CBRC Digest Recompute operation takes a long time to complete for the Horizon linked-clone desktops.

Products

• VMware vSphere ESXi

Issue/Introduction

Symptoms:
This issue found in the following versions.
ESXi 6.7 U1  build 10302608 or later and lower than ESXi 6.7 P01.
ESXi 6.5 P03 build 10884925 or later and lower than ESXi 6.5 P04.

You see high network usage and high storage read latency.

When the desktops are refreshed\recomposed, a recompute operation follows which could take a long time to complete.
              
1. Digest attach failure due to missing "scsix:x.digest = TRUE" entry in the .vmx file.
in VMX logs, the entry should be missing when VM powers on

     scsi0:0.deviceType = "scsi-hardDisk"
     scsi0:0.fileName = "test.vmdk"
     scsi0:0.present = "TRUE"

And due to which attach failure will occur: 
SCSIFilterESXAttachCBRCInt: CBRC not enabled or opened without filters,skipping CBRC           filter attach.

This log should be for the digest enable disk.

2. CID mismatch
You see:  DigestLibFileOpenInt: CID mismatch -> disk=0x3d2bdd0e, header=0x384d3a1a in vmware.log or hostd.log

Environment

VMware vSphere ESXi 6.5
VMware vSphere ESXi 6.7

Cause

Recompose operation leads to miss-configuration of the desktop's digest files.
This results in all the re-compute IO ending up in the replica disk.
This leads to IO congestion in the replica and finally longer recompute times.

Resolution

Fixed in ESXi 6.5 P04 and ESXi 6.7 P01