Restart=always not working with ExecStopPost and dependency on service with StopWhenUnneeded=true

[omgold]

In found that in a special situation a service is not restarted, although marked as Restart=always. I narrowed it down to the combination of the two service files below.

If I do a kill -9 on the sleep command of myd2 the service enters failed state instead of restarting. But if I remove either the StopWhenUnneeded from myd1 or the ExecStopPost from myd2 it works as expected.

I believe this is a bug.

--- myd1.service ---
[Unit]
Description=example daemon 1
StopWhenUnneeded=true

[Service]
Type=oneshot
ExecStart=/bin/true
ExecStop=/bin/true
RemainAfterExit=true

--- myd2.service ---
[Unit]
Description=example daemon 2
Requires=myd1.service
After=myd1.service

[Service]
Type=simple
ExecStart=/usr/bin/sleep 100000000
ExecStopPost=/bin/true
Restart=always

[poettering]

Which systemd version is this happening with?

[omgold]

I run version 224.

[digrouz]

same issue on centos 7.2 running version 219.

Is there any fix or workaround ?

[Amyyum]

Has the problem been solved?

[omgold]

No. Just tested it with version v238-76-gf398c546c (f398c54). Still same behavior.

[Amyyum]

Ok thanks! Do you know any other way to get know when s systemd service restart ? Because I'm now trying to figure out if the application somehow silently failed and got respawned by systemd, or whether a user purposely restarted the service via systemctl.(Which Restart=always) Thank you very much.

…

On Thu, May 3, 2018 at 5:37 PM, omgold ***@***.***> wrote: No. Just tested it with version v238-76-gf398c546c ( f398c54). Still same behavior. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1154 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/Aiqns4XzMXOdMzhrTNyvV8p1fjkRb-2Mks5tus_agaJpZM4F4ZXP> .

-- *Amy Yu* *TVU Networks* 1279, ZhongShan Road West, Building 6, 10th floor <https://maps.google.com/?q=1279,+ZhongShan+Road+West,+Building+6,+10th+floor&entry=gmail&source=g> Shanghai, P.R.China amyyu@tvunetworks.com *Making every transmission a success* www.tvunetworks.com Facebook <https://www.facebook.com/TVUNetworks/> | Twitter <https://twitter.com/TVUNetworks> | Instagram <https://www.instagram.com/tvu_networks/> | LinkedIn <https://www.linkedin.com/in/carolinemccarty/>

[bl33pbl0p]

@omgold
This is very similar to #11456. I know what happens here too:

You have StopWhenUnneeded=true in myd2.service, you have Requires=myd2.service in myd1.service, Requires= forwards stop/restart requests on required unit to requiree, so when you go down, what happens is that the other unit has a stop job triggered from systemd due to referenced units not being active anymore, meanwhile, your ExecStopPost= keeps it hung in deactivating stage for a while. OTOH, this stop job on myd2.service is propagated to your unit back when it was triggered.

This creates a race: now, when the stop job installs itself in your unit, the state change (to failed) is attributed to it. Since Restart= only triggers when state change happens without any job's involvement, it will not fire a restart job for the same unit.

Fix is to probably introduce a version of Requires= that only has the strong effect at start and not propagation effect like PartOf=.

[bl33pbl0p]

@omgold Note that your bug was fixed with a3c1168 so this works in v240 nicely.

@poettering This can be closed.

[dtardon]

ping