RE: 回复： [qfs-devel:138] some questions of QFS

[Jeremy Fishman]

Hi,

Yes, it has actually always supported read and write, with caveats.

To enable write, currently you must pass option 'rrw'. I am planning on changing this requirement, though will maintain backwards compatibility.

Enabling Reed-Solomon encoding on files created via fuse requires a code change, and does not support seeking, but should work for any sequential I/O. I am planning a mount option to change the default behavior, but do not know of a way to specify this on a per-file basis through the fuse API.

Simultaneous read and write (O_RDWR) on a file is not supported.

I have some patches on jfishman/qfs master branch to enable overwrite-on-rename (per POSIX spec), and implement utime (set file mtime).

I will continue pushing my FUSE-related changes their until ready to integrate with master. My next change will likely be to support chroot'd mounts (a path prefix) via the mount URI.

With the changes I have made so far I am able to clone a git repository into a Fuse-mounted qfs directory and run git status, though commit operations aren't working yet. I figure if git can operate without complaint we are most of the way to full compliance, including pread-able pack files but no hard links (yet).

- Jeremy

Sent from my Android phone using TouchDown (www.nitrodesk.com)

-----Original Message-----
From: 飞 [duro...@qq.com]
Received: Wednesday, 20 Mar 2013, 8:35pm
To: Jeremy Fishman [jfis...@quantcast.com]
Subject: 回复： [qfs-devel:138] some questions of QFS

Hello Jeremy, dose the QFS fuse support read and write now ? thank you 

------------------ 原始邮件 ------------------

发件人: "Jeremy Fishman"<jfis...@quantcast.com>;

发送时间: 2013年2月22日(星期五) 凌晨2:40

收件人: "qfs-...@googlegroups.com"<qfs-...@googlegroups.com>;

抄送: "飞"<duro...@qq.com>;

主题: Re: [qfs-devel:138] some questions of QFS

I would like to amend answer (1) to mention that HDFS writes are not seekable and must be ordered.

  - Jeremy

On 02/21/2013 10:34 AM, Jeremy Fishman wrote:

Hello,

Welcome and thanks for trying QFS.  Your questions,

1. Yes, QFS supports the Hadoop FileSystem API, including seekable FSDataInputStream and FSDataOutputStream implementations.

2. Files are broken into 64MB chunks, so there is no limit on total file size beyond a practical limit on the number of chunks the meta server (namenode) can address in the memory allocated to it.

3. At this time the only traditional filesystem interface we support is a limited FUSE implementation wrapping the C++ client. This may be able to support virtual machine images, but as currently built will mount read-only.

It looks like there is a greater external support for FUSE and NFS mounts of HDFS.  See MountableHDFS, specifically Fuse-DFS, which may work with QFS since it uses a JNI interface to Hadoop, and HDFS NFS Proxy, which eliminates FUSE from buggy the NFS -> FUSE -> HDFS write path.

  - Jeremy

 

On 02/21/2013 12:58 AM, 飞 wrote:

Hello, I am looking for Distributed File System now, I saw QFS on github, I think it's a greate project.
but I have some question , can you give me any help ?  thank you

1. dose the QFS supports random file access ?
2. dose the QFS have a filesize limit ?
3. I want to using QFS to KVM Guests, QFS is right?

--
You received this message because you are subscribed to the Google Groups "QFS Development" group.
To unsubscribe from this group and stop receiving emails from it, send an email to qfs-devel+...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

-- 
Jeremy Fishman | Senior Software Engineer
Quantcast Corporation | Big Data Platform
201 3rd Street FL 2, San Francisco, CA 94103

-- 
Jeremy Fishman | Senior Software Engineer
Quantcast Corporation | Big Data Platform
201 3rd Street FL 2, San Francisco, CA 94103