VMware Cloud Community
Mateusz-Konopni
Contributor
Contributor
‎05-23-2018 02:42 AM

VSAN 6.6 - Read Cache Reservation BUG

Hi,

Some info about the environment.

3 x ESXi 6.5u2

VCSA 6.7 (was 6.5u2 but still the same issue)

I'm really puzzled by the problem that I have. I had some storage policies configured with Read Cache reservation which was at 50-60% level most of the time. As it was no longer needed I have removed Read Cache Reservation from those policies and reapplied them. After doing so I got this bug:

vsan.jpg

As you can see VSAN Health Report is now showing some ridiculous number of GB being presumably reserved for Read Cache, it must be bug since my entire VSAN datastore is around 27 TB, 12 TB used and around 22 TB provisioned.

As you can imagine first thing I did was to double check that no VM or policy has got any Read Cache Reservation configured. Then I thought that maybe fact that it went to 0 tipped of the bug so I created small reservation for 10 GB hoping it will fix this but no success.

I also checked all physical SSDs but Reservation size is being reported correctly there as well.

Reason that I need to solve this is that now no SSD cache is being used on this VSAN cluster, as it thinks it all needs to be reserved.

Have you got any idea on how this number from VSAN Health Check is generated and how can I go about fixing that?

0 Kudos
1 Reply
TheBobkin
Champion
Champion
‎05-23-2018 02:01 PM

Hello Mateusz-Konopnicki​

Welcome to Communities! Some useful info on participating here:

https://communities.vmware.com/docs/DOC-12286

Going to assume this is a Hybrid cluster as Read-cache reservation isn't supported on All-Flash.

I saw this once before long long ago but I cannot recall the cause, I will see if I can dig it out tomorrow.

Is this a Horizon View environment?

Please do confirm at the Object level that no Objects have cache reservation applied to them:

# python /usr/lib/vmware/vsan/bin/vsan-health-status.pyc > /tmp/placeholdernamehere.txt

# less /tmp/placeholdernamehere.txt | grep cacheReservation

"I have removed Read Cache Reservation from those policies and reapplied them"

Can you try cloning the current Storage Policy, adding a cache reservation rule, setting it to 0 and re-applying this to the VMs/Objects? I tested this in lab, it should not cause full Object rebuild (as would striping or changing FTM etc.) but do test this with 1-2 VMs before batch applying.

"Reason that I need to solve this is that now no SSD cache is being used on this VSAN cluster"

As in there are currently no IO write operations possible? If this is a production cluster do open a Support Request.

Bob

0 Kudos