[
https://issues.apache.org/jira/browse/HDFS-13663?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16537647#comment-16537647
]
Xiao Chen commented on HDFS-13663:
----------------------------------
Thanks Yongjun for reporting this, and Shweta for working on it.
I agree it's better to fail explicitly here rather than depending on assertion,
which may silently goes by if the NN is not {{-ea}}'ed.
Comments on the patch:
- I understand the existing class has some inconsistent styles. Let's follow
the convention of having a space around the brackets.
- We can omit the else statement, because the if statement would throw.
- No need to keep the original assert as a comment. We can just remove it.
> Should throw exception when incorrect block size is set
> -------------------------------------------------------
>
> Key: HDFS-13663
> URL: https://issues.apache.org/jira/browse/HDFS-13663
> Project: Hadoop HDFS
> Issue Type: Bug
> Reporter: Yongjun Zhang
> Assignee: Shweta
> Priority: Major
> Attachments: HDFS-13663.001.patch
>
>
> See
> ./hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/BlockRecoveryWorker.java
> {code}
> void syncBlock(List<BlockRecord> syncList) throws IOException {
> newBlock.setNumBytes(finalizedLength);
> break;
> case RBW:
> case RWR:
> long minLength = Long.MAX_VALUE;
> for(BlockRecord r : syncList) {
> ReplicaState rState = r.rInfo.getOriginalReplicaState();
> if(rState == bestState) {
> minLength = Math.min(minLength, r.rInfo.getNumBytes());
> participatingList.add(r);
> }
> if (LOG.isDebugEnabled()) {
> LOG.debug("syncBlock replicaInfo: block=" + block +
> ", from datanode " + r.id + ", receivedState=" +
> rState.name() +
> ", receivedLength=" + r.rInfo.getNumBytes() + ", bestState=" +
> bestState.name());
> }
> }
> // recover() guarantees syncList will have at least one replica with
> RWR
> // or better state.
> assert minLength != Long.MAX_VALUE : "wrong minLength"; <= should
> throw exception
> newBlock.setNumBytes(minLength);
> break;
> case RUR:
> case TEMPORARY:
> assert false : "bad replica state: " + bestState;
> default:
> break; // we have 'case' all enum values
> }
> {code}
> when minLength is Long.MAX_VALUE, it should throw exception.
> There might be other places like this.
> Otherwise, we would see the following WARN in datanode log
> {code}
> WARN org.apache.hadoop.hdfs.server.datanode.DataNode: Can't replicate block
> xyz because on-disk length 11852203 is shorter than NameNode recorded length
> 9223372036854775807
> {code}
> where 9223372036854775807 is Long.MAX_VALUE.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
