On Tue, Sep 02, 2014 at 12:21:20PM +0200, Vladimír Čunát wrote: > On 09/02/2014 11:36 AM, Eelco Dolstra wrote: > >However, there is a long-standing issue with stdout/stderr logging: if the > >process dies before systemd has had a change to process its log message, then > >systemd may not be able to figure out the unit corresponding to the message, > >so > >it may be lost (or not attributed to the unit). See > > > > http://lists.freedesktop.org/archives/systemd-devel/2012-June/005590.html > > I didn't investigate details, but I suspect that losing the last > parts of logs could be caused by the fact that stdout is buffered by > default (and stderr is not). I don't know if the contents of the > buffer residing in the crashed process can be well recovered. > Perhaps they have some hack to force the flush externally, I don't > know. > > See e.g. > http://stackoverflow.com/questions/7876660/how-to-turn-off-buffering-of-stdout-in-c
If that is the case, maybe we can run the daemons with "stdbuf -o 0 daemon". I understood from previous letters that *stderr* messages were lost, not *stdout*, but that maybe was a reference to upstart. The systemd archive reference mentions some unlink between unit and log, which would be a completely differen case. Regards, Lluís. _______________________________________________ nix-dev mailing list nix-dev@lists.science.uu.nl http://lists.science.uu.nl/mailman/listinfo/nix-dev