[Gluster-devel] Trash translator for GlusterFS

[Vijay Bellur]

On 02/17/2014 03:06 PM, Pranith Kumar Karampuri wrote:
>
>
> ----- Original Message -----
>> From: "Chetan Risbud" <crisbud at redhat.com>
>> To: "Pranith Kumar Karampuri" <pkarampu at redhat.com>
>> Cc: "Jiffin Thottan" <jthottan at redhat.com>, gluster-devel at nongnu.org
>> Sent: Monday, February 17, 2014 3:00:07 PM
>> Subject: Re: [Gluster-devel] Trash translator for GlusterFS
>>
>>
>>
>> ----- Original Message -----
>> From: "Pranith Kumar Karampuri" <pkarampu at redhat.com>
>> To: "Jiffin Thottan" <jthottan at redhat.com>
>> Cc: gluster-devel at nongnu.org
>> Sent: Monday, February 17, 2014 2:51:44 PM
>> Subject: Re: [Gluster-devel] Trash translator for GlusterFS
>>
>>
>>
>> ----- Original Message -----
>>> From: "Jiffin Thottan" <jthottan at redhat.com>
>>> To: "Niels de Vos" <ndevos at redhat.com>
>>> Cc: gluster-devel at nongnu.org
>>> Sent: Monday, February 17, 2014 2:27:29 PM
>>> Subject: Re: [Gluster-devel] Trash translator for GlusterFS
>>>
>>>
>>> Hi Niels,
>>>
>>> The deleted files are moved to trash directory by appending timestamp to
>>> file
>>> name.So there is no chance of name collisions.The trash directory will be
>>> present in every brick.
>>>
>>> In the given scenario,both README files are shown separately  in trash
>>> directory with different timestamps.
>>
>> What if two files are created, deleted with in a second? What if the file is
>> already 254 characters? (maximum file name length is 255 I think)
>>
>>
>>>>>> A HASH of the 'metadata' of the original file appended to the name of
>>>>>> the file to be deleted would help differentiate ?
>
> It still does not help in file-names of length 254.
>

Another possibility is to always truncate the file name to NAME_MAX 
minus the size required for timestamp. The real file name could be 
stored as an extended attribute of the file in trash.

Yet another possibility would be to have the files stored as gfid with 
the timestamp appended. A separate external mapping mechanism (files, 
sqlite, mongo) could provide gfid ==> file name translations.


-Vijay