[
https://issues.apache.org/jira/browse/HDFS-14661?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16890660#comment-16890660
]
Ayush Saxena commented on HDFS-14661:
-------------------------------------
I guess, we shouldn't stop from updating, we allow adding a mount entry too
when the destination is not available. We should just handle the quota
exception and ignore, It is admin's discretion, if he wants to create the
destination directory after the mount entry. You can do same as in HDFS-14331
i.e suppress any exception from Quota sync, while updating mount entry, Just
LOG it. For the test also you may use the existing one similarly add behind.
> RBF: updateMountTableEntry shouldn't update mountTableEntry if targetPath not
> exist
> -----------------------------------------------------------------------------------
>
> Key: HDFS-14661
> URL: https://issues.apache.org/jira/browse/HDFS-14661
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: rbf
> Affects Versions: 3.1.2
> Reporter: xuzq
> Priority: Major
> Attachments: HDFS-14661-HDFS-13891-001.patch,
> HDFS-14661-trunk-001.patch
>
>
> The updateMountTableEntry shouldn't update the mountEntry if the targetPath
> not exist.
> {code:java}
> @Override
> public UpdateMountTableEntryResponse updateMountTableEntry(
> UpdateMountTableEntryRequest request) throws IOException {
> UpdateMountTableEntryResponse response =
> getMountTableStore().updateMountTableEntry(request);
> MountTable mountTable = request.getEntry();
> if (mountTable != null && router.isQuotaEnabled()) {
> synchronizeQuota(mountTable.getSourcePath(),
> mountTable.getQuota().getQuota(),
> mountTable.getQuota().getSpaceQuota());
> }
> return response;
> }
> /**
> * Synchronize the quota value across mount table and subclusters.
> * @param path Source path in given mount table.
> * @param nsQuota Name quota definition in given mount table.
> * @param ssQuota Space quota definition in given mount table.
> * @throws IOException
> */
> private void synchronizeQuota(String path, long nsQuota, long ssQuota)
> throws IOException {
> if (router.isQuotaEnabled() &&
> (nsQuota != HdfsConstants.QUOTA_DONT_SET
> || ssQuota != HdfsConstants.QUOTA_DONT_SET)) {
> HdfsFileStatus ret = this.router.getRpcServer().getFileInfo(path);
> if (ret != null) {
> this.router.getRpcServer().getQuotaModule().setQuota(path, nsQuota,
> ssQuota, null);
> }
> }
> }
> {code}
> As above, updateMountTableEntry update one mountEntry:
> # update mountEntry in zookeeper
> # synchronizeQuota(maybe throw some exception like "Directory does not
> exist")
>
> if synchronizeQuota throw some exception, will return some exception to
> dfsRouterAdmin, but the new mountEntry has been updated to zk.
> It's clearly not what we would expect.
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
