[mythtv-users] Disk space is not freed from deleted recording until mythbackend exits

[Jim O'Neill]

Michael T. Dean wrote:
> On 08/15/2006 10:06 PM, Jim O'Neill wrote:
>
>   
>> I've been having a problem for about two to three weeks now where the 
>> space from a recording manually deleted from within mythfrontend is not 
>> returned to the system until i exit mythbackend.  It was working prior 
>> to this.   The recording appears to be deleted according to mythfrontend 
>> and the file system but myth and df do not show the space being 
>> returned.  The file system is used by mythtv alone and only for 
>> recordings.  As a test i have deleted a recording from konqueror and 
>> space is returned after a few moments without exiting konqueror but i 
>> have no clue as to whether this test has any validity.
>>
>> I've found one thread on the list mentioning similar symptoms but in 
>> that case having mythbackend and mythfrontend running on different boxes 
>> was determined to be a contributing factor and i'm running both on the 
>> same box.
>>
>> I'd appreciate any insight anyone can share that may lead to a solution.
>>
>>     
>
> In SVN trunk, due to the new feature that performs deletes in chunks (to 
> prevent I/O starvation), you won't see the disk space show up for quite 
> some time--unless you stop mythbackend (which, BTW, should result in a 
> flurry of I/O activity as the delete is completed in one chunk on 
> killing mythbackend).  So, if you're using SVN trunk, delete a video and 
> check your disk space in a few hours and think of all the benefits 
> you'll reap from the quicker response times in the GUI after deletion, 
> the lack of prebuffering pauses due to deletion/autoexpiration, and the 
> lack of recording errors/artifacts from I/O starvation during recording.
>
> Mike
>   
Mike, i think you may be right.  I am running SVN trunk and was aware of 
this new method for deleting files but i wasn't aware that it is as slow 
as it is and therefore thought that space wasn't being returned. 

I just created a script to execute df every 60 seconds and ran this for 
about 20 minutes between recordings.  I'm seeing 500 KB +/- a few bytes 
being deleted every minute.  If my math is correct this works out to 16 
hours to completely delete a 500 MB recording.  For the test i deleted 
about 4 GB of recording which represents over 5 days before space is 
returned. Does this sound right or do i have a problem somewhere else?  
If it is correct, any chance of changes to allow the chunk size to be 
configurable or to be determined heuristically.  My hardware 
configuration isn't particularly low-end so performance wouldn't be the 
concern it may for others so larger chunks would not unacceptably impact 
performance.

I appreciate your assistance,

Jim