vsphere-iso: POST "/sdk": 502 Proxy Error

[xchotard-talentsoft]

Overview of the Issue

I've been using Packer / vsphere-iso for a long time without any issues but somehow it broke a few months ago and now I can't build new images:

errored after 5 minutes 30 seconds: POST "/sdk": 502 Proxy Error

Notes:

• It used to work but it suddenly broke a few months ago (summer 2025).
• I suspect this is related to vSphere 8.x. It worked well under vSphere 7.x.x.
• I have no idea of the content of the POST thats trigger the 502 Error.
• Guest installation runs fine but is interrupted by the shutdown.
• Reproducible on multiple vSphere clusters.

Reproduction Steps

Run "packer build ." with the provided demo project bellow, connect to a 8.0.x vSphere.

Packer Version

Packer v1.14.2 (Linux amd64)

Plugin Version and Builders

packer-plugin-vsphere_v2.0.0_x5.0_linux_amd64.

Please select the builder.

• vsphere-iso
• vsphere-clone

VMware vSphere Version

8.0.3

Guest Operating System

• Windows Server Datacenter 2025
• Windows Server Datacenter 2022

Simplified Packer Buildfile

source "vsphere-iso" "datacenter-efi" {
  vm_name              = "windows-server-2025-datacenter-efi-packer"
  CPUs                 = 2
  RAM                  = 8192
  disk_controller_type = ["lsilogic-sas"]
  guest_os_type        = "windows2022srvNext_64Guest"
  firmware             = "efi"
  iso_url              = # private url to iso
  iso_checksum         = # iso checksum  
  boot_wait            = "2s"
  boot_command         = ["<spacebar>"]
  network_adapters {
    network      = var.vcenter_network
    network_card = "vmxnet3"
  }
  storage {
    disk_size             = 32768
    disk_thin_provisioned = true
  }
  cdrom_type   = "sata"
  remove_cdrom = true
  floppy_content = {
    "autounattend.xml" = templatefile("./autounattend_efi.xml.pkrtpl")
  }
}

Omitted parameters: host,password,username,vcenter_server,datacenter,datastore,cluster, winrm communicator.

Operating System and Environment Details

Launched build from:

• Debian 13 amd64

Log Fragments and crash.log Files

Build results:

[...]
==> vsphere-iso.datacenter-efi: Powering on virtual machine...
==> vsphere-iso.datacenter-efi: Waiting 2s for boot...
==> vsphere-iso.datacenter-efi: Typing boot command...
==> vsphere-iso.datacenter-efi: Waiting for IP...
==> vsphere-iso.datacenter-efi: Clearing boot order...
==> vsphere-iso.datacenter-efi: Powering off virtual machine...
==> vsphere-iso.datacenter-efi: Deleting floppy image...
==> vsphere-iso.datacenter-efi: Destroying VM...
==> vsphere-iso.datacenter-efi: Closing sessions...
Build 'vsphere-iso.datacenter-efi' errored after 5 minutes 28 seconds: POST "/sdk": 502 Proxy Error

==> Wait completed after 5 minutes 28 seconds

==> Some builds didn't complete successfully and had errors:
--> vsphere-iso.datacenter-efi: POST "/sdk": 502 Proxy Error

==> Builds finished but no artifacts were created.

This is the log without "PACKER_LOG=1" because the later is more verbose but does not provides more info about that 502 error.

[tenthirtyam]

The HTTP 502 typically indicates a server-side issue, such as a server overload, network problems, or a misconfiguration between the endpoints involved in processing the request. This is likely an issue in the upstream endpoint.

Did the behavior begin to occur immediately after an upgrade from 7x to 8x or was it gradual?

Try running the following on the VCSA to get the status of the services:

service-control --status

You may need to restart the API facing services on the VCSA network instances:

service-control --stop vsphere-ui
service-control --stop vpxd
service-control --start vpxd
service-control --start vsphere-ui

Recommend reaching out to Broadcom Support to investigate the reason for the HTTP 502 response from the vCenter instance.

Ryan Johnson, VMware by Broadcom

[xchotard-talentsoft]

Thanks Ryan.

The HTTP 502 typically indicates a server-side issue, such as a server overload, network problems, or a misconfiguration between the endpoints involved in processing the request. This is likely an issue in the upstream endpoint.

That was also my though so I did contact our provider to open a case. Currently awaiting for their feedback.

Did the behavior begin to occur immediately after an upgrade from 7x to 8x or was it gradual?

I don't run daily image builds so I'm not 100% sure.
It did work during spring on every vSphere clusters, and now it does not.
Same Packer / vsphere plugin version, but all vSphere clusters have been upgraded to 8x during summer.
I'll be waiting for feedback from our hosting provider.

[xchotard-talentsoft]

The problem is no longer present.
vSphere was fine, the root cause was a timeout on a custom reverse proxy.
They increased some timeout thresholds and now it works.
Thanks again for your time.