RemainAfterExit should respect return codes and/or Restart= modes

[binarybana]

Submission type

• Bug report
• Request for enhancement (RFE)

systemd version the issue has been seen with

225 (but this is an RFE, and I didn't see anything in NEWS about changes to this behavior)

Used distribution

CoreOS

Using something like

[Service]
ExecStart=/usr/bin/false
Type=simple
RemainAfterExit=yes
Restart=on-failure
RestartSec=10

Despite the fact that the service fails, it does not get restarted. Ideally I could specify something like RemainAfterExit=on-success to only remain if exiting with a success return code.

My primary use case for this is using Type=simple jobs for one off scripts at bootup (like formatting drives) where I need Restart=on-failure but also RemainAfterExit. This relates to (but is distinct from) #2582, because even with that issue fixed, you'd still want to add this behavior.

[arvidjaar]

Well, strictly speaking service had never been started in the first place, so Restart logically does not apply here. This would need new directive with semantic "keep starting this service even if initial attempt fails". Although exact semantic is rather hairy - what to do with all ExecStartPre? Should they be repeated or not? Should any cleanup be attempted (i.e. ExecStop*)?

[binarybana]

Perhaps something like RetryStarting=yes.

And my votes on the semantics would be: repeat ExecStartPre directives yet ExecStop* are only for explicit stop commands.

[Keruspe]

Note that another side-effect is that while the exit code of the unit is 1, its state is "active", not "failed" which makes systemctl is-failed claim that it's not.

I think that RemainAfterExit=on-succes at least makes sense, and I'd say that's what I would expect anyway, even with yes.

[arvidjaar]

@binarybana I apologize, I misunderstood it. In case of simple service it is already started as soon as main process is spawned. So the problem is indeed interaction with RemainAfterExit. The same issue will have Type=forking when forked process terminates.

@Keruspe Yes, and when service is stopped (systemctl stop ...) it suddenly becomes failed.

I wonder whether RemainAfterExit should be allowed with non-oneshot services in the first place. @binarybana , could you explain why you need it and not Type=oneshot?

[gegenschall]

I stumbled across this when creating a tmux.service unit. What I want to achieve is having the RemainAfterExit behavior but also have the service restarting when it exists (any exit code is fine).
This way I don't need to systemctl --user restart tmux when I accidentally quit my tmux session.

There is currently no way to achieve this, right?

edit: Actually, If you tell tmux to remain-on-exit on this all works out pretty nicely. Still, a less application specific way to achieve this would be greatly appreciated.