[linux-lvm] Converting existing thick LVs into thin LVs possible?

Discussion:

Jeremy Smith

2014-02-14 19:21:40 UTC

I have a system currently using LVM2 that was originally set up on
RHEL5 (where thin provisioning was not supported).
This system would greatly benefit from the newer thin provisioning
features, and so I am investigating upgrading to RHEL6 so that I may
take advantage. My question is this: Can I convert existing LVM2
"thick" logical volumes into thinly-provisioned volumes on the fly?

This page:
https://access.redhat.com/site/documentation/en-US/Red_Hat_Enterprise_Linux/6/html/Logical_Volume_Manager_Administration/thinly_provisioned_volume_creation.html

lead me to believe that it is possible:
"You can use the --thinpool parameter of the lvconvert command to
convert an existing logical volume to a thin volume."

with the following:

lvconvert --thinpool vg001/lv1 --poolmetadata vg001/lv2

However, I seem to not understand how that works.

Zdenek Kabelac

2014-02-17 09:00:26 UTC

Permalink

Post by Jeremy Smith
I have a system currently using LVM2 that was originally set up on
RHEL5 (where thin provisioning was not supported).
This system would greatly benefit from the newer thin provisioning
features, and so I am investigating upgrading to RHEL6 so that I may
take advantage. My question is this: Can I convert existing LVM2
"thick" logical volumes into thinly-provisioned volumes on the fly?
https://access.redhat.com/site/documentation/en-US/Red_Hat_Enterprise_Linux/6/html/Logical_Volume_Manager_Administration/thinly_provisioned_volume_creation.html
"You can use the --thinpool parameter of the lvconvert command to
convert an existing logical volume to a thin volume."
lvconvert --thinpool vg001/lv1 --poolmetadata vg001/lv2

Sounds like misunderstanding which needs fix/enhancement of this doc page -
we will need to enhance lvconvert command to write big warning about this
operation.

This conversion is converting volume to 'data' volume - but it's not about
preserving content of LV1 as LV1 - it will serve as en empty pool.

The primary purpose is - you can build 'any' type of LV for data volume
and 'any' type of LV for metadata volume (i.e. special layout for raid/mirror)
and then you 'create' for these 2 volumes your thin pool - since normally
you can't create mirrored data & metadata volumes and thin pool in 1 command,
and you can't easily select where the metadata should be placed
(i.e. data on spindle, metadata on SSD).

Post by Jeremy Smith
Am I doing something wrong here?
Or am I going to have to create a new thin pool, new thin logical
volumes, and copy the data over?

There is not yet support for direct conversion of LV to thin-LV.
The only currently supported method is to use your thick-LV as an external
origin LV for your thin-LV volume.

i.e. you create your thin-pool LV + and then thin-LV volume with external
origin that will use original thick-LV (external origin need to be read-only
volume)

Here are the conversion steps:

lvcreate -Lsize_of_data_pool -T vg/pool
lvconvert -T vg/thick_LV --originname new_thick_LV_name --thinpool vg/pool

Disadvantage here is you cannot merge changes from thin volume back to
original thick volume. You can however remove thin-LV and use again your
thick-LV as normal volume - i.e. thin-LV is basically snapshot.

Now hint for recovery - if you have not yet written anything to thin-LV from
newly create thin-pool. You could restore back to your previous configuration
via vgcfgrestore --force.

Zdenek

Jeremy Smith

2014-04-04 16:50:16 UTC

Permalink

Post by Zdenek Kabelac

Sounds like misunderstanding which needs fix/enhancement of this doc page -
we will need to enhance lvconvert command to write big warning about this
operation.
This conversion is converting volume to 'data' volume - but it's not about
preserving content of LV1 as LV1 - it will serve as en empty pool.
The primary purpose is - you can build 'any' type of LV for data volume
and 'any' type of LV for metadata volume (i.e. special layout for raid/mirror)
and then you 'create' for these 2 volumes your thin pool - since normally
you can't create mirrored data & metadata volumes and thin pool in 1 command,
and you can't easily select where the metadata should be placed
(i.e. data on spindle, metadata on SSD).

Post by Jeremy Smith
Am I doing something wrong here?
Or am I going to have to create a new thin pool, new thin logical
volumes, and copy the data over?

There is not yet support for direct conversion of LV to thin-LV.
The only currently supported method is to use your thick-LV as an external
origin LV for your thin-LV volume.
i.e. you create your thin-pool LV + and then thin-LV volume with external
origin that will use original thick-LV (external origin need to be
read-only volume)
lvcreate -Lsize_of_data_pool -T vg/pool
lvconvert -T vg/thick_LV --originname new_thick_LV_name --thinpool vg/pool
Disadvantage here is you cannot merge changes from thin volume back to
original thick volume. You can however remove thin-LV and use again your
thick-LV as normal volume - i.e. thin-LV is basically snapshot.
Now hint for recovery - if you have not yet written anything to thin-LV from
newly create thin-pool. You could restore back to your previous configuration
via vgcfgrestore --force.
Zdenek

-- does not support live
resizing of the pool metadata.
You mention in that talk that there is a manual offline pool metadata
resize method that can be done on kernels prior to 3.14. I was not
able to find details anywhere on how to accomplish this. Could you
point me in the right direction on how to fix a thin pool (and its
thin logical volumes) that has reached the metadata size limit? It
would be greatly appreciated.

Regards,
--
Jeremy Smith
Systems Administrator
eSilo, LLC.
1530 Cypress Drive, Suite H
Jupiter, FL. 33469
561-747-6880 x 412 (Phone)
800-747-9255 x 412 (Toll Free)
866-598-1358 (After Hours)
561-747-6488 (Fax)
***@esilo.com

Zdenek Kabelac

2014-04-14 08:01:00 UTC

Permalink

Post by Jeremy Smith

Post by Zdenek Kabelac

Sounds like misunderstanding which needs fix/enhancement of this doc page -
we will need to enhance lvconvert command to write big warning about this
operation.
This conversion is converting volume to 'data' volume - but it's not about
preserving content of LV1 as LV1 - it will serve as en empty pool.
The primary purpose is - you can build 'any' type of LV for data volume
and 'any' type of LV for metadata volume (i.e. special layout for raid/mirror)
and then you 'create' for these 2 volumes your thin pool - since normally
you can't create mirrored data & metadata volumes and thin pool in 1 command,
and you can't easily select where the metadata should be placed
(i.e. data on spindle, metadata on SSD).

Post by Jeremy Smith
Am I doing something wrong here?
Or am I going to have to create a new thin pool, new thin logical
volumes, and copy the data over?

There is not yet support for direct conversion of LV to thin-LV.
The only currently supported method is to use your thick-LV as an external
origin LV for your thin-LV volume.
i.e. you create your thin-pool LV + and then thin-LV volume with external
origin that will use original thick-LV (external origin need to be
read-only volume)
lvcreate -Lsize_of_data_pool -T vg/pool
lvconvert -T vg/thick_LV --originname new_thick_LV_name --thinpool vg/pool
Disadvantage here is you cannot merge changes from thin volume back to
original thick volume. You can however remove thin-LV and use again your
thick-LV as normal volume - i.e. thin-LV is basically snapshot.
Now hint for recovery - if you have not yet written anything to thin-LV from
newly create thin-pool. You could restore back to your previous configuration
via vgcfgrestore --force.
Zdenek

Zdenek,
Thank you. I was able to perform migration of some of my LVs, but now
I have ran into a problem: the thin pool metadata has ran out of space
:-(
I am running LVM 2.02.98(2) (2012-10-15) on kernel 3.11.0-18-generic
(Ubuntu server 13.10), which -- as I understand from your talk at
http://youtu.be/dGax09aw7WI -- does not support live
resizing of the pool metadata.
You mention in that talk that there is a manual offline pool metadata
resize method that can be done on kernels prior to 3.14. I was not
able to find details anywhere on how to accomplish this. Could you
point me in the right direction on how to fix a thin pool (and its
thin logical volumes) that has reached the metadata size limit? It
would be greatly appreciated.

Offline resize is basically metadata recovery into bigger LV.
So step for resize apply in the same way as for recovery.

1. create temp LV
2. replace temp LV with metadata pool LV
3. now you can activate your metadata LV as regular LV
4. create bigger LV for restore
5. thin_restore -i old_metadata_lv_path -o new_metadata_lv_path
6. swap new_metadata_lv into your pool
7. activate pool

yep - about the time to integrate this offline resize into lvm2 I guess... :)

Zdenek

Zdenek Kabelac

2014-04-14 08:02:53 UTC

Permalink

Post by Zdenek Kabelac

Post by Jeremy Smith

Post by Zdenek Kabelac

Sounds like misunderstanding which needs fix/enhancement of this doc page -
we will need to enhance lvconvert command to write big warning about this
operation.
This conversion is converting volume to 'data' volume - but it's not about
preserving content of LV1 as LV1 - it will serve as en empty pool.
The primary purpose is - you can build 'any' type of LV for data volume
and 'any' type of LV for metadata volume (i.e. special layout for raid/mirror)
and then you 'create' for these 2 volumes your thin pool - since normally
you can't create mirrored data & metadata volumes and thin pool in 1 command,
and you can't easily select where the metadata should be placed
(i.e. data on spindle, metadata on SSD).

Post by Jeremy Smith
Am I doing something wrong here?
Or am I going to have to create a new thin pool, new thin logical
volumes, and copy the data over?

There is not yet support for direct conversion of LV to thin-LV.
The only currently supported method is to use your thick-LV as an external
origin LV for your thin-LV volume.
i.e. you create your thin-pool LV + and then thin-LV volume with external
origin that will use original thick-LV (external origin need to be
read-only volume)
lvcreate -Lsize_of_data_pool -T vg/pool
lvconvert -T vg/thick_LV --originname new_thick_LV_name --thinpool vg/pool
Disadvantage here is you cannot merge changes from thin volume back to
original thick volume. You can however remove thin-LV and use again your
thick-LV as normal volume - i.e. thin-LV is basically snapshot.
Now hint for recovery - if you have not yet written anything to thin-LV from
newly create thin-pool. You could restore back to your previous configuration
via vgcfgrestore --force.
Zdenek

oops hit send button to early...
of course thin_repair

Post by Zdenek Kabelac
6. swap new_metadata_lv into your pool
7. activate pool
yep - about the time to integrate this offline resize into lvm2 I guess... :)
Zdenek