krun not stopping sometimes (1.17.4)

[marmitar]

Arch Linux just updated to libkrun-1.17.4 for testing (f7015b3) and I noticed that some short lived containers are not stopping after finishing the command. It doesn't happen every time, but something closer to 10% to 30% of the time. In the following logs, I got to reproduce it on the fourth trial:

• krun-not-stopping.log

The logs after the command ends are just this on repeat every few seconds:

[2026-02-19T04:51:27Z DEBUG devices::virtio::rng::event_handler] rng: request queue event
[2026-02-19T04:51:27Z DEBUG devices::virtio::rng::device] rng: process_req()

The only exception are logs like this after I press something on the keyboard (e.g. Ctrl+C):

[2026-02-19T04:51:55Z DEBUG devices::virtio::mmio[console]] interrupt: signal_used_queue

[slp]

I've just left a script running for quite a while continuously launching short-lived VMs which just run /bin/true, but couldn't reproduce the issue. I think we need more data about your environment:

• Arch (x86_64 or aarch64)
• libkrunfw version
• Which software is using libkrun
• If the above it's not podman+crun, which devices are configured in the VM.
• Which software is running inside the short-lived VMs (or, at least, which syscalls is primarily using).

[marmitar]

Sure, sorry about letting the details out like that.

• Arch: x86_64 (AMD Zen 4)
  • Host Kernel: 6.18.9-zen1-2-zen #1 ZEN SMP PREEMPT_DYNAMIC Mon, 09 Feb 2026 17:45:06 +0000 x86_64 GNU/Linux
• libkrunfw version: 5.2.1
  • Guest Kernel: 6.12.68 #1 SMP PREEMPT_DYNAMIC Tue Feb 17 16:15:12 CET 2026 x86_64 Linux
• Which software is using libkrun:
  • podman 5.7.1 (rootless) (podman-info)
  • crun 1.26
  • passt 2026_01_20.386b5f5
• which devices are configured in the VM: Arch builds with BLK=1 NET=1 EFI=0 GPU=1 SND=1 INPUT=1 (see PKGBUILD)
• Which software is running inside the short-lived VMs: I was running podman run --rm -it --runtime=krun archlinux curl ifconfig.me

It seems like the issue only happens with curl, actually. Later today I'll try more commands and different images to see if I can reproduce it in another way.

[marmitar]

I cleaned up everything (sudo rm -rf ~/.local/share/containers/ ~/.cache/containers/ $XDG_RUNTIME_DIR/containers/) and rebooted the machine, but the issue persists for me. I can only reproduce it with curl specifically, but not with wget or other base tools I found in common images. I was able to reproduce it with quay.io/curl/curl, though, which is based on Alpine, so it seems less likely to be something from the software inside the container (and crun is working fine too).

• Logs: env RUST_LOG=debug podman run --rm -it --runtime=krun --env RUST_LOG=debug quay.io/curl/curl:latest curl ipinfo.io

I also failed to mention before, but I have some config overrides. Hopefully they shouldn't impact too much on this issue:

[containers]
log_driver = "journald"
log_tag = "container/{{.Name}}"
dns_options = ["edns0", "trust-ad"]

[engine]
image_parallel_copies = 12
compression_format = "zstd"
compression_level = 5
compose_warning_logs = true