[linux-lvm] lvreduce thin-pool

[linux-lvm] lvreduce thin-pool

[Paul Dann]

[-- Attachment #1: Type: text/plain, Size: 741 bytes --]

Hi there,

I've been following Ede Wolf's thread with interest, but it hasn't quite
covered my query, and I've not been able to find much about this in the
archives:

Are there any plans to implement reducing of thin pools? I've been bitten
several times by carelessly allowing a thin pool to expand onto unsuitable
storage and subsequently being unable to reduce the pool again to ensure IO
performance of the pool. (I know that better forward planning can mitigate
this.)

Mainly, I'm wondering if there are technical reasons this hasn't been
implemented yet? On the face of it, I'd have expected this to be a
relatively straight-forward procedure? But I'd be interested in why this
might not be the case.

Cheers,
Paul

[-- Attachment #2: Type: text/html, Size: 905 bytes --]

Re: [linux-lvm] lvreduce thin-pool

[Zdenek Kabelac]

Dne 14. 01. 20 v 10:52 Paul Dann napsal(a):
> Hi there,
> 
> I've been following Ede Wolf's thread with interest, but it hasn't quite 
> covered my query, and I've not been able to find much about this in the archives:
> 
> Are there any plans to implement reducing of thin pools? I've been bitten 
> several times by carelessly allowing a thin pool to expand onto unsuitable 
> storage and subsequently being unable to reduce the pool again to ensure IO 
> performance of the pool. (I know that better forward planning can mitigate this.)
> 
> Mainly, I'm wondering if there are technical reasons this hasn't been 
> implemented yet? On the face of it, I'd have expected this to be a relatively 
> straight-forward procedure? But I'd be interested in why this might not be the 
> case.

Hi

There is no support for 'online' reduction of thin-pool. Mainly because there 
is not much interest into this.

For 'offline' reduction - you could probably write a tool working like this:

In kernel metadata figure out who holds the last 'chunk' and find what is the 
first 'empty' chunk - and copy the chunk there and remap all the references
to this last chunk to the new position.

Quite simple isn't  :) - the task would be probably simpler to code if it 
would be directly using thin tools from d-m-p-d package (i.e. thin_repair 
background)

So basically a straight forward process, someone just needs to do all the 
coding for this.

Regards

Zdenek