[
https://issues.apache.org/jira/browse/VFS-698?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16865089#comment-16865089
]
David Septimus commented on VFS-698:
------------------------------------
As far as I can tell, this is only an issue when your {{FileSystem}} is using
{{CacheStrategy.ON_RESOLVE}}. This forces the {{FileSystem}} to call
{{refresh()}} on the newly created {{SftpFileObject}} (see:
{{AbstractFileSystem.resolveFile(...)}}) and because {{attrs}} is null at the
the time of resolution, a call is made to the server in order to determine the
file's {{FileType}} (see: {{SftpFileObject.refresh()}} and
{{SftpFileObject.doGetType()}}).
> SFTP file attributes are fetched multiple times leading to very slow
> directory listing
> --------------------------------------------------------------------------------------
>
> Key: VFS-698
> URL: https://issues.apache.org/jira/browse/VFS-698
> Project: Commons VFS
> Issue Type: Bug
> Affects Versions: 2.3
> Reporter: krishnan
> Priority: Major
>
> getChildren() applied on SftpFileObject is very slow compared to JSCH
> implementation. This is because, the SftpATTRS which is readily available for
> the children after an "ls" call is again fetched for each child file since
> they are independently resolved. So if a directory contains 10 files, it
> results in 1 (ls) + 10 (stat) calls to server.
> For a folder with 100 files (AWS), it took about 35 secs instead of 1.5 secs
> to getChildren().
>
> *doListChildrenResolved:*
> {{final FileObject fo =
> getFileSystem().resolveFile(getFileSystem().getFileSystemManager()}}
> \{{ .resolveName(getName(), UriParser.encode(name), NameScope.CHILD));}}
> {{{color:#ff0000}((SftpFileObject)
> FileObjectUtils.getAbstractFileObject(fo)).setStat(stat.getAttrs());{color}}}
>
> The resolveFile call, creates a SftpFileObject and calls its resolve method,
> which results in getting the (stats) SftpATTRS for each child file. This stat
> is already available as part of the 'ls' call we made. The setStat call above
> (highlighted is red) is redundant, since stat for each child file is already
> fetched one at a time.
> The solution would be to avoid getting the stat for each child file after an
> 'ls' call. May be, the framework makes it difficult to do this easily.
>
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
