[Gluster-users] Gluster vs NFS export sync

Santosh Pradhan spradhan at redhat.com
Wed Jul 16 03:30:45 UTC 2014

On 07/16/2014 08:41 AM, Franco Broi wrote:
> On Wed, 2014-07-16 at 08:32 +0530, Santosh Pradhan wrote:
>> On 07/15/2014 06:24 AM, Franco Broi wrote:
>>> I think the option I need is noac.
>> noac is no-attribute-caching in NFS. By default, NFS client would cache
>> the meta-data/attributes of FILEs for 3 seconds and directory for 30
>> seconds (default value). NFS client revalidates the attr cache by
>> sending periodic GETATTR call. But if the client is mounted with noac,
>> then it wont cache attrs. I dont think it's available in FUSE mount.
>> This is not applicable to WRITE I/O which keeps changing the attrs. This
>> cant make I/Os synchronous. I/Os may get cached (VM subsystem cache) in
>> either side (NFS client and NFS server side).
> The nfs man page says:
>   In addition to preventing the client from  caching  file
>   attributes, the noac option forces application writes to
>   become synchronous so  that  local  changes  to  a  file
>   become  visible  on  the  server immediately.  That way,
>   other clients can quickly detect recent writes when they
>   check the file's attributes.

My bad, I missed the man page. Yes, with noac option, all the WRITEs 
goes with FILE_SYNC (STABLE write) flag which means synchronous I/O.

> I'm guessing that an NFS disk exported with sync, forces all clients to
> be mounted as if they had used noac...
>> Thanks,
>> Santosh
>>> On Fri, 2014-07-11 at 16:10 +0530, Santosh Pradhan wrote:
>>>> On 07/11/2014 02:37 PM, Franco Broi wrote:
>>>>> On Fri, 2014-07-11 at 14:22 +0530, Santosh Pradhan wrote:
>>>>>> On 07/11/2014 06:10 AM, Franco Broi wrote:
>>>>>>> Hi
>>>>>>> Is there any way to make Gluster emulate the behaviour of a NFS
>>>>>>> filesystem exported with the sync option? By that I mean is it possible
>>>>>>> to write a file from one client and guarantee that the data will be
>>>>>>> instantly available on close to all other clients?
>>>>>> I think FUSE also supports the sync option with mount similar to NFS.
>>>>> There's 2 types of NFS sync, server and client but they behave very
>>>>> differently. With sync at the client end, NFS just calls fsync on close
>>>>> (I think),
>>>> No. If the NFS client is mounted with sync flag i.e. -o sync, then all
>>>> the WRITEs will be synchronous i.e. they wont be cached in client side
>>>> and they(WRITEs) will hit the server with FILE_SYNC (stable write) flag
>>>> which is nothing but FILE I/O with O_SYNC flag.
>>>>> with sync at the server end, all writes are synchronous.
>>>>> Opening the file with O_SYNC also doesn't give the desired effect
>>>>> although you'd think it would.
>>>> It should be same as above.
>>>> Thanks,
>>>> Santosh
>>>>>     NFS must have some way of communicating
>>>>> the file attribute changes to all the clients immediately the write
>>>>> completes.
>>>>>> Man page snippet for fuse(8):
>>>>>> OPTIONS
>>>>>>            Most of the generic mount options described in mount are
>>>>>> supported (ro, rw, suid, nosuid, dev,
>>>>>>            nodev, exec, noexec, atime, noatime, sync, async, dirsync).
>>>>>> Filesystems are mounted  with
>>>>>>            nodev,nosuid by default, which can only be overridden by a
>>>>>> privileged user.
>>>>>> Thanks,
>>>>>> Santosh
>>>>>>> Cheers,
>>>>>>> _______________________________________________
>>>>>>> Gluster-users mailing list
>>>>>>> Gluster-users at gluster.org
>>>>>>> http://supercolony.gluster.org/mailman/listinfo/gluster-users

More information about the Gluster-users mailing list