[Linux-ha-dev] logging daemon

[Lars Marowsky-Bree]

On 2004-10-12T14:26:48, Guochun Shi <gshi at ncsa.uiuc.edu> wrote:

> 1. The log daemon will be a separate daemon (of course) which can be
> started/stopped similar to other system services The daemon uses a
> local domain socket /var/lib/log_daemon to communicate with others.

Yes.

> 2. Add an entry to ha.cf to enable/disable using the log daemon If it
> is disabled, then it is the same as now (default) If it is enabled and
> log daemon is running, we are happy to continue, but if the log daemon
> is not running, we quit.
> 
> There is a question here: if heartbeat is using log daemon and somehow
> the log daemon dies or gets killed, I assume heartbeat should quit
> since we have added the dependency on the log daemon, right? My
> concern is the more heartbeat depends on, the less robust it becomes.
> And we could switch to normal logging when the log daemon dies.

That would be my suggestion: to switch to normal logging if/while the
logging daemon is unavaiable and periodically try to reconnect.

This is also the only way how we can log during startup before we have
started the logging daemon ;-)

> 3. I suggest messages in IPC should not go through log daemon since
> log daemon itself is depending on IPC. Otherwise when something bad
> happened to log daemon,  it will keep calling itself forever.

That's a very good point. The cl_log() calls associated with the
connection to the logging daemon should not go to the log daemon.
Recursion like that would be suboptimal...

> 4. Is there anything we need do with those log messages from scripts?
> Those messages go directly to log files right now.

Now that you mention it, this has irked me for quite some time ;-) If
you could give us a commandline "cl_log" tool which we could call from
the scripts, that would be good.


Sincerely,
    Lars Marowsky-Brée <lmb at suse.de>

-- 
High Availability & Clustering
SUSE Labs, Research and Development
SUSE LINUX AG - A Novell company