On Monday 21 November 2011 04:12:57 Jon Choate wrote:
----- Original Message -----
> From: "Doron Fediuck" <dfediuck(a)redhat.com>
> To: engine-devel(a)ovirt.org
> Cc: "Laszlo Hornyak" <lhornyak(a)redhat.com>, "Jon Choate"
> Sent: Sunday, November 20, 2011 7:35:53 AM
> Subject: Re: [Engine-devel] logging in engine
> On Friday 18 November 2011 17:36:16 Laszlo Hornyak wrote:
> > Yep, good question. Most Unix/linux sysadmins prefer syslog,
> > whatever we have on the java+jboss side.
> > ----- Original Message -----
> > > From: "Jon Choate" <jchoate(a)redhat.com>
> > > To: engine-devel(a)ovirt.org
> > > Sent: Friday, November 18, 2011 3:16:11 PM
> > > Subject: Re: [Engine-devel] logging in engine
> > >
> > > On 11/18/2011 06:42 AM, Laszlo Hornyak wrote:
> > > > Hi,
> > > >
> > > > In oVirt's logging configuration we direct org.ovirt.* loggers
> > > > to
> > > > the engine log appender, org.ovirt.engine.ui to the ui's log
> > > > and
> > > > then the rest to the server log.
> > > > The difficulties come when you start to do cross-component
> > > > debugging. E.g. you see the logs of the engine in the engine
> > > > log,
> > > > but you would also like to see the logs of the components the
> > > > engine is building upon: persistence API's (jdbc, hibernate)
> > > > network communication components (e.g. httpclient, vdsm, ssh
> > > > client), data processing API's (xml or json-parsers, apache
> > > > commons apis) and all these logs go to the generic engine log.
> > > > So
> > > > when debugging you would have to merge these logs, otherwise
> > > > you
> > > > may miss some important details.
> > > >
> > > > I believe Moti's patch
> > > >
> > > > is one step in the another (I believe more useful) direction.
> > > > But rather than specifying all the components that the engine
> > > > uses,
> > > > wouldn't it make sense to take just let all the logs flow into
> > > > a
> > > > single appender?
> > > >
> > > > What is your opinion?
> > > >
> > > > Thank you,
> > > > Laszlo
> > > > _______________________________________________
> > > > Engine-devel mailing list
> > > > Engine-devel(a)ovirt.org
> > > > http://lists.ovirt.org/mailman/listinfo/engine-devel
> > > Has any consideration been given to using the syslog appender and
> > > sending all messages there? Most data centers already have tools
> > > in
> > > place for analyzing the contents of syslog.
> 1. We have a specific log collector tool to collect the relevant logs
> and configurations
> from the setup, since this is not a simple setup. We have hosts we'd
> like to monitor,
> core dumps, etc. So this is not a standalone machine debugging /
> logging which may
> use syslog.
That sounds like we've recreated our own syslog since syslog aggregates logs from
several different applications and across different machines.
Not exactly, we're collecting configurations and coredumps as well. This is
not only logging data.
> 2. Moreover, Java's logging allows you to control log
> runtime. This is
> a feature we'd like to preserve, and not re-create all namespaces as
> filters in syslog...
Log4J has a syslog appender so switching to syslog would require no code changes other
than injecting a new appender. You would still be able to tweak verbosity at runtime.
Syslog would also allow you to tee the messages to both a syslog for forwarding to a
syslog aggregator and to an application specific log file on the local machine.
Moreover, syslog is something our users (sysadmins) are comfortable and familiar with and
have often built data center monitoring solutions around.
This may cause flooding of the syslog, filling it with messages which may mask
issues such as hardware issues, etc. So you can create a filter on message source
and direct it to a specific file, then you end up with what we already have....
To make is simple, if someone needs it, he can submit a patch with an alternate solution
and the relevant changes on /etc/syslog.conf, and those who need it can use it.
"Who's General Failure and why's he reading my disk?"