<div dir="ltr">I'm observing it on the ubuntu bionic that comes with systemd 237</div><div class="gmail_extra"> <div class="gmail_quote">On 19 May 2018 at 07:27, Lennart Poettering <<a href="mailto:lennart@poettering.net" target="_blank">lennart@poettering.net</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="HOEnZb"><div class="h5">On Fr, 18.05.18 09:02, Ivan Kurnosov (<a href="mailto:zerkms@zerkms.ru">zerkms@zerkms.ru</a>) wrote: > Hi, > > I originally asked this question on the stackoverflow ( > <a href="https://stackoverflow.com/q/50400145/251311" rel="noreferrer" target="_blank">https://stackoverflow.com/q/50400145/251311</a>) but I feel I wouldn't get an > answer there. > > The simplified example that demonstrates a problem is > > #include <stdio.h> > > int main(void) > { > printf("foo\n"); > fflush(stdout); > getchar(); > return 0; > } > > (it's not programming language specific though - the original application I > discovered the problem with is in Go). > > If it's compiled and is run by the following unit: > > [Unit] > Description=description > After=network.target > > [Service] > ExecStart=/path/to/binary > > [Install] > WantedBy=multi-user.target > > then on `systemctl restart` `journald` would in most cases not get the > `foo\n` output, while `syslog` successfully writes it down to the > `/var/log/syslog`. > > Here is the example of the `journald` output for the service: > > May 18 08:30:38 hostname systemd[1]: Stopped servicename > May 18 08:30:38 hostname systemd[1]: Started servicename > May 18 08:30:38 hostname systemd[1]: Stopped servicename > May 18 08:30:38 hostname systemd[1]: Started servicename > May 18 08:30:38 hostname servicename[7701]: foo > May 18 08:30:41 hostname systemd[1]: Stopped servicename > May 18 08:30:41 hostname systemd[1]: Started servicename > May 18 08:30:46 hostname systemd[1]: Stopped servicename > May 18 08:30:46 hostname systemd[1]: Started servicename > > and then the corresponding part of the `/var/log/syslog`: > > May 18 08:30:38 hostname systemd[1]: Stopped servicename. > May 18 08:30:38 hostname systemd[1]: Started servicename. > May 18 08:30:38 hostname servicename[7682]: foo > May 18 08:30:38 hostname systemd[1]: Stopped servicename. > May 18 08:30:38 hostname systemd[1]: Started servicename. > May 18 08:30:38 hostname servicename[7701]: foo > May 18 08:30:41 hostname systemd[1]: Stopped servicename. > May 18 08:30:41 hostname systemd[1]: Started servicename. > May 18 08:30:41 hostname servicename[7720]: foo > May 18 08:30:46 hostname systemd[1]: Stopped servicename. > May 18 08:30:46 hostname systemd[1]: Started servicename. > May 18 08:30:46 hostname servicename[7739]: foo > > What would be a way to **guarantee** journald received the `stdout`, if any? </div></div>It is guaranteed that it wil get the data. However, you are probably running into this: <a href="https://github.com/systemd/systemd/issues/2913" rel="noreferrer" target="_blank">https://github.com/systemd/systemd/issues/2913</a> i.e. stuff that is logged right before a process exits might miss the metadata we need for "-u" to work correctly. The message itself is not lost though, you should see it if you drop the "-u"... The situation is improved a ton with systemd 235 where metadata is cached in journald if possible, which increases the chance we wil have the metadata around even if the sending process is already gone. It fixes most common cases, but there are still some cases around which aren't covered by that. hence, make sure to run 235 (or some release that backported the caching). Lennart -- Lennart Poettering, Red Hat </blockquote></div> <div> </div>-- <div class="gmail_signature" data-smartmail="gmail_signature">With best regards, Ivan Kurnosov</div> </div>