fstrim not working and passing through the DISCARD command

Software-based VM-centric and flash-friendly VM storage + free version

Moderators: anton (staff), art (staff), Max (staff), Anatoly (staff)

Post Reply
ashley
Posts: 4
Joined: Thu Jul 18, 2019 2:44 pm

Thu Jul 18, 2019 2:59 pm

I'm running on a paid for version of the 4tb isci software and its been rock solid on the whole.
I've not managed to convince the management to pay for the support, but this may well sway it!

We're running starwind VSAN v8.8730 hosting a 2Tb drive for 25 VMs. Each has 80Gb thin provisioned drives and the VM platform is Ovirt 4.1.1
we've only marginally over provisioned the drives assuming that we'd not use all space on the Centos servers we run on it.

But now, after a year they all growing to use the maximum disk space and we are getting very close to the limits.
With our without the Linux "FStrim" service running the VMs wont release the space back to the block storage
Most of the machines only use 4Gb of disk space.
Currently our mitigation is to shut down the VMs and clonezilla them onto a fresh drive and discard the old. This is not sustainable though.

Does the version we have support the pass through of the discard function, which. i believe, is what's required here?

From some of Ovirt's support pages https://ovirt.org/develop/release-manag ... orage.html it mentions;
If all the restrictions are met and Pass Discard is enabled, when a discard command (UNMAP SCSI command) is sent from the guest, qemu will not ignore it and will pass it on to the underlying storage. Then, the storage will set the unused blocks as free so that others can use them, and the reported consumed space of the underlying thinly provisioned lun will reduce.
Do you have any advice for us on this matter
Last edited by ashley on Thu Jul 18, 2019 4:09 pm, edited 1 time in total.
ashley
Posts: 4
Joined: Thu Jul 18, 2019 2:44 pm

Thu Jul 18, 2019 4:08 pm

I can see in this release notes:
Release Notes V8, build is 11404 (07 July 2017)
Cache
Fixed processing of TRIM/UNMAP command. When cache memory allocation was still in progress, but virtual devices have already been set to the functional state, service could crash on receiving TRIM/UNMAP command.
Is this relevant?
Oleg(staff)
Staff
Posts: 568
Joined: Fri Nov 24, 2017 7:52 am

Fri Jul 19, 2019 12:27 pm

Hi Ashley,
StarWind supports TRIM/UNMAP commands.
Please try to upgrade your nodes to the latest StarWind build.
ashley
Posts: 4
Joined: Thu Jul 18, 2019 2:44 pm

Fri Jul 19, 2019 3:20 pm

Oleg thats great news.
Which version had this feature added?
ashley
Posts: 4
Joined: Thu Jul 18, 2019 2:44 pm

Fri Jul 19, 2019 3:46 pm

also, will it pass through the trim if the storage is recognised as non flash drive??

Code: Select all

tcp: [10] 10.10.1.1:3260,1 iqn.2008-08.com.starwindsoftware:xxxx-st2a-target1 (non-flash)
Oleg(staff)
Staff
Posts: 568
Joined: Fri Nov 24, 2017 7:52 am

Wed Jul 24, 2019 7:47 am

Hi Ashley,
You can change disk type in StarWind.
It can be done by adding "SolidStateDrive" parameter to device string in config file StarWind.cfg:
Stop StarWind service on one node;
Start WordPad with administrator privileges, open StarWind.cfg file, find <device... record by value of device name or .swdsk file parameter.
Add SolidStateDrive="1" parameter before closing tag, for example:

<device file="My Computer\D\img\Storage4\Storage4.swdsk" name="imagefile1" SolidStateDrive="1"/>

Save file, start StarWind service, do the same on the second node.
Which version had this feature added?
As I told, StarWind supports TRIM/UNMAP commands.
Post Reply