snap auto-import causes part-probe to hang while running in a VMware VM

[Lee Trager]

A customer is trying to setup a Jenkins job running on Ubuntu 18.04 in VMware ESXi 6.7 to build the VMware ESXi 6.7 MAAS image with packer-maas[1]. I have not been able to reproduce this issue on metal.

packer-maas is a Packer[2] template which uses QEMU to install VMware ESXi onto a disk image. After the installation is complete a post-processing script[3] is run which mounts the disk image and injects Curtin hooks. The post install script uses qemu-nbd to load the disk image and partprobe to ensure device entries are created for all partitions before mounting.

When partprobe is run it calls udevadm settle which hangs causing the build to fail. While partprobe is running /usr/bin/snap auto-import --mount=/dev/nbd4pX keeps spawning for the various partitions. Because udevadm settle never becomes unstuck the build fails.

If I remove snapd from the system I can successfully build the image with udevadm blocking. I can only produce this on a fresh install, apt purge snapd; apt install snapd seems to fix this as well. The install I reproduced on was from MAAS so this may be an issue with how the cloud-image is being generated.

[1] https://code.launchpad.net/~maas-committers/maas/+git/packer-maas
[2] https://www.packer.io/
[3] https://git.launchpad.net/~maas-committers/maas/+git/packer-maas/tree/vmware-esxi/vmware-esxi.json#n32
[4] http://cloud-images.ubuntu.com/bionic/current/

[Ian Johnson (anonymouse67)]

Hi, can you clarify that you are seeing `snap auto-import ...` processes start and never return? What's the output of `pstree` for example on an affected system?

[Launchpad Janitor (janitor)]

[Expired for snapd because there has been no activity for 60 days.]