Review: Needs Fixing
> I still have to test this simulating a hung kernel, but the panic +
> no-reboot logic looks nice and solid.
I suppose should add a vmtest (not enabled by default, I think) which triggers
a panic. For curtin, we'd just run an early command like so:
echo c > /proc/sysrq-trigger
> The crashdump part is still not fully clear to me, as I don't understand why
> we need to install linux-image-generic (the VM is already running a kernel,
> isn't it?) and why we need MODULES=most to the point of patching
> /etc/kernel/postinst.d/kdump-tools..
Curtin vmtests boot via -kernel -initrd and get root from http url. crashdump
expects that your kernel is installed via dpkg; so certain files it looks for
are only present if you install a kernel. So we install a kernel.
For the MODULES=most, crashdump tries to determine the correct storage
modules it might need to mount root, however, since we booted via -kernel
we don't have a root blockdevice that can be found. As a workaround, I
modify crashdump to enable initramfs generation with most. This allows us
to properly configure crashdump (it does take longer and uses more ram).
I'll also include this in the crashdump config.
Diff comments:
> diff --git a/examples/tests/crashdump.cfg b/examples/tests/crashdump.cfg
> new file mode 100644
> index 0000000..e010961
> --- /dev/null
> +++ b/examples/tests/crashdump.cfg
> @@ -0,0 +1,19 @@
> +_install_crashdump:
> + - &install_crashdump |
> + command -v apt &>/dev/null && {
> + DEBIAN_FRONTEND=noninteractive apt-get -qy install linux-image-generic
> + debconf-set-selections <<< "kexec-tools kexec-tools/load_kexec
> boolean true"
The question is indeed confusing; all I know is that I wanted to use kexec;
thats how crashdump works!
> + debconf-set-selections <<< "kdump-tools kdump-tools/use_kdname
> boolean true"
> + DEBIAN_FRONTEND=noninteractive apt-get -qy install linux-crashdump;
> + mkdir -p /var/lib/kdump
> + # fix up crashdump post-inst to just put all of the modules in
> + sed -i -e 's,MODULES=dep,MODULES=most,'
> /etc/kernel/postinst.d/kdump-tools
vmtest boot with out a root block device, so dep cannot 'find' a root disk to
generate deps from.
> + kdump-config load
> + kdump-config show
> + }
> + exit 0
> +
> +
> +early_commands:
> + # run before other install commands
> + 0000_aaaa_install_crashdump: ['bash', '-c', *install_crashdump]
> diff --git a/tests/vmtests/__init__.py b/tests/vmtests/__init__.py
> index 222adcc..e102b6d 100644
> --- a/tests/vmtests/__init__.py
> +++ b/tests/vmtests/__init__.py
> @@ -967,6 +968,25 @@ class VMBaseClass(TestCase):
> for service in ["systemd.mask=snapd.seeded.service",
> "systemd.mask=snapd.service"]])
>
> + # We set guest kernel panic=1 to trigger immediate rebooot, combined
haha, I tried to use the longer version of reboot!
> + # with the (xkvm) -no-reboot qemu parameter should prevent vmtests
> from
> + # wasting time in a soft-lockup loop. Add the params after the '---'
> + # separator to extend the parameters to the target system as well.
> + cmd.extend(["--no-reboot", "--append=panic=-1",
> + "--append=softlockup_panic=1",
> + "--append=hung_task_panic=1",
> + "--append=nmi_watchdog=panic,1"])
> +
> + # configure guest with crashdump to capture kernel failures for debug
> + if cls.crashdump:
> + # we need to install a kernel and modules so bump the memory by
> 2g
> + # for the ephemeral environment to hold it all
> + cls.mem = int(cls.mem) + 2048
> + logger.info(
> + 'Enabling linux-crashdump during install, mem += 2048 = %s',
> + cls.mem)
> + cmd.extend(["--append=crashkernel=384M-5000M:192M"])
> +
> # getting resolvconf configured is only fixed in bionic
> # the iscsi_auto handles resolvconf setup via call to
> # configure_networking in initramfs
--
https://code.launchpad.net/~raharper/curtin/+git/curtin/+merge/383805
Your team curtin developers is subscribed to branch curtin:master.
--
Mailing list: https://launchpad.net/~curtin-dev
Post to : curtin-dev@lists.launchpad.net
Unsubscribe : https://launchpad.net/~curtin-dev
More help : https://help.launchpad.net/ListHelp
