Skip to contentSkip to conversation

LVM: Split mirror capability

PATCH In review Jonathan Brassow 3 participants 4 messages Oct 13, 2009 – Oct 26, 2009 Device mapper

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 LVM: Split mirror capability

Adding code to allow users to split legs off of a mirror wasn't all that hard. However, I could use some help regarding the nature of the arguments that should be used to invoke this action.

The following patches introduce a new (horrible) argument '--split' to lvconvert that signals the intent to split and keep a desired leg of a mirror. Example run:

[root@bp-01 ~]# lvcreate -m2 -L 500M -n lv vg
Logical volume "lv" created
[root@bp-01 ~]# lvs -a -o name,copy_percent,devices

  LV            Copy%  Devices
  LogVol00             /dev/sda2(0)
  LogVol01             /dev/sda2(4451)
  lv            100.00 lv_mimage_0(0),lv_mimage_1(0),lv_mimage_2(0)
  [lv_mimage_0]        /dev/sdb1(0)
  [lv_mimage_1]        /dev/sdc1(0)
  [lv_mimage_2]        /dev/sdd1(0)
  [lv_mlog]            /dev/sdi1(0)
[root@bp-01 ~]# lvconvert -m1 --split vg/lv /dev/sdc1
  Logical volume lv converted.
[root@bp-01 ~]# lvs
  LV          VG         Attr   LSize   Origin Snap%  Move Log     Copy%  Convert
  LogVol00    VolGroup00 -wi-ao 139.09g
  LogVol01    VolGroup00 -wi-ao   9.81g
  lv          vg         mwi-a- 500.00m                    lv_mlog 100.00

lv_mimage_1 vg -wi-a- 500.00m

You can see from the above that I went as far as to specify the leg I wanted split off.

Aside from the new '--split' argument not being very good, we also end up with a new logical volume named, 'lv_mimage_1', which also isn't very good. Should I change the name via s/_mimage_1/-copy-%d/, or should I allow the name to be specified (or left alone)?

Here is the current syntax and some others I have thought of:

# New '--split' argument.
# This is suppose to give the user the ability to signify a split

# of the mirror, rather than a reduction/increase in mirror images.
# However, something like '--keep_images' might better signify that
# although we are reducing the number of mirror legs, we want to
# have them presented rather than removed.
prompt> lvconvert -m <n> --split vg/lv <removable device(s)>

# New '--splitmirror <n>' argument
# Replaces '-m <n> --split', and rather than specifying the number

# of legs you want remaining in the mirror when finished, you specify
# the number of legs you want split off.
#
# BTW, you can only split off one leg at a time right now, but I
# don't see a reason why we couldn't split a 4-way mirror into
# 2 2-way mirrors at some point in the future.  (Being able to
# specify the removable devices also allows us to choose every
# other device if we chose to.)
prompt> lvconvert --splitmirror 1 vg/lv <removable device(s)>

Please added better ideas if you have them... I am leaning toward the second example (even though I implemented the 1st in the patches).

Concerning the naming of the newly split off device, we could use the '-n <name>' argument. If unspecified, we would have to choose some default.

brassow

--
lvm-devel mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/lvm-devel

#1 lore raw
Quoted 11 lines · Jonathan Brassow [[email protected]] wrote:

[root@bp-01 ~]# lvconvert -m1 --split vg/lv /dev/sdc1
Logical volume lv converted.
[root@bp-01 ~]# lvs

  LV          VG         Attr   LSize   Origin Snap%  Move Log     Copy%  Convert
  LogVol00    VolGroup00 -wi-ao 139.09g
  LogVol01    VolGroup00 -wi-ao   9.81g
  lv          vg         mwi-a- 500.00m                    lv_mlog 100.00

lv_mimage_1 vg -wi-a- 500.00m

You can see from the above that I went as far as to specify the leg I wanted split off.

How about specifying the hidden volume name of legs rather than PVs? I imagine a leg could be allocated on two different PVs.

Quoted 4 lines

Aside from the new '--split' argument not being very good, we also end up with a new logical volume named, 'lv_mimage_1', which also isn't very good. Should I change the name via s/_mimage_1/-copy-%d/, or should I allow the name to be specified (or left alone)?

We should allow the name to be specified. If not specified, we should do something like: lvname-split-%date%

Quoted 10 lines

# New '--split' argument.
# This is suppose to give the user the ability to signify a split

# of the mirror, rather than a reduction/increase in mirror images.
# However, something like '--keep_images' might better signify that
# although we are reducing the number of mirror legs, we want to
# have them presented rather than removed.
prompt> lvconvert -m <n> --split vg/lv <removable device(s)>

# New '--splitmirror <n>' argument
# Replaces '-m <n> --split', and rather than specifying the number

Quoted 9 lines
# of legs you want remaining in the mirror when finished, you specify
# the number of legs you want split off.
#
# BTW, you can only split off one leg at a time right now, but I
# don't see a reason why we couldn't split a 4-way mirror into
# 2 2-way mirrors at some point in the future.  (Being able to
# specify the removable devices also allows us to choose every
# other device if we chose to.)
prompt> lvconvert --splitmirror 1 vg/lv <removable device(s)>

I like the current syntax if you go with 'lvconvert' command. It already has -m option, and convert is all about the current LV rather than the resulting split LV.

I like #2 better but only if you go with a different command like lvsplit:

promt> lvsplit --splitnum <n> vg/lv <list of hidden lv names>

<n> is the number of mirror legs the user wants to split out and the number of hidden volume names specified can be less than or equal to <n> but can't be more.

Thanks, Malahal.

--
lvm-devel mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/lvm-devel

#2 lore raw

PATCH 1/1 LVM: Split mirror capability

In reply to Jonathan Brassow

Based on the first proposal, things do seem to fit together nice and simply.

Quoted 1 line · Jon Brassow

[root at bp-01 ~]# lvconvert -m1 --split vg/lv /dev/sdc1

(1) One lvconvert cmd can only split a mirror into two pieces.

Starting with a 2-way mirror:

lvconvert -m1 -n lvcopy --retainas copy vg/lv lv_mimage1

Splits off lv_mimage1, promotes it to a visible LV and renames it 'lvcopy'.

Starting with a 3-way mirror, it would split off a 2-way mirror (which you could split further if you wished).

Without -n it would rename it to lv_msplit_N where N is the next unused digit and msplit is *not* a reserved name.

lvconvert -m1 --retainas unsynced vg/lv lv_mimage1

promotes lv_mimage1 to a read-only visible LV & renames it to lv_mimage1_unsynced (still a reserved name - you cannot rename it while in this state). You can now lvremove it, or further use

lvconvert --retainas copy vg/lv_mimage1

to split it properly, rename it and make it writeable (unless the mirror is itself readonly).

Alasdair

#3 lore raw

PATCH 1/1 LVM: Split mirror capability

In reply to Jonathan Brassow

Second proposal.

Quoted 4 lines · Jon Brassow

# New '--splitmirror <n>' argument
# Replaces '-m <n> --split', and rather than specifying the number

# of legs you want remaining in the mirror when finished, you specify
# the number of legs you want split off.

Actually, that's really a general argument alternative to -m. Either say -m for what is needed in the final state, or some new argument for what to change to reach the desired final state, applicable to other lvconvert functionality too.

So we could have both option 1 and option 2...

Alasdair