QA (public) » openQA Project (public) » openQA Infrastructure (public)

According to https://discourse.nixos.org/t/how-do-i-configure-nfs-server-to-start-after-the-required-filesystem-mounts-are-ready/22540/10 as we now use "nofail" we should still add x-systemd.before=local-fs.target to each mount point definition to prevent the continuation of services but I want to check if that still leaves sshd to start.

I ran

qemu-system-x86_64 -m 8192 -snapshot -drive file=/dev/vda,if=virtio -drive file=/dev/vdb,if=virtio -drive file=/dev/vdc,if=virtio -drive file=/dev/vdd,if=virtio -drive file=/dev/vde,if=virtio -nographic -serial mon:stdio -smp 4

on OSD while setting the second number on each mount in /etc/fstab for vdb…e to 0 so the system actually booted but then a lot of journal errors cropped up so I stopped again. Maybe the available storage for "snapshot" was exhausted. Trying to simulate locally with

qemu-system-x86_64 -m 8192 -enable-kvm -snapshot -drive file=SLES15-SP6-Minimal-VM.x86_64-kvm-and-xen-GM.qcow2,if=virtio -drive file=/dev/null,if=virtio -drive file=/dev/null,if=virtio -drive file=/dev/null,if=virtio -drive file=/dev/null,if=virtio -nographic -serial mon:stdio -smp 4

but this shows nothing usable after the initial grub loading. I guess the serial terminal isn't properly enabled.

Booted locally with graphical interface using

for i in vdb vdc vdd vde ; do qemu-img create -f qcow2 $i.qcow2 2G; done
qemu-system-x86_64 -m 8192 -enable-kvm -drive file=SLES15-SP6-Minimal-VM.x86_64-kvm-and-xen-GM.qcow2,if=virtio -drive file=vdb.qcow2,if=virtio -drive file=vdc.qcow2,if=virtio -drive file=vdd.qcow2,if=virtio -drive file=vde.qcow2,if=virtio -smp 4

and then in /etc/default/grub enable "serial console" instead of "gfxterm" and GRUB_SERIAL_COMMAND="serial --unit=0 --speed=115200" and called update-bootloader. Also zypper ref && zypper -n in nfs-kernel-server so that I can test with that.

Afterwards booted again with

qemu-system-x86_64 -m 8192 -enable-kvm -drive file=SLES15-SP6-Minimal-VM.x86_64-kvm-and-xen-GM.qcow2,if=virtio -drive file=vdb.qcow2,if=virtio -drive file=vdc.qcow2,if=virtio -drive file=vdd.qcow2,if=virtio -drive file=vde.qcow2,if=virtio -smp 4 -nographic -serial mon:stdio

Then copied over the content of /etc/exports and /etc/fstab from OSD, replacing UUIDs with my clean simulation devices vdb…e. I just added the mount point definitions but have not created a filesystem yet to simulate problems on those devies. On boot system was stuck in emergency mode, not entirely expected as we had "nofail" on all partitions. The error:

[    4.393865][  T440] systemd-fstab-generator[440]: Failed to create unit file '/run/systemd/generator/srv.mount', as it already exists. Duplica?
[    4.399227][  T434] (sd-execu[434]: /usr/lib/systemd/system-generators/systemd-fstab-generator failed with exit status 1.

I disabled the original mount for /srv with the btrfs subvolume. But I don't see kernel and systemd boot output anymore on the serial terminal. Don't know what broke. ok, the "quiet" flag was set so removed that from /etc/default/grub and called update-bootloader.

Anyway, the system boots and eventually also shows a login prompt. In there it looks like nfs-server is not started because the mount point isn't populated so how it should be. Ok, but I never enabled nfs-server. Did that now and created filesystems and the complete directory structure to populate all mount points as on OSD. Enabled nfs-server, rebooted, all good. Now disabled the mount for /assets in /etc/fstab. I guess the problem is that the mount point for /assets /var/lib/openqa/share none x-systemd.requires=/var/lib/openqa,x-systemd.automount,bind,nofail 0 0 can work regardless if something is mounted on /assets or not. Let's see. So systemctl cat var-lib-openqa-share.mount shows that we require var-lib-openqa.mount but not something like assets.mount. Should we also require that?

I did

/dev/vdc /assets xfs_foo defaults,logbsize=256k,noatime,nodiratime,nofail 1 2
/assets /var/lib/openqa/share none x-systemd.requires=/var/lib/openqa,x-systemd.automount,bind,nofail 0 0

so forcing a condition where a mount unit for /assets is defined but fails to mount as "xfs_foo" is purposely invalid. This caused the system to still boot up to the point of reaching a login prompt and started sshd but systemctl status nfs-server shows that the NFS server did not start with message

Job nfs-server.service: Job nfs-server.service/start failed with result 'dependency'

which we wanted. Same for var-lib-openqa-share.mount . Doesn't say which dependency but ok. systemctl --failed clearly shows that assets.mount failed so it should be clear where to start investigating. Now, what about other devices, like vdb, vdd, vde? /srv/PSQL also has no dependency on /srv as apparent in systemctl cat var-lib-pgsql.mount. First simulating that all vdb…e fail with "xfs_foo". Got on systemctl --failed

# systemctl --failed
  UNIT                LOAD   ACTIVE SUB    DESCRIPTION
● assets.mount        loaded failed failed /assets
● results.mount       loaded failed failed /results
● space\x2dslow.mount loaded failed failed /space-slow
● srv.mount           loaded failed failed /srv

so far as expected. And all dependant mount points are not mounted which is good. systemctl cat var-lib-pgsql.mount for example says that it's inactive and mentions the missing dependency but systemctl cat … does not explicitly list it so apparently the What=/srv/PSQL is enough. Maybe we can actually remove some requires then? Looks like everything still works fine. I assume the original problem for this ticket happened because we went a bit back and forth disabling complete lines in fstab. So let's see what happens if I disable lines completely. Then nfs-server still does not start as the directory /var/lib/openqa/share does not exist. I installed openQA in my environment which actually created the directory structure which nfs-server would expect even though that is not the correct one. We could manually delete the directories in the unmounted directories.

What about

[Unit]
ConditionPathIsMountPoint=/var/lib/openqa/share

Now trying to start fails with

NFS server and services was skipped because of an unmet condition check (ConditionPathIsMountPoint=/var/lib/openqa/share).

that sounds promising. Enabling all mounts in /etc/fstab again. Triggered reboot and verified that this worked fine. So we can add that extra dependency I guess. Put into salt and tested with

salt --no-color --state-output=changes 'openqa.suse.de' state.test etc.master.init | grep -v 'Result: Clean'
salt --no-color --state-output=changes 'openqa.suse.de' state.apply etc.master.init | grep -v 'Result: Clean'

https://gitlab.suse.de/openqa/salt-states-openqa/-/merge_requests/1213