[jira] [Resolved] (HDDS-11223) Datanode checksum validation in EC bucket does not work

["Tsz-wo Sze (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/HDDS-11223?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Tsz-wo Sze resolved HDDS-11223.
-------------------------------
    Fix Version/s: 1.5.0
       Resolution: Fixed

The pull request is now merged.  Thanks, [~Siberian Sky]!

> Datanode checksum validation in EC bucket does not work
> -------------------------------------------------------
>
>                 Key: HDDS-11223
>                 URL: https://issues.apache.org/jira/browse/HDDS-11223
>             Project: Apache Ozone
>          Issue Type: Bug
>    Affects Versions: 1.5.0
>            Reporter: Kirill Sizov
>            Assignee: Kirill Sizov
>            Priority: Major
>              Labels: pull-request-available
>             Fix For: 1.5.0
>
>
> *Steps to reproduce*
> - Enable hdds.datanode.chunk.data.validation.check property.
> - put any file into a EC bucket
> The output a client gets is:
> {noformat}
> INTERNAL_ERROR No enough datanodes to choose. TotalNodes = 10 AvailableNodes = 4 RequiredNodes = 5 ExcludedNodes = 6 UsedNodes = 0
> {noformat}
> *Investigation details*
> The issue comes from incorrect checksum validation on the datanode side.
> Here are the details:
> 1. {{KeyValueHandler.validateChunkChecksumData}} calls {{Checksum.verifyChecksum}} 
> 2. {{Checksum.verifyChecksum}} calls {{checksum.computeChecksum(data)}} to calculate a checksum over the received data
> 3.  {{checksum.computeChecksum}} has the following block:
> {code}
> for (ByteBuffer b : data.iterate(bytesPerChecksum)) {
>       checksumList.add(computeChecksum(b, function, bytesPerChecksum));
>     }
> {code}
> The issue is that {{data.iterate()}} may return byte buffers of an arbitrary size. 
> Specifically, it is observed on {{ChunkBufferImplWithByteBufferList}} which was created from a {{RopeByteString}} in {{KeyValueHandler.handleWriteChunk}}:
> {code}
>  data =
>             ChunkBuffer.wrap(writeChunk.getData().asReadOnlyByteBufferList());
> {code}
> So checksums rely on the fact the the requested data buffer is split into parts of the requested size (and a smaller final one),
> but {{ChunkBufferImplWithByteBufferList}} for some reason declares the opposite.
> {code}
> // See ChunkBufferImplWithByteBufferList:
>   /**
>    * Returns the next buffer in the list irrespective of the bufferSize.
>    */
>   @Override
>   public Iterable<ByteBuffer> iterate(int bufferSize) {
> {code}
> And that implementation contradicts with the interface:
> {code}
> // ChunkBuffer:
>   /**
>    * Iterate the buffer from the current position to the current limit.
>    * Upon the iteration complete,
>    * the buffer's position will be equal to its limit.
>    *
>    * @param bufferSize the size of each buffer in the iteration.
>    */
>   Iterable<ByteBuffer> iterate(int bufferSize);
> {code}



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@ozone.apache.org
For additional commands, e-mail: issues-help@ozone.apache.org