systemd does not serialize/deserialize the ExecStart= or ExecStartPre= line number across daemon reloads #518
Description
(systemd 220)
I use a systemd unit to load and start fleet units inside a CoreOS cluster. The unit is pretty straightforward but I observe strange behaviours with successive ExecStart directives:
[Unit]
Description=Deploy Kubernetes units via fleet
Requires=fleet.service
After=fleet.service
[Service]
ExecStart=/usr/bin/bash -c '/usr/bin/fleetctl load /opt/fleet/*.service'
ExecStart=/usr/bin/fleetctl start \
kube-apiserver.service \
kube-scheduler.service \
kube-controller-manager.service \
kubelet.service \
kube-proxy.service
Type=oneshot
RemainAfterExit=true
The first ExecStart exits in 2-3s but the second ExecStart is not executed, the unit simply exits after the first command:
# systemctl status kube-fleet
● kube-fleet.service - Deploy Kubernetes units via fleet
Loaded: loaded (/etc/systemd/system/kube-fleet.service; static; vendor preset: disabled)
Active: active (exited) since Wed 2015-07-08 14:01:23 UTC; 2s ago
Main PID: 1230 (code=exited, status=0/SUCCESS)
Jul 08 14:01:21 ip-10-0-21-132.eu-west-1.compute.internal bash[1230]: Triggered global unit kubelet.service load
Jul 08 14:01:21 ip-10-0-21-132.eu-west-1.compute.internal bash[1230]: Triggered global unit provision-k8s-secrets.service load
Jul 08 14:01:23 ip-10-0-21-132.eu-west-1.compute.internal bash[1230]: Unit kube-scheduler.service loaded on 3d5044dd.../10.0.21.132
Jul 08 14:01:23 ip-10-0-21-132.eu-west-1.compute.internal bash[1230]: Unit kube-controller-manager.service loaded on 3d5044dd.../10.0.21.132
Jul 08 14:01:23 ip-10-0-21-132.eu-west-1.compute.internal bash[1230]: Unit generate-k8s-certs.service loaded on 3d5044dd.../10.0.21.132
Jul 08 14:01:23 ip-10-0-21-132.eu-west-1.compute.internal bash[1230]: Unit generate-k8s-tokens.service loaded on 3d5044dd.../10.0.21.132
Jul 08 14:01:23 ip-10-0-21-132.eu-west-1.compute.internal bash[1230]: Unit kube-addons.service loaded on 3d5044dd.../10.0.21.132
Jul 08 14:01:23 ip-10-0-21-132.eu-west-1.compute.internal bash[1230]: Unit kube-apiserver.service loaded on 3d5044dd.../10.0.21.132
Jul 08 14:01:23 ip-10-0-21-132.eu-west-1.compute.internal bash[1230]: Unit get-kubectl.service loaded on 3d5044dd.../10.0.21.132
Jul 08 14:01:23 ip-10-0-21-132.eu-west-1.compute.internal systemd[1]: Started Deploy Kubernetes units via fleet.
It works if I replace the first ExecStart directive with ExecStartPre:
# systemctl status kube-fleet
● kube-fleet.service - Deploy Kubernetes units via fleet
Loaded: loaded (/etc/systemd/system/kube-fleet.service; static; vendor preset: disabled)
Active: active (exited) since Wed 2015-07-08 14:04:43 UTC; 2s ago
Process: 1416 ExecStart=/usr/bin/fleetctl start kube-apiserver.service kube-scheduler.service kube-controller-manager.service kubelet.service kube-proxy.service (code=exited, status=0/SUCCESS)
Process: 1392 ExecStartPre=/usr/bin/bash -c /usr/bin/fleetctl load /opt/fleet/*.service (code=exited, status=0/SUCCESS)
Main PID: 1416 (code=exited, status=0/SUCCESS)
Jul 08 14:04:41 ip-10-0-21-132.eu-west-1.compute.internal bash[1392]: Unit kube-scheduler.service loaded on 3d5044dd.../10.0.21.132
Jul 08 14:04:41 ip-10-0-21-132.eu-west-1.compute.internal bash[1392]: Unit kube-apiserver.service loaded on 3d5044dd.../10.0.21.132
Jul 08 14:04:41 ip-10-0-21-132.eu-west-1.compute.internal bash[1392]: Unit generate-k8s-tokens.service loaded on 3d5044dd.../10.0.21.132
Jul 08 14:04:41 ip-10-0-21-132.eu-west-1.compute.internal bash[1392]: Unit kube-addons.service loaded on 3d5044dd.../10.0.21.132
Jul 08 14:04:41 ip-10-0-21-132.eu-west-1.compute.internal fleetctl[1416]: Triggered global unit kubelet.service start
Jul 08 14:04:41 ip-10-0-21-132.eu-west-1.compute.internal fleetctl[1416]: Triggered global unit kube-proxy.service start
Jul 08 14:04:43 ip-10-0-21-132.eu-west-1.compute.internal fleetctl[1416]: Unit kube-apiserver.service launched on 3d5044dd.../10.0.21.132
Jul 08 14:04:43 ip-10-0-21-132.eu-west-1.compute.internal fleetctl[1416]: Unit kube-controller-manager.service launched on 3d5044dd.../10.0.21.132
Jul 08 14:04:43 ip-10-0-21-132.eu-west-1.compute.internal fleetctl[1416]: Unit kube-scheduler.service launched on 3d5044dd.../10.0.21.132
Jul 08 14:04:43 ip-10-0-21-132.eu-west-1.compute.internal systemd[1]: Started Deploy Kubernetes units via fleet.
From my understanding of the documentation both directives should behave the same way as long as the Service type is oneshot.
Another confusing detail: if I replace the first ExecStart by another, faster to execute, bash command like bash -c 'echo hello' both commands are executed.