Open Stack

On 7/15/13 6:52 AM, Leandro Reox wrote:
> Hi guys,
>
> I was wondering if there any plans or maybe already exist a feature or a
> flag for the account reaping process to make it gradually. We had one
> case with an account with millions of files, that was plain deleted, an
> that triggers the deletion of every object, then containers to finally
> delete the account. This process added a LOT of load on the cluster
> causing bad performance issues. Maybe we can add a flag like
> reaping_progress = x% or something like that, maybe that hits recon to
> get the progress and when 5% its done, move on with the next 5% and so
> on. Would be great to have something like this without manual intervention.

There is a concurrency setting (imaginatively named "concurrency") in 
the account-reaper config that controls the maximum number of concurrent 
DELETE requests. Default is 25; sounds like you might want to lower it.

Note that the value should be a perfect square (so 1, 4, 9, etc.); if 
not, the reaper will still work, but the number will effectively be 
rounded up to the next perfect square (so 20 -> 25, for example).