[heketi-devel] Should 'heketi-cli device add ...' wipe devices?

[Jose A. Rivera]

Since I last write in this thread, I had an interesting conversation
with Dustin Black about an issue he ran into. He hit an edge case
where he was creating LVM partitions, wiping the drives, then
recreating the same partitions, and the LVM metadata from the previous
partitions was still there and caused errors with udev. Now this was
rather specific to his case since he was retrying deployments of RHSO
while reusing the previous configuration. His solution ended up being
to write a few MB of zeros at the predicted offset of the partitions
he would be using.

I don't think we'll run into this scenario, but putting this out there
so we have it in our minds in case someone else does. Like, say,
upstream. :)

--Jose

On Tue, May 15, 2018 at 6:24 AM, Michael Adam <obnox at samba.org> wrote:
> On 2018-05-15 at 12:14 +0200, Niels de Vos wrote:
>> On Wed, May 09, 2018 at 01:44:06PM -0500, Jose A. Rivera wrote:
>> > I would very much prefer not making it the default and requiring a
>> > --force flag. But I would want the feature in general, since I already
>> > effectively have such a flag in openshift-ansible.
>>
>> Thanks Jose!
>>
>> It seems that there is an agreement that a --force option would be
>> useful. When passing this option, the data on the device will be
>> destroyed with 'wipefs --all'.
>
> Yes, please require an option!
>
>> Most likely the option will be called --destroy-existing-data so that
>> users are warned about the potential danger of using it.
>
> Might be a little clumsy but very explicit.
> Something like "--wipe-data" might work as well.
>
>> I'll have a look at implementing this.
>
> Great!
>
> Michael
>
>> Niels
>>
>> >
>> > --Jose
>> >
>> > On Wed, May 9, 2018 at 8:30 AM, Niels de Vos <ndevos at redhat.com> wrote:
>> > > Hi,
>> > >
>> > > In https://bugzilla.redhat.com/1433614 there is the suggestion that upon
>> > > tolopogy load the devices of the nodes should get wiped and not abort if
>> > > 'pvcreate' detects existing data. I am wondering if this is a good idea
>> > > to be the default.
>> > >
>> > > My preference would be to add a --force flag to the 'device add ...'
>> > > command, and a matching key/value in the topology.json when 'topology
>> > > load' is run.
>> > >
>> > > Any suggestions or preferences from others? Comments in the bug or a
>> > > reply to this email are most welcome.
>> > >
>> > > Thanks,
>> > > Niels
>> > > _______________________________________________
>> > > heketi-devel mailing list
>> > > heketi-devel at gluster.org
>> > > http://lists.gluster.org/mailman/listinfo/heketi-devel
>> _______________________________________________
>> heketi-devel mailing list
>> heketi-devel at gluster.org
>> http://lists.gluster.org/mailman/listinfo/heketi-devel