On 03/04/2016 12:50 AM, Richard Yao wrote:
> On 03/02/2016 12:31 PM, Matthew Ahrens wrote:
>> On Wed, Mar 2, 2016 at 12:37 AM, Erik Sørnes <[email protected]> wrote:
>>
>>> Hi!
>>>
>>> Are there any plans on implementing functionality to remove an entire vdev
>>> from a pool ?
>>>
>>> One could write zpool remove <zpoolname> vdev <vdev-name>, and it would
>>> move the data on <vdev-name> to other vdevs in the pool and then remove the
>>> vdev <vdev-name> entirely.
>>> This is very usefull for many use cases. I've googled a lot, and haven't
>>> found any information on this, apart from a very old one from SUN.
>>>
>>
>> Yes, we (Delphix) have implemented this and it is used in production.  It
>> is not yet upstreamed.  For more info, see our talk at the 2014 OpenZFS
>> Developer Summit:
>>
>> slides:
>> http://open-zfs.org/w/images/b/b4/Device_Removal-Alex_Reece_%26_Matt_Ahrens.pdf
>> video:
>> http://www.youtube.com/watch?v=Xs6MsJ9kKKE&list=PLaUVvul17xSdOhJ-wDugoCAIPJZHloVoq&index=12
>>
>> Here is the first commit for Device Removal from the Delphix repo (there
>> are several follow on bug fixes and additional features):
>> https://github.com/delphix/delphix-os/commit/db775effdda128d8e14216abde06281513b03c37
>>
>> We haven't yet upstreamed this because it only supports removing
>> non-redundant top-level vdevs (i.e. you can't remove a mirror or RAID-Z
>> vdev).  Additional work is needed to make sure that we don't lose any good
>> copies of the data, and to ensure proper raid-z allocation alignment.  We
>> would welcome anyone who's interested in implementing these.
>>
>> We'd also appreciate opinions of "Please upstream even without
>> mirror/RAID-Z support" or "Please don't integrate without mirror/RAID-Z
>> support - all ZFS features should work together."
> 
> My vote is for "Please don't integrate without mirror/RAID-Z
> support - all ZFS features should work together."

I probably should elaborate on this. ZFS is intended to make
administration simple. If we start introducing features that are
unnecessarily incompatible with existing features, we risk a future
where the non-interoperability of various features makes OpenZFS
administration into a NP-complete boolean satisfiability problem.

In addition, some ideas for features are far more incompatible with
other things (e.g. killing checksums, redundancy, compression, etcetera
for O_DIRECT support) have been discussed in private. If this is merged,
the idea that "all ZFS features should work together" is not going to be
a great argument against implementing even less compatible features.

As useful as this is, I would rather that we not take the first step
toward either scenario.

> 
>>
>> --matt
>>
>>
>>>
>>> kind regards
>>>
>>> Erik Sørnes, IT-support, Nilu
>>> P Please consider the environment before printing this email and
>>> attachments
>>>
>> 
>> 

Attachment: signature.asc
Description: OpenPGP digital signature




-------------------------------------------
openzfs-developer
Archives: https://www.listbox.com/member/archive/274414/=now
RSS Feed: https://www.listbox.com/member/archive/rss/274414/28015062-cce53afa
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=28015062&id_secret=28015062-f966d51c
Powered by Listbox: http://www.listbox.com

Reply via email to