Since Stretch with its newer systemd this service had a regression
when stopping. The timeout for stop changed from infinity to 90
seconds, meaning that if it is not finished until then it receives a
sigkill shortly therafter.
As it shutdowns (first gracefully, then after 180 seconds
ungracefully) all VMs and CTs on host powerdown this is an issue.
A single blocking VM could let all later ones get killed, even if they
would shut down gracefully.
This shortcomming gets addressed in patch 1, which is probably the
most important of this series.
Patch 2 - 3 try to improve the service and its contact with users,
by refusing manual start/stop/restart through systemctl – if such
action is desired the API/pvesh can be still used. More logging, so
that errors and synchronization points from the API call do not get
thrown away – this does not help anybody.
Moving the open-iscsi service to the pve-storage.target, seems to fit
better there - no strong feeling about this change.
The last one changes the name of this unit – while keeping the old one
as an alias for the 5.x release (legacy reasons, yay), as this may see
some critique I'd explicitly put it last, so that the other patches
can be still applied.
Thomas Lamprecht (5):
pve-manager.service: do not allow stop to timeout
pve-manager.service: refuse manual start/stop
pve-manager.service: allow logging STDERR and STDOUT
pve-manager.service: move open-iscsi ordering to pve-storage.target
rename pve-manager.service to pve-guests
bin/init.d/Makefile | 3 +-
.../{pve-manager.service => pve-guests.service} | 11 ++---
bin/init.d/pve-manager | 51 ----------------------
bin/init.d/pve-storage.target | 1 +
debian/postinst | 3 +-
debian/postrm | 2 +-
6 files changed, 11 insertions(+), 60 deletions(-)
rename bin/init.d/{pve-manager.service => pve-guests.service} (73%)
delete mode 100755 bin/init.d/pve-manager
--
2.11.0
_______________________________________________
pve-devel mailing list
pve-devel@pve.proxmox.com
https://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel
