[SOLVED] LVM `issue_discards` is disabled after upgrade from 7 to 8.

J

jamesharr

Member
Proxmox Subscriber
Jan 8, 2022
32
4
13
Hi,

I'm not sure if this is an "issue" per-say, but I noticed that during an upgrade the `issue_discards = 1` config line is removed from lvm.conf and the new default is `issue_discards = 0`. I also could not find the setting anywhere else in /etc/lvm, for example in lvmlocal.conf.

Is this an issue and/or intended? In most clusters I don't imagine it'll cause an immediate issue, but it could cause an issue long-term on SSDs.

These questions are also likely relevant:
1) has anyone else has encountered this? It's possible this is unique to me.
2) Is this file the same on a fresh 8.x install? Or is this unique to upgrades? -- If someone has a fresh install to look at it might save some work, otherwise I might try a fresh install later
3) is there something else going on that might make this a non-issue? (ie: LVM behavior change that's not documented in lvm.conf)

Config file diff during `dist-upgrade`:

Diff: 
        # Configuration option devices/issue_discards.
        # Issue discards to PVs that are no longer used by an LV.
@@ -286,7 +379,8 @@
        # benefit from discards, but SSDs and thinly provisioned LUNs
        # generally do. If enabled, discards will only be issued if both the
        # storage and kernel provide support.
-       issue_discards = 1
+       # This configuration option has an automatic default value.
+       # issue_discards = 0

I've attached full config files for reference:
- lvm.conf - LVM config installed as part of the upgrade (I opted to install the package maintainers version)
- lvm.conf.bak - LVM Config prior to the upgrade.
 
did you by chance get prompted about config file changes and selected to use the maintainer's version?
 
I had a chance to do a fresh Proxmox-VE install (inside a VM if that makes any difference). The defaults are to leave issue_discards set to 0 (off). The install was using proxmox-ve-8.0-2.iso.

Bash: 
root@pve-test:~# lvmconfig --typeconfig full | grep issue_discards
        issue_discards=0
       
root@pve-test:~# grep issue_discards /etc/lvm/lvm.conf
        # Configuration option devices/issue_discards.
        # issue_discards = 0
 
You know... When I installed 7.4 (from proxmox-ve_7.4-1.iso), the defaults are to set issue_discards to 0 as well. This system was originally a 5.x install and I've done several upgrades so I'm wondering if it was a setting from a long time ago. I don't believe it's anything I tuned myself.
 
yes, it defaults to 0. I am not sure whether the Debian package had a different default at some point in the past, that would be possible. for PVE it doesn't make much difference - the option is for discarding segments when a regular LV is removed (and thus, space on a PV becomes "free" again). this rarely happens in PVE, unless you are using regular LVM and not LVM thin.
 
  • Like
Reactions: Dmitrius7
For completeness, a fresh install from the 6.4-1 ISO (which installed the proxmox-ve:6.3-1 package) had issue_discards set to 1.

Bash: 
root@pve-test:~# lvmconfig --typeconfig full | grep issue_discards

        issue_discards=1

root@pve-test:~# grep issue_discards /etc/lvm/lvm.conf

        # Configuration option devices/issue_discards.

        issue_discards = 1
 
  • Like
Reactions: fabian
Egbahan koissi said:
I have the same problem. what solution did you find please
Click to expand...
I determined that it wasn't necessary to set this value. Most people use LVM thin provisioning.

To quote @fabian,
for PVE it doesn't make much difference - the option is for discarding segments when a regular LV is removed (and thus, space on a PV becomes "free" again). this rarely happens in PVE, unless you are using regular LVM and not LVM thin.
Click to expand...

If you do use LVM thin provisioning, you can always edit /etc/lvm/lvm.conf to insert this line:

Code: 
issue_discards = 1
 
The lvm volume is displayed with a question mark after updating and the output of the pvdisplay command is:

root@pve2:~# pvdisplay
--- Physical volume ---
PV Name /dev/sda3
VG Name pve
PV Size 278.86 GiB / not usable <1.28 MiB
Allocatable yes
PE Size 4.00 MiB
Total PE 71389
Free PE 4095
Allocated PE 67294
PV UUID L1v4sc-a1Bk-NNoI-KFb7-98jl-aB4P-HUHhAp

"/dev/sdb" is a new physical volume of "44.88 TiB"
--- NEW Physical volume ---
PV Name /dev/sdb
VG Name
PV Size 44.88 TiB
Allocatable NO
PE Size 0
Total PE 0
Free PE 0
Allocated PE 0
PV UUID eQyJrR-qRck-55nT-FC3i-Thd9-M3au-amt55M
 
What "pvdisplay" shows is a physical volumes and doesn't show whether or not you're using volume thin provisioning.

What does "lvdisplay" show? This will show if you're using lvm thin provisioning or not
 
surely I have the wrong discussion thread but my problem is that all my vms are on an LVM volume which is connected to an ISCI, after the upgrade
at level 7 to 8, my LVM has marked a question mark and it is no longer accessible, my VMs no longer start either, here is the error message when I try to start the VM:

-------------------------------------
Command failed with status code 5.
command '/sbin/vgscan --ignorelockingfailure --mknodes' failed: exit code 5
Volume group "bnetd-stockage" not found
TASK ERROR: can't activate LV '/dev/bnetd-stockage/vm-250-disk-0': Cannot process volume group bnetd-stockage

-------------------------------

I have nearly 110 VMs running in my environment and their mounted on this volume
 
create storage failed: command '/sbin/pvs --separator : --noheadings --units k --unbuffered --nosuffix --options pv_name,pv_size,vg_name,pv_uuid /dev/disk/by-id/scsi-36006016024984e00cf23665e71e47afa' failed: exit code 5 (500)

when i want to create new LVM
 
fabian said:
yes, it defaults to 0. I am not sure whether the Debian package had a different default at some point in the past, that would be possible. for PVE it doesn't make much difference - the option is for discarding segments when a regular LV is removed (and thus, space on a PV becomes "free" again). this rarely happens in PVE, unless you are using regular LVM and not LVM thin.
Click to expand...
Just curious... Why isn't it necessary to set issue_discards=1 when used with PVE and lvm thin provisioning? Surely this is still needed when a thin volume is deleted (as there's always stuff left lying around in the volume etc)?
 
Last edited:
issue_discards just controls whether the PV extents are discarded when an LV is removed. with a thin pool, the full size of the pool itself is allocated on the VG/PV and belongs to the pool, and removing a thin LV within the pool has no effect outside of the pool, no physical space becomes free as a result, so no discard can happen.

for thin pool there is a separate mechanism for preventing data leakage of previously used extents - since all the extents belong to the thin pool, whenever an extent is first assigned for usage by a thin LV, it is written over with zeroes. you can see this in effect when doing a live migration with a big thin volume - there will be a pretty hefty I/O spike at the start of the storage migration when the full image is overwritten on the target side. this behaviour is controlled by "thin_pool_zero" in /etc/lvm/lvm.conf. additionally, you can also control what happens with discard/trim on a thin LV (e.g., like a guest OS might issue on a guest volume) - this is "thin_pool_discards". the default here is to mark discarded extents for re-use in the pool *and* also issue a discard on the underlying device.

"man lvmthin" contains a lot more information about thin pools and the various tuning options. just be careful - it's easy to misconfigure a pool and lose all your data!
 
  • Like
Reactions: chrcoluk, Dmitrius7 and Tony C
You must log in or register to reply here.
Top Bottom