[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Cluster-devel] logsys in fenced



On Fri, 2008-06-27 at 20:27 -0700, Joel Becker wrote:
> On Fri, Jun 27, 2008 at 10:06:01AM -0500, David Teigland wrote:
> > The hope that I do still hold out for using logsys, is that we may be able
> > to develop a simple syslog-like api for it, that would either parallel or
> > sit above the plugin-oriented api that's there now.  Steve claims that the
> > syslog(3) api is fundamentally broken and didn't sound at all interested
> > in the idea, but maybe others are more open to the idea?
> 
> 	I wonder what's fundamentally broken about it.  I'm sorry that
> logsys doesn't use syslog internally - how does it log remotely, for
> example? :-)
> 
> Joel
> 

The syslog api has served thousands of applications for 20+ years.  But
the fact is every single application that is significant develops their
own logging solution because, frankly, the syslog API is a POS.

The logsys API does use the syslogd daemon's ipc connection but not the
c library api which is fundamentally flawed.  As a result we support all
of the syslogd features that system administrators have come to know and
love, and doesn't change any of those features at all!

For logging remotely, all of that is managed by syslogd.  Syslogd is a
great system.  My sole issue is with the c library api
syslog/openlog/closelog friends.  They were designed in 1970's and are
missing some key features.  This is the motivation behind logsys.
LOGSYS IS A REPLACEMENT for SYSLOG(3) NOT SYSLOGD.  It is 100%
compatabile with the syslogd daemon shipped in every single distribution
in the world.

The logsys api is not "plugin oriented"; it is designed for large
software projects that have several components each of which may desire
to log specific information at different levels, to different files, or
with different tracing levels about different subsystems.  Also the
syslog api itself is broken because is does not allow nonblocking IO
without the loss of log messages.

None of this is supported by the syslog api which is why it is
insufficient.

But I can tell I'm talking to a BRICK WALL now since I've been repeating
this for 6 months or more, and after everyone seemly agrees logsys is
the way to roll, NOW IT IS CRITICAL TO REWRITE THE API because it uses a
macro to support subsystems.  I try not to take personal insult when
these sorts of things happen but seriously, the api has been published
for over 6 months and now it is not sufficient?  Where were these posts
6 months ago?  IMO the naysayers are a day late and a dollar short.

sigh
-steve


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]