[systemd-devel] One failed restart leaves dependent units permanently dead

[Shimin Guo]

Summary: If bar depends on foo via Requires and After, a failed restart on
foo leaves bar permanently in the failed state, even if a later autorestart
successfully brings up foo, even if bar is configured to always restart. Is
this how systemd restart and dependency is supposed to work?

To repro:

foo.service:

[Service]
Restart=always
RestartSec=5s
Type=notify
# sleep forever if /tmp/foo is present, otherwise fail
ExecStart=/usr/bin/bash -c "if [[ -e /tmp/foo ]]; then systemd-notify
--ready; while true; do sleep 99999; done; else exit 1; fi"

bar.service

[Unit]
Requires=foo.service
After=foo.service

[Service]
Restart=always
RestartSec=5s
ExecStart=/usr/bin/bash -c "while true; do /usr/bin/sleep 99999; done"

- Initially, both foo and bar are running.
- Then I rm'ed /tmp/foo and killed the main process for foo.service.
- Systemd noticed that foo is dead and attempted to restart it.
- In doing so, it stopped bar first, then foo.
- Then it tried to start foo but failed because /tmp/foo was missing.
- That put bar in a failed state with result 'dependency'.
- Systemd then tried to restart foo every 5 seconds.
- At this point I touched /tmp/foo, so foo started successfully.
- However, systemd didn't bother with starting bar at all in the restart
attempts after the first one.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/systemd-devel/attachments/20160610/1de9c6c8/attachment.html>