VMware Cloud Community
jasonh83
Contributor
Contributor

Why did thin-provisioned VMDK suddenly grow, and why doesn't WebUI reflect new size after 'punchzero'?

This is a two part question;

First, I have an ESX 6.5.0 Update 1 (Build 5969303) host, no vCenter, with a VM containing two thin-provisioned VMDKs (150GB and 5TB) on a local storage (SSD RAID array) datastore, both of which ballooned up to 100% of their provisioned capacity. I shrunk them back down with vmkfstools --punchzero and they are now consuming 2GB and 2.3TB (as per provisioned space shown in datastore browser and output of stat from the shell.) However in the VM list and the VM summary page within the ESXi web interface it still shows this VM consuming 5.15TB. Veeam Backup & Replication also sees this VM size as 5.15TB, so presumably it's going to replicate 5.15TB of blank space - or maybe not, and it's just reading the same metadata or whatever ESXi is using to see the wrong size? This VM was previously only consuming 2.5TB, and the replicas with Veeam were the same. Why doesn't the WebUI reflect the shrunken file size, and is there a way to fix this?

Second, I don't understand why the VMDks suddenly consumed all their space? I know what triggered it, which was creating a snapshot, installing Windows Updates, then deleting the snapshot. However the sparse file for this snapshot was only 20GB. The "delete snapshot" task ran for multiple hours, and at some point the datastore ran out of space and the disks were marked as requiring consolidation. I freed up some space by removing a few replica VMs then ran the Consolidate Disks action, which took multiple more hours to complete. But why did it fill up the datastore like that? Is there something in the logs I can look for to help understand what happened, or is this expected for some reason? I did the same set of Windows Updates on two other VMs (one of which was on the same host and datastore) and their VMDKs did not grow like this.

Reply
0 Kudos
2 Replies
jasonh83
Contributor
Contributor

Shortly after posting this, ESXi (and by extent, Veeam) suddenly recognized the correct amount consumed space. I guess the WebUI keeps a cache and only checks it every few hours.

But if anyone can help me understand why they grew so much, that would be greatly appreciated.

Reply
0 Kudos
richard612
Enthusiast
Enthusiast

I witnessed the same behavior yesterday in my vSphere 6.7 homelab.  I spent more than an hour toying with Sdelete, UNMAP from the guest, host unmap configuration, and punchzero with no results.  Or so I thought.  Turns out I'd shrunk the VMDK by 25% or so but the web view of the VM (used space vs provisoned) was stuck at the original size.

Wondering if this refresh interval is configurable...

Reply
0 Kudos