[
https://issues.apache.org/jira/browse/HDFS-15386?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Stephen O'Donnell resolved HDFS-15386.
--------------------------------------
Resolution: Fixed
> ReplicaNotFoundException keeps happening in DN after removing multiple DN's
> data directories
> --------------------------------------------------------------------------------------------
>
> Key: HDFS-15386
> URL: https://issues.apache.org/jira/browse/HDFS-15386
> Project: Hadoop HDFS
> Issue Type: Bug
> Reporter: Toshihiro Suzuki
> Assignee: Toshihiro Suzuki
> Priority: Major
> Fix For: 3.0.4, 3.2.2, 2.10.1, 3.3.1, 3.4.0, 3.1.5
>
>
> When removing volumes, we need to invalidate all the blocks in the volumes.
> In the following code (FsDatasetImpl), we keep the blocks that will be
> invalidate in *blkToInvalidate* map. However as the key of the map is *bpid*
> (Block Pool ID), it will be overwritten by other removed volumes. As a
> result, the map will have only the blocks of the last volume we are removing,
> and invalidate only them:
> {code:java}
> for (String bpid : volumeMap.getBlockPoolList()) {
> List<ReplicaInfo> blocks = new ArrayList<>();
> for (Iterator<ReplicaInfo> it =
> volumeMap.replicas(bpid).iterator(); it.hasNext();) {
> ReplicaInfo block = it.next();
> final StorageLocation blockStorageLocation =
> block.getVolume().getStorageLocation();
> LOG.trace("checking for block " + block.getBlockId() +
> " with storageLocation " + blockStorageLocation);
> if (blockStorageLocation.equals(sdLocation)) {
> blocks.add(block);
> it.remove();
> }
> }
> blkToInvalidate.put(bpid, blocks);
> }
> {code}
> [https://github.com/apache/hadoop/blob/704409d53bf7ebf717a3c2e988ede80f623bbad3/hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/fsdataset/impl/FsDatasetImpl.java#L580-L595]
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
