Comment 0 for bug 1739107

This is a request to make a change in the hv-kvp-daemon systemd service which is part of the linux-cloud-tools-common package to ensure the hv-kvp-daemon service starts before the walinuxagent service. The default dependencies make hv-kvp-daemon wait until the whole system is up before it can start.

Currently the /lib/systemd/system/hv-kvp-daemon.service file looks like this:

====================
# On Azure/Hyper-V systems start the hv_kvp_daemon
#
# author "Andy Whitcroft <email address hidden>"
[Unit]
Description=Hyper-V KVP Protocol Daemon
ConditionVirtualization=microsoft

[Service]
ExecStart=/usr/sbin/hv_kvp_daemon -n

[Install]
WantedBy=multi-user.target
====================

The suggested modification is to make the [Unit] section look like this:

[Unit]
Description=Hyper-V KVP Protocol Daemon
ConditionVirtualization=microsoft
DefaultDependencies=no
After=systemd-remount-fs.service
Before=shutdown.target cloud-init-local.service walinuxagent.service
Conflicts=shutdown.target
RequiresMountsFor=/var/lib/hyperv

The hv-kvp-daemon service is not currently part of the critical-chain:

$ systemd-analyze critical-chain
The time after the unit is active or started is printed after the "@" character.
The time the unit takes to start is printed after the "+" character.

graphical.target @10.809s
└─multi-user.target @10.723s
└─ephemeral-disk-warning.service @10.538s +31ms
└─cloud-config.service @8.249s +2.252s
└─basic.target @8.044s
└─sockets.target @8.019s
└─snapd.socket @7.692s +264ms
└─sysinit.target @6.719s
└─cloud-init.service @5.803s +842ms
└─networking.service @5.137s +612ms
└─network-pre.target @5.074s
└─cloud-init-local.service @2.257s +2.783s
└─systemd-remount-fs.service @1.368s +656ms
└─systemd-journald.socket @1.218s
└─-.mount @649ms
└─system.slice @653ms
└─-.slice @649ms

In an Azure VM, the current startup time of my test is:
$ systemd-analyze
Startup finished in 10.375s (kernel) + 12.352s (userspace) = 22.728s

After making the suggested change, the startup time is similar:

$ systemd-analyze
Startup finished in 9.759s (kernel) + 11.867s (userspace) = 21.627s

And the service is now in the critical-chain:

$ systemd-analyze critical-chain
The time after the unit is active or started is printed after the "@" character.
The time the unit takes to start is printed after the "+" character.

graphical.target @10.666s
└─multi-user.target @10.636s
└─ephemeral-disk-warning.service @10.556s +36ms
└─cloud-config.service @8.423s +2.095s
└─basic.target @8.124s
└─sockets.target @8.101s
└─lxd.socket @7.677s +326ms
└─sysinit.target @6.755s
└─cloud-init.service @5.814s +908ms
└─networking.service @5.111s +651ms
└─network-pre.target @5.087s
└─cloud-init-local.service @2.345s +2.707s
└─hv-kvp-daemon.service @2.316s
└─systemd-remount-fs.service @1.253s +680ms
└─system.slice @1.225s
└─-.slice @650ms

The ConditionVirtualization=microsoft line makes it so that this doesn't affect non microsoft virtualization environments (ie. qemu, kvm, vmware, xen, etc.)