * between 'start' and 'end'
+   */
+  public List<ChunkMetadata> getStoreKeysInByteRange(long start, long end) {


we might want to add unit tests for CompositeBlobInfo testing for edge cases of this method.
e.g.:

0 byte chunks in the middle of the blob (Potentially a possibility with segmented blobs)

start/end = last byte of blob

agreed, actually writing that now. CompositeBlobInfo didn't have a unit test before because it was a POJO class but with this method it should be tested.

Can you explain "0 byte chunks in the middle of the blob"?

Sure. So I was thinking that since we allow 0 byte simple blob uploads, they are valid candidates for stitching. So, someone could potentially create a blob with chunk sizes like: 25, 30, 0, 0, 70, 0, 5

If we call this method with a start offset of 30, would the range start from the second or third chunk?

As discussed in person, going to change the code so that stitched blobs will not allow blobs of 0 length to be stitched (will throw an exception if that happens)

cgtz · 2019-06-04T22:19:44Z

ambry-router/src/main/java/com.github.ambry.router/GetBlobOperation.java

     */
-    void initialize(int index, BlobId id) {
+    void initialize(int index, CompositeBlobInfo.ChunkMetadata keyAndSizeAndOffset) {


change variable names in this class (and others) from keyAndSizeAndOffset to chunkMetadata

cgtz · 2019-06-04T22:49:38Z

ambry-router/src/main/java/com.github.ambry.router/GetBlobOperation.java

@@ -937,9 +943,9 @@ protected ByteBuffer filterChunkToRange(ByteBuffer buf) {
        buf.position(0);
        buf.limit(0);
      } else {
-        long startOffsetInThisChunk = chunkIndex == 0 ? resolvedByteRange.getStartOffset() % chunkSize : 0;


Sorry about the confusion around chunkIndex. Since we now have the offset variable, you could consider changing this to long startOffsetInThisChunk = Math.max(resolvedByteRange.getStartOffset() - offset, 0) if you prefer.

cgtz · 2019-06-04T22:53:13Z

ambry-router/src/main/java/com.github.ambry.router/GetBlobOperation.java

        long endOffsetInThisChunk =
-            chunkIndex == (numChunksTotal - 1) ? (resolvedByteRange.getEndOffset() % chunkSize) + 1 : chunkSize;


If you feel that it is cleaner, you could also express this as long endOffsetInThisChunkExclusive = Math.min(resolvedByteRange.getEndOffset() - offset + 1, chunkSize)

dharju · 2019-06-05T20:09:08Z

made the max, min changes suggested above

cgtz · 2019-06-07T20:50:19Z

ambry-router/src/main/java/com.github.ambry.router/GetBlobOperation.java

-        chunkIdIterator = keys.listIterator();
-        numChunksTotal = keys.size();
-        dataChunks = new GetChunk[Math.min(keys.size(), NonBlockingRouter.MAX_IN_MEM_CHUNKS)];
+        chunkIdIterator = keysAndSizesAndOffsets.listIterator();


change this name to chunkMetadataList

thought I caught them all... done

cgtz · 2019-06-07T20:55:04Z

ambry-router/src/main/java/com.github.ambry.router/GetBlobOperation.java

@@ -943,8 +943,9 @@ protected ByteBuffer filterChunkToRange(ByteBuffer buf) {
        buf.position(0);
        buf.limit(0);
      } else {
-        long startOffsetInThisChunk = Math.max(resolvedByteRange.getStartOffset() - offset, 0);
-        long endOffsetInThisChunkExclusive = Math.min(resolvedByteRange.getEndOffset() - offset + 1, chunkSize);
+        long startOffsetInThisChunk = chunkIndex == 0 ? resolvedByteRange.getStartOffset() - offset : 0;


Why did you decide to revert the min/max change?

It broke a GetBlobOperationTest test so I changed it back

cgtz · 2019-06-07T20:56:13Z

ambry-router/src/main/java/com.github.ambry.router/GetBlobOperation.java

@@ -1051,7 +1057,6 @@ protected void maybeProcessCallbacks() {
                  decryptCallbackResultInfo.result.getDecryptedUserMetadata().array());
            }
            totalSize = decryptCallbackResultInfo.result.getDecryptedBlobContent().remaining();
-            chunkSize = totalSize;


If you choose to use min/max in filterChunkToRange, I think it will be required to set chunkSize appropriately on simple blobs

…List

cgtz

One last comment, but other than that LGTM

cgtz · 2019-06-10T16:43:39Z

ambry-router/src/main/java/com.github.ambry.router/PutOperation.java

@@ -1604,12 +1612,21 @@ private void finalizeMetadataChunk() {
              passedInBlobProperties.isEncrypted(), passedInBlobProperties.getExternalAssetTag());
      if (isStitchOperation() || getNumDataChunks() > 1) {
        // values returned are in the right order as TreeMap returns them in key-order.
-        List<StoreKey> orderedChunkIdList = new ArrayList<>(indexToChunkIds.values());
-        buf = MetadataContentSerDe.serializeMetadataContent(intermediateChunkSize, getBlobSize(), orderedChunkIdList);
+        List<Pair<StoreKey, Long>> orderedChunkIdList = new ArrayList<>(indexToChunkIdsAndChunkSizes.values());


Move this to inside the else if (line 1620) so that we don't have to construct it for v2

dharju added 3 commits May 23, 2019 03:57

Changed stitched blobs to work with blobs with different data content…

6d89c74

… sizes

Added comments

3de424e

Added comment for serializeMetadataContentV3

c01996c

dharju requested review from cgtz and jsjtzyy May 23, 2019 11:28

jsjtzyy requested changes May 26, 2019

View reviewed changes

jsjtzyy reviewed May 26, 2019

View reviewed changes

cgtz suggested changes May 28, 2019

View reviewed changes

dharju added 3 commits May 29, 2019 10:48

Added more unit tests for V3 metadata content record, renamed a metho…

f9e62b4

…d to be V2 specific, fixed V3 metadata content record size calculation

Reformatted code

8440428

review comments

d6f1cb8

dharju added 2 commits May 30, 2019 12:49

Changed PutOperation to work with versions 2 and 3 of metadata conten…

4a85e75

…t, fixed tests to test both versions of metadata content

Added back some V2 asserts in PutManagerTest

047ba93

Fixed test that could give unexepcted wrong input

8c6b0af

jsjtzyy reviewed Jun 4, 2019

View reviewed changes

cgtz reviewed Jun 4, 2019

View reviewed changes

dharju added 2 commits June 4, 2019 18:58

Added tests for CompositeBlobInfo.getStoreKeysInByteRange

4f3de93

Added license header

8e76753

dharju added 2 commits June 5, 2019 13:16

review comments

c4e6c6a

Reverting max-min change in GetBlobOperation#filterChunkToRange

3c66429

cgtz reviewed Jun 7, 2019

View reviewed changes

Changed name of variable from keysAndSizesAndOffsets to chunkMetadata…

c05a702

…List

cgtz approved these changes Jun 10, 2019

View reviewed changes

jsjtzyy approved these changes Jun 10, 2019

View reviewed changes

Moved line per review comment

870c22e

cgtz merged commit 580097e into linkedin:master Jun 10, 2019

		long endOffsetInThisChunk =
		chunkIndex == (numChunksTotal - 1) ? (resolvedByteRange.getEndOffset() % chunkSize) + 1 : chunkSize;

		@@ -62,4 +126,49 @@ public int getChunkSize() {
		public List<StoreKey> getKeys() {

Stitched blobs support different sized component blobs #1175

Stitched blobs support different sized component blobs #1175

Conversation

dharju commented May 23, 2019

codecov-io commented May 23, 2019 • edited Loading

Codecov Report

jsjtzyy left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jsjtzyy left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cgtz left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dharju commented May 28, 2019

dharju commented May 30, 2019

dharju commented May 30, 2019

dharju commented May 30, 2019

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jsjtzyy Jun 4, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cgtz Jun 5, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov-io commented May 23, 2019 •

edited

Loading

jsjtzyy Jun 4, 2019 •

edited

Loading

cgtz Jun 5, 2019 •

edited

Loading