[lvm-devel] Batch/queue LVM operations

[Zdenek Kabelac]

Dne 1.8.2017 v 19:24 Eric Wheeler napsal(a):
> On Fri, 28 Jul 2017, Zdenek Kabelac wrote:
> 
>> Dne 28.7.2017 v 21:22 Eric Wheeler napsal(a):
>>> Hello,
>>>
>>> Is there an option to batch LVM operations?
>>>
>>> For example, I would like to delete 100 thin snapshots without updating
>>> the vgmeta 100 times.
>>
>> You could possibly use --select feature to handle all removals with
>> just one lvremove command.
> 
> I tried --select, but there is a long pause between removals so it is
> clearly doing a metadata operation for each removal.
> 
> Is it possible to do something like this:
> 
> 1. Hold a lock (how?)
> 2. vgcfgbackup -f /my-backup/file --ignorelocking-failure
> 3. Modify /my-backup/file (remove LVs)
> 4. dmsetup message pool 0 delete ID (each ID)
> 5. vgcfgrestore --ignorelocking-failure
> 6. Release lock
> 
> --
> Eric Wheeler
> 

Hi

Unfortunately this optimization is currently not possible.

lvm2 is strictly working on 1-by-1 logic - since resolving more complex
recovery path is beyond capabilities of this tool.

We do plan to 'group' 'lvremove' operation in future -
so i.e. if you pass multiple  LVs on cmdline -
sort them by  VG and do a single VG commit to remove them all.

But with thinLV this is different level as we would need to also
join all delete thin-pool transactions together - and we do not have good 
interface for this to resolve 'what have been delete and what's been left in 
case of error' so there is some space for improvements ATM - but I could see 
some possibilities, but it's still in queue behind bigger fishes for hunting...

What we would could possibly improve more easily is to remove ALL thin LVs
from a pool - such 'pool' reset could be implemented in more simple way.

So far the remove or creation of large numbers of LVs was not seen
as time-critical operation - so we are rather focused on simpler code here.


Regards

Zdenek