[pve-devel] [PATCH 0/2] Fix systemd ordering for qmeventd.service

[pve-devel] [PATCH 0/2] Fix systemd ordering for qmeventd.service

[Stefan Reiter]

Reported in the forum:
https://forum.proxmox.com/threads/shutdown-reboot-error-nodes.78906/

Since qmeventd is now responsible for stopping QEMU instances after they shut
down, it needs to keep running for as long as there are guests, otherwise the
host will hang waiting for guests to exit.

Fix this in pve-guests.service for regular VMs and pve-ha-lrm.service for HA.

I'm not too familiar with systemd services, but I figured the After is enough,
since Wants implies that qmeventd must be running for the services to work,
which is not true, the VMs will work fine, as long as the user takes care of
cleanup themselves somehow or doesn't use shutdown (and by default
qmeventd.service is enabled anyway). But easy enough to change if that's wrong.

Not sure how to handle version bumping, I'd assume qemu-server needs a depends
on new pve-manager and pve-ha-manager to ensure non-broken behaviour?


manager: Stefan Reiter (1):
  stop qmeventd.service after pve-guests.service

 services/pve-guests.service | 1 +
 1 file changed, 1 insertion(+)

ha-manager: Stefan Reiter (1):
  stop qmeventd.service after pve-ha-lrm.service

 debian/pve-ha-lrm.service | 1 +
 1 file changed, 1 insertion(+)

-- 
2.20.1

[pve-devel] [PATCH manager 1/2] stop qmeventd.service after pve-guests.service

[Stefan Reiter]

Otherwise stopped guests can't be cleaned up and will prevent host
shutdown.

Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
---
 services/pve-guests.service | 1 +
 1 file changed, 1 insertion(+)

diff --git a/services/pve-guests.service b/services/pve-guests.service
index 6b044a5f..a2f8e068 100644
--- a/services/pve-guests.service
+++ b/services/pve-guests.service
@@ -14,6 +14,7 @@ After=spiceproxy.service
 After=pve-firewall.service
 After=lxc.service
 After=pve-ha-crm.service pve-ha-lrm.service
+After=qmeventd.service
 
 [Service]
 Environment="PVE_LOG_ID=pve-guests"
-- 
2.20.1

[pve-devel] [PATCH ha-manager 2/2] stop qmeventd.service after pve-ha-lrm.service

[Stefan Reiter]

Otherwise stopped HA guests can't be cleaned up and will prevent host
shutdown.

Signed-off-by: Stefan Reiter <s.reiter@proxmox.com>
---
 debian/pve-ha-lrm.service | 1 +
 1 file changed, 1 insertion(+)

diff --git a/debian/pve-ha-lrm.service b/debian/pve-ha-lrm.service
index fb1b5db..367a93d 100644
--- a/debian/pve-ha-lrm.service
+++ b/debian/pve-ha-lrm.service
@@ -17,6 +17,7 @@ After=pveproxy.service
 After=ssh.service
 After=syslog.service
 After=watchdog-mux.service
+After=qmeventd.service
 
 [Service]
 ExecStart=/usr/sbin/pve-ha-lrm start
-- 
2.20.1

[pve-devel] applied: [PATCH 0/2] Fix systemd ordering for qmeventd.service

[Thomas Lamprecht]

On 12.11.20 16:43, Stefan Reiter wrote:
> Reported in the forum:
> https://forum.proxmox.com/threads/shutdown-reboot-error-nodes.78906/
> 
> Since qmeventd is now responsible for stopping QEMU instances after they shut
> down, it needs to keep running for as long as there are guests, otherwise the
> host will hang waiting for guests to exit.
> 
> Fix this in pve-guests.service for regular VMs and pve-ha-lrm.service for HA.
> 
> I'm not too familiar with systemd services, but I figured the After is enough,
> since Wants implies that qmeventd must be running for the services to work,
> which is not true, the VMs will work fine, as long as the user takes care of
> cleanup themselves somehow or doesn't use shutdown (and by default
> qmeventd.service is enabled anyway). But easy enough to change if that's wrong.
> 
> Not sure how to handle version bumping, I'd assume qemu-server needs a depends
> on new pve-manager and pve-ha-manager to ensure non-broken behaviour?
> 
> 
> manager: Stefan Reiter (1):
>   stop qmeventd.service after pve-guests.service
> 
>  services/pve-guests.service | 1 +
>  1 file changed, 1 insertion(+)
> 
> ha-manager: Stefan Reiter (1):
>   stop qmeventd.service after pve-ha-lrm.service
> 
>  debian/pve-ha-lrm.service | 1 +
>  1 file changed, 1 insertion(+)
> 



applied the reverse to qemu-server, thanks!