[linux-lvm] thin discards

[linux-lvm] thin discards

[Raymond Jennings]

[-- Attachment #1: Type: text/plain, Size: 123 bytes --]

I tried to run an fstrim on a mounted thin volume, but got FITRIM:
 operation not supported.

Did I screw up or something?

[-- Attachment #2: Type: text/html, Size: 170 bytes --]

Re: [linux-lvm] thin discards

[Zdenek Kabelac]

Dne 15.7.2013 17:32, Raymond Jennings napsal(a):
> I tried to run an fstrim on a mounted thin volume, but got FITRIM:  operation
> not supported.
>
> Did I screw up or something?


Is the pool device capable of TRIM support ?

Which version of lvm2 and kernel is in use ?


Zdenek

Re: [linux-lvm] thin discards

[Raymond Jennings]

[-- Attachment #1: Type: text/plain, Size: 779 bytes --]

Gentoo

sys-kernel/gentoo-sources-3.10.1
sys-fs/lvm2-2.02.97-r1

I don't know if the pool device is capable of trim support.  I do know that
the underlying device, a seagate hard drive, does not support discard.
 However one would think that thin volumes have their own discard handling
to deallocate internally, possibly passing a discard down to the underlying
block device as needed.


On Tue, Jul 16, 2013 at 3:16 AM, Zdenek Kabelac <zkabelac@redhat.com> wrote:

> Dne 15.7.2013 17:32, Raymond Jennings napsal(a):
>
>  I tried to run an fstrim on a mounted thin volume, but got FITRIM:
>>  operation
>> not supported.
>>
>> Did I screw up or something?
>>
>
>
> Is the pool device capable of TRIM support ?
>
> Which version of lvm2 and kernel is in use ?
>
>
> Zdenek
>
>
>

[-- Attachment #2: Type: text/html, Size: 1390 bytes --]

Re: [linux-lvm] thin discards

[Raymond Jennings]

[-- Attachment #1: Type: text/plain, Size: 1035 bytes --]

So can thin volumes support discards independently of the device underneath
them?

I expected fstrim to free space in the thin pool.


On Tue, Jul 16, 2013 at 4:33 AM, Raymond Jennings <shentino@gmail.com>wrote:

> Gentoo
>
> sys-kernel/gentoo-sources-3.10.1
> sys-fs/lvm2-2.02.97-r1
>
> I don't know if the pool device is capable of trim support.  I do know
> that the underlying device, a seagate hard drive, does not support discard.
>  However one would think that thin volumes have their own discard handling
> to deallocate internally, possibly passing a discard down to the underlying
> block device as needed.
>
>
> On Tue, Jul 16, 2013 at 3:16 AM, Zdenek Kabelac <zkabelac@redhat.com>wrote:
>
>> Dne 15.7.2013 17:32, Raymond Jennings napsal(a):
>>
>>  I tried to run an fstrim on a mounted thin volume, but got FITRIM:
>>>  operation
>>> not supported.
>>>
>>> Did I screw up or something?
>>>
>>
>>
>> Is the pool device capable of TRIM support ?
>>
>> Which version of lvm2 and kernel is in use ?
>>
>>
>> Zdenek
>>
>>
>>
>

[-- Attachment #2: Type: text/html, Size: 1958 bytes --]

Re: [linux-lvm] thin discards

[Zdenek Kabelac]

Dne 21.7.2013 09:10, Raymond Jennings napsal(a):
> So can thin volumes support discards independently of the device underneath them?
>
> I expected fstrim to free space in the thin pool.

Yes - you are expecting it correctly.

Ignore is mostly usable in case you do not want to free already provisioned 
space - which might be in certain cases better for performance.

But you need to use recent enough kernels and lvm2 tools - since discards 
support required kernel updates.

Upstream has this:

passdown   TRIM ->  thin-vol -> thin-pool -> PV
(if the PV doesn't support discard - you get kernel report error report
and it fallbacks to nopassdown behaviour)

nopassdown TRIM ->  thin-vol -> thin-pool

ignore     TRIM ->  thin-vol



Zdenek

Re: [linux-lvm] thin discards

[Raymond Jennings]

[-- Attachment #1: Type: text/plain, Size: 1006 bytes --]

Do snapshots behave the same way as thin volumes wrt discards?


On Sun, Jul 21, 2013 at 3:47 AM, Zdenek Kabelac <zkabelac@redhat.com> wrote:

> Dne 21.7.2013 09:10, Raymond Jennings napsal(a):
>
>  So can thin volumes support discards independently of the device
>> underneath them?
>>
>> I expected fstrim to free space in the thin pool.
>>
>
> Yes - you are expecting it correctly.
>
> Ignore is mostly usable in case you do not want to free already
> provisioned space - which might be in certain cases better for performance.
>
> But you need to use recent enough kernels and lvm2 tools - since discards
> support required kernel updates.
>

Kernel: 3.10.1-gento
LVM2:  2.02.97

Are these versions recent enough?

Upstream has this:
>
> passdown   TRIM ->  thin-vol -> thin-pool -> PV
> (if the PV doesn't support discard - you get kernel report error report
> and it fallbacks to nopassdown behaviour)
>
> nopassdown TRIM ->  thin-vol -> thin-pool
>
> ignore     TRIM ->  thin-vol
>
>
>
> Zdenek
>
>

[-- Attachment #2: Type: text/html, Size: 1954 bytes --]

Re: [linux-lvm] thin discards

[Zdenek Kabelac]

Dne 21.7.2013 13:01, Raymond Jennings napsal(a):
> Do snapshots behave the same way as thin volumes wrt discards?
>
>
> On Sun, Jul 21, 2013 at 3:47 AM, Zdenek Kabelac <zkabelac@redhat.com
> <mailto:zkabelac@redhat.com>> wrote:
>
>     Dne 21.7.2013 09:10, Raymond Jennings napsal(a):
>
>         So can thin volumes support discards independently of the device
>         underneath them?
>
>         I expected fstrim to free space in the thin pool.
>
>
>     Yes - you are expecting it correctly.
>
>     Ignore is mostly usable in case you do not want to free already
>     provisioned space - which might be in certain cases better for performance.
>
>     But you need to use recent enough kernels and lvm2 tools - since discards
>     support required kernel updates.
>
>
> Kernel: 3.10.1-gento
> LVM2:  2.02.97
>
> Are these versions recent enough?

Nope - 2.02.98 has fixes for discards - it's been just introduced in 2.02.97 
and had some problems - possible those you can observe.

Zdenek

Re: [linux-lvm] thin discards

[Zdenek Kabelac]

Dne 21.7.2013 21:40, Zdenek Kabelac napsal(a):
> Dne 21.7.2013 13:01, Raymond Jennings napsal(a):
>> Do snapshots behave the same way as thin volumes wrt discards?
>>

Yes, snapshots are just like any other thin volumes.

Just a side note - until I think 3.9 kernel there used to be
bug, which had wrong ref-counting of shared blocks.

Zdenek

Re: [linux-lvm] thin discards

[Raymond Jennings]

[-- Attachment #1: Type: text/plain, Size: 548 bytes --]

I tried to run a fstrim on a mounted snapshot (regular snapshot, not thin
snapshot) and got zip.



On Mon, Jul 22, 2013 at 2:47 AM, Zdenek Kabelac <zkabelac@redhat.com> wrote:

> Dne 21.7.2013 21:40, Zdenek Kabelac napsal(a):
>
>  Dne 21.7.2013 13:01, Raymond Jennings napsal(a):
>>
>>> Do snapshots behave the same way as thin volumes wrt discards?
>>>
>>>
> Yes, snapshots are just like any other thin volumes.
>
> Just a side note - until I think 3.9 kernel there used to be
> bug, which had wrong ref-counting of shared blocks.
>
> Zdenek
>
>

[-- Attachment #2: Type: text/html, Size: 1205 bytes --]

Re: [linux-lvm] thin discards

[Zdenek Kabelac]

Dne 22.7.2013 12:12, Raymond Jennings napsal(a):
> I tried to run a fstrim on a mounted snapshot (regular snapshot, not thin
> snapshot) and got zip.
>

I've been talking about  thin snapshots.

Of course - for old non-thin snapshot - TRIM is not supported.
(And it doesn't look like anything simple to implement)

Zdenek


>
>
> On Mon, Jul 22, 2013 at 2:47 AM, Zdenek Kabelac <zkabelac@redhat.com
> <mailto:zkabelac@redhat.com>> wrote:
>
>     Dne 21.7.2013 21:40, Zdenek Kabelac napsal(a):
>
>         Dne 21.7.2013 13:01, Raymond Jennings napsal(a):
>
>             Do snapshots behave the same way as thin volumes wrt discards?
>
>
>     Yes, snapshots are just like any other thin volumes.
>
>     Just a side note - until I think 3.9 kernel there used to be
>     bug, which had wrong ref-counting of shared blocks.
>
>     Zdenek
>
>

Re: [linux-lvm] thin discards

[Raymond Jennings]

[-- Attachment #1: Type: text/plain, Size: 1379 bytes --]

As a pure matter of opinion, I enjoy "fat provisioning" on my filesystems,
since I run a desktop and in theory it really cuts down on physical
fragmentation.

Snapshots are something I use whenever I need to make a fundamental
reconfiguration to the system as a whole that can't be done live, such as
resizing or reconfiguring the root filesystem.

This would preclude the unquestionable usage of thin volumes.



On Mon, Jul 22, 2013 at 3:16 AM, Zdenek Kabelac <zkabelac@redhat.com> wrote:

> Dne 22.7.2013 12:12, Raymond Jennings napsal(a):
>
>  I tried to run a fstrim on a mounted snapshot (regular snapshot, not thin
>> snapshot) and got zip.
>>
>>
> I've been talking about  thin snapshots.
>
> Of course - for old non-thin snapshot - TRIM is not supported.
> (And it doesn't look like anything simple to implement)
>
> Zdenek
>
>
>
>>
>> On Mon, Jul 22, 2013 at 2:47 AM, Zdenek Kabelac <zkabelac@redhat.com
>> <mailto:zkabelac@redhat.com>> wrote:
>>
>>     Dne 21.7.2013 21:40, Zdenek Kabelac napsal(a):
>>
>>         Dne 21.7.2013 13:01, Raymond Jennings napsal(a):
>>
>>             Do snapshots behave the same way as thin volumes wrt discards?
>>
>>
>>     Yes, snapshots are just like any other thin volumes.
>>
>>     Just a side note - until I think 3.9 kernel there used to be
>>     bug, which had wrong ref-counting of shared blocks.
>>
>>     Zdenek
>>
>>
>>
>

[-- Attachment #2: Type: text/html, Size: 2370 bytes --]