On 24 May 2011 16:23, Ewan MacMahon <[log in to unmask]> wrote:
>> -----Original Message-----
>> From: GRIDPP2: Deployment and support of SRM and local storage management
>> [mailto:[log in to unmask]] On Behalf Of Sam Skipsey
>> Sent: 24 May 2011 15:22
>>
>> (I'd RDONLY the fuller filesystems now in any case, to stop them
>> getting worse before you drain them).
>>
> Do bear in mind though that making filesystems read-only reduces
> the free space available on the DPM (still by more than it should,
> I think), and if the free space as reported by dpm-listspaces
> falls to zero then you'll fail SAM/Nagios tests.
>
That was what I thought - except that we added some of our storage
back into the DPM as RDONLY for testing purposes recently, and it
seems to be less problematic now.
> If you've got spacetokens occupying most of the DPM (like we do),
> you may need to shrink them.
>
This *is* a valid concern, however.
And this kind of horrible complication is why I was holding out for a
proper targeted dpm-replicate command in the API so I could make a
proper rebalancing script (the existing one basically just juggles
RDONLY and other filesystem statuses while doing drains of the fullest
filesystems in the system).
Sam
> Ewan
>
|