Is there any way to make it only care about sigkill from the bash command that is the root of the process tree?
I can't reproduce this failure. Can you give me a simple test case?
This seems to make it happen on my box with ~100% reliability
achew22@springs:~$ sh -c 'sleep 1000s' &
achew22@springs:~$ kill -9 %1
Container my-container terminated by signal KILL.
If I run that on other boxes it doesn't seem to have a problem with it killing the parent.
I cannot reproduce this in my toolbox (PRETTY_NAME="Fedora 23 (Twenty Three)"). Which version of Fedora is in your toolbox? Can you try recreating the toolbox with the latest?
PRETTY_NAME="Fedora 23 (Twenty Three)"
@achew22 The same for me. I've tried with an image of ubuntu, debian and fedora (latest versions). Any workarround to avoid this weird bug? Thank you all.
Edit: Not in fedora:latest, but when I try to run byobu (tmux wrapper), toolbox exits with the same message.
I'm getting the same error with pritunl/archlinux any time I try to install a package (e.g. pacman -S tmux). But I tried @achew22's test and that doesn't kill it.
pacman -S tmux
There is a workaround linked several issues over. just copying it over to this issue since it took me a while to track it down again and the other issues are closed as dupes of this one.
Use the previous fedora tag
echo "TOOLBOX_DOCKER_TAG=24" >>$HOME/.toolboxrc
When one actually needs Fedora 25 for whatever reasons, this workaround from #1676 (comment) worked for me:
rpm --import /etc/pki/rpm-gpg/RPM*
There is another workaround: remove --share-system from systemd-nspawn command line. Somehow it propagates SIGKILL. I reference #1610 here