On Mon, Feb 23, 2015 at 1:31 PM, <cov...@ccs.covici.com> wrote: > > Marc Joliet <mar...@gmx.de> wrote: > > > Am Mon, 23 Feb 2015 12:10:18 -0600 > > schrieb Canek Peláez Valdés <can...@gmail.com>: > > > > > On Mon, Feb 23, 2015 at 11:49 AM, <cov...@ccs.covici.com> wrote: > > > > > > > > Canek Peláez Valdés <can...@gmail.com> wrote: > > > > > > > > > On Mon, Feb 23, 2015 at 3:41 AM, <cov...@ccs.covici.com> wrote: > > > > > > > > > > > > Marc Joliet <mar...@gmx.de> wrote: > > > > > > > > > > > > > Am Mon, 23 Feb 2015 00:41:50 +0100 > > > > > > > schrieb lee <l...@yagibdah.de>: > > > > > > > > > > > > > > > Neil Bothwick <n...@digimed.co.uk> writes: > > > > > > > > > > > > > > > > > On Wed, 18 Feb 2015 21:49:54 +0100, lee wrote: > > > > > > > > > > > > > > > > > >> > I wonder if the OP is using systemd and trying to read the > > > > > journal > > > > > > > > >> > files? > > > > > > > > >> > > > > > > > > >> Nooo, I hate systemd ... > > > > > > > > >> > > > > > > > > >> What good are log files you can't read? > > > > > > > > > > > > > > > > > > You can't read syslog-ng log files without some reading > > > software, > > > > > usually > > > > > > > > > a combination of cat, grep and less. systemd does it all with > > > > > journalctl. > > > > > > > > > > > > > > > > > > There are good reasons to not use systemd, this isn't one of > > > them. > > > > > > > > > > > > > > > > To me it is one of the good reasons, and an important one. Plain > > > text > > > > > > > > can usually always be read without further ado, be it from rescue > > > > > > > > systems you booted or with software available on different > > > operating > > > > > > > > systems. It can be also be processed with scripts and sent as > > > email. > > > > > > > > You can probably even read it on your cell phone. You can still > > > read > > > > > > > > log files that were created 20 years ago when they are plain text. > > > > > > > > > > > > > > > > Can you do all that with the binary files created by systemd? I > > > can't > > > > > > > > even read them on a working system. > > > > > > > > > > > > > > What Canek and Rich already said is good, but I'll just add this: > > > it's > > > > > not like > > > > > > > you can't run a classic syslog implementation alongside the systemd > > > > > journal. > > > > > > > On my systems, by *default*, syslog-ng kept working as usual, > > > getting > > > > > the logs > > > > > > > from the systemd journal. If you want to go further, you can even > > > > > configure > > > > > > > the journal to not store logs permanently, so that you *only* end up > > > > > with > > > > > > > plain-text logs on your system (Duncan on gentoo-amd64 went this > > > way). > > > > > > > > > > > > > > So no, the format that the systemd journal uses is most decidedly > > > *not* > > > > > a reason > > > > > > > against using systemd. > > > > > > > > > > > > > > Personally, I'm probably going to uninstall syslog-ng, because > > > > > journalctl is > > > > > > > *such* a nice way to read logs, so why run something whose output > > > I'll > > > > > never > > > > > > > read again? I recommend reading > > > > > > > http://0pointer.net/blog/projects/journalctl.html for examples of > > > the > > > > > kind of > > > > > > > stuff you can do that would be cumbersome, if not *impossible* with > > > > > regular > > > > > > > syslog. > > > > > > > > > > > > Except that I get lots of messages about the system journal missing > > > > > > messages when forwarding to syslog, so how can I make sure this does > > > not > > > > > > happening? > > > > > > > > > > Could you please show those messages? systemd sends *everything* to the > > > > > journal, and then the journal (optionally) can send it too to a regular > > > > > syslog. In that sense, it's impossible for the journal to miss any > > > message. > > > > > > > > > > The only way in which the journal could miss messages is at very early > > > boot > > > > > stages; but with a proper initramfs (like the ones generated with > > > dracut), > > > > > even those get caught. You get to put an instance of systemd and the > > > > > journal inside the initramfs, and so it's available almost from the > > > > > beginning. > > > > > > > > > > And if you use gummiboot, then you can even log from the moment the UEFI > > > > > firmware comes to life. > > > > > > > > So, I get lots of messages in my regular syslog-ng /var/log/messages > > > > like the following: > > > > Feb 23 12:47:52 ccs.covici.com systemd-journal[715]: Forwarding to > > > > syslog missed 15 messages. > > > > > > > > So, I saw a post on Google to up the queue length, and I uped it to 200, > > > > but no joy, still get the messages like the one above. > > > > > > Are you using the unit file provided by syslog-ng (systemd-delta doesn't > > > mention syslog)? Also, is /etc/systemd/system/syslog.service is a link > > > to /usr/lib/systemd/system/syslog-ng.service? > > > > > > I do, and I don't get any of those messages. I use the default journal > > > configuration. According to [1], this should be fixed. > > > > I remember getting a small number of messages like that, too, on my laptop. > > However, it's at the university, so I can't check now to see what types of > > messages were missed (if any; if I understand [1] correctly, those messages are > > most likely bogus?). > > > > But yeah, that's any idea, Covici: see what's in /var/log/messages, compare that > > to the journalctl output, and check if any messages were actually missed ("diff > > -U" might be of help here). And if/once you did that, what kinds of messages > > were missed, if any? If those messages really are bogus, you shouldn't see any > > differences between the two. > > > > > Regards. > > > > > > https://github.com/balabit/syslog-ng/issues/314 > > > > Note that that fix would only be in the ~arch version of syslog-ng, the current > > stable version (3.4.8) is a few months too old. > > I am up to 3.6 something, so the fix should be there. But my unit file > is different, so that remains to check.
I would try the provided unit file. It seems that the only difference with yours is that it doesn't comment the Restart=on-failure line, and that it has StandardOutput=null. I think the general idea is always to use upstream's unit files. They write the software, supposedly they should know better. Regards. -- Canek Peláez Valdés Profesor de asignatura, Facultad de Ciencias Universidad Nacional Autónoma de México
