Force systemd to stop lvm2-lvmetad.service before shutting down.

[DarkWav]

I noticed that systemd sometimes reaches shutdown.target before lvm2-lvmetad.service has been stopped. Since lvm2-lvmetad.service (a.k.a. lvm2_lvmetad_systemd_red_hat.service.in in the "scripts" folder) has

[Unit]
Conflicts=shutdown.target

but no

[Unit]
Before=shutdown.target

in it, this problem can easily occur if the service doesn't react to SIGTERM fast enough.
In case it does happen, with kernels 5.0.0 and up the shutdown/reboot procedure can be interrupted.
If we would add

[Unit]
Before=shutdown.target

to the service file we could avoid the issue. I heared a some Arch Linux users are suffering from this occasional problem, as described in this bug report.
This issue at systemd is also related to the problem, as @patrakov describes.
Thanks for your efforts in advance.

[wangwang3210]

It seems that we encounter the same problem.
#16

[prajnoha]

Thanks for reporting the issue, it should be fixed now with:

(master tree)
https://sourceware.org/git/?p=lvm2.git;a=commit;h=0cab341e1d0e8f9089d3c62d3adbec24dfd5e124

(stable-2.02 tree)
https://sourceware.org/git/?p=lvm2.git;a=commit;h=0a726a7e268b31856615491809af73bda5d4d6f9

[DarkWav]

Thanks a lot, you're awesome!

[intelfx]

I think something went wrong with the backport in 0a726a7e268b31856615491809af73bda5d4d6f9.

diff --git a/scripts/lvm2_lvmetad_systemd_red_hat.socket.in b/scripts/lvm2_lvmetad_systemd_red_hat.socket.in
index 2663c72..3201537 100644 (file)
--- a/scripts/lvm2_lvmetad_systemd_red_hat.socket.in
+++ b/scripts/lvm2_lvmetad_systemd_red_hat.socket.in
@@ -1,8 +1,6 @@
 [Unit]
 Description=LVM2 metadata daemon socket
 Documentation=man:lvmetad(8)
-DefaultDependencies=no
-Conflicts=shutdown.target
 
 [Socket]
 ListenStream=@DEFAULT_RUN_DIR@/lvmetad.socket

Is it actually correct to do stuff like this? Doesn't look like it, cherry-picking this commit breaks boot in Arch (FS#62302).

[kouros17]

I'm using manjaro and after the last update of of lvm2 ( 2.02.184-2) which include the fix commit the "long shutdown" issue solved but now from journalctl I get:

manjaro systemd[1]: boot-efi.mount: Job lvm2-lvmetad.socket/start deleted to break ordering cycle starting with boot-efi.mount/start
manjaro systemd[1]: boot-efi.mount: Job lvm2-monitor.service/start deleted to break ordering cycle starting with boot-efi.mount/start

[prajnoha]

My mistake, sorry for that. I expected the sockets.target to be reached sooner during boot, but it's even after sysinit.target (https://www.freedesktop.org/software/systemd/man/bootup.html). So yes, we still need the DefaultDependencies=no there.

(master tree)
https://sourceware.org/git/?p=lvm2.git;a=commit;h=25f231cf06c3938838a64f5b88834d831f45c1c2

(stable-2.02 tree)
https://sourceware.org/git/?p=lvm2.git;a=commit;h=03d6cfdd99fb1c2fd85721ab6faebbb1b6658873

[kouros17]

It's fixed for me with lvm2 2.02.184-4 in arch.
@prajnoha thanks a lot!