shared logs and multiple configurations

[Chris Curvey]

Hi all,

I've apparently tied myself up a bit using the logging package.

In my project, I have a core set of model and controller classes that
set up their logging using logging.fileCon fig(). So far, so good.

But I use these core classes from a bunch of different places.
Sometimes from within a CherryPy server, sometimes batch jobs run from
a command line, sometimes from Windows services (also written in
Python). Ah, the beauty of OO design.

I would like to have a series of logs: a shared log for the core
classes (which would log messages from the core classes, regardless of
how they were invoked), a different one for CherryPy related stuff, a
third one for batch jobs run from the command line, a fourth for the
windows services. I'd alsolike to have the logs rotate automatically
using the TimedRotatingFi leHandler classes. Oh, and I'd like to do as
much setup thru the configuration files and logging.fileCon fig() as
possible.

So my questions are:

1) Can I share log files between processes? Log messages seem to get
written, but when it comes time to roll over a log, I generally get a
"IO operation on closed file" error. (I'm thinking that I may have to
set up a logging service if I want to do this, but I'm hoping there's a
simpler way.)

2) When I want to use logging from within a multi-threaded server
(built using the Threading module), do I create one global logger
reference, and then have each thread refer to that instance, or do I
have each thread grab it's own reference using getLogger()?

3) It seems like you can't make multiple calls to logging.fileCon fig().
If I call this twice with different ini files, it appears that any
handlers set up in the first call are silently dropped when the second
call is made. (Strangely, log messages sent to loggers set up in the
first call seem to just be silently dropped.) Am I diagnosing this
behavior properly? Is there a way
to work around it?

Platform is Python 2.4.2 on Windows.

Many thanks!

Chris

[Vinay Sajip]

Re: shared logs and multiple configurations

Chris Curvey wrote:
[color=blue]
> 1) Can I share log files between processes? Log messages seem to get
> written, but when it comes time to roll over a log, I generally get a
> "IO operation on closed file" error. (I'm thinking that I may have to
> set up a logging service if I want to do this, but I'm hoping there's a
> simpler way.)[/color]

For multi-process logging, I would advise that all processes use a
SocketHandler to send their events to a listener. You can use the
example code in the documentation at

http://docs.python.org/lib/network-logging.html

to get a working example of a server which listens on a socket and
processes logging events received via the socket.
[color=blue]
> 2) When I want to use logging from within a multi-threaded server
> (built using the Threading module), do I create one global logger
> reference, and then have each thread refer to that instance, or do I
> have each thread grab it's own reference using getLogger()?[/color]

Logger instances are singletons, so multiple calls to getLogger() with
a given name will always return the same instance. There's usually no
need to create a single global reference, though you could for example
create a logger as a class attribute in your Thread subclass (assuming
you're subclassing Thread), or an appopriate class which implements the
functionality which runs in different threads.
[color=blue]
> 3) It seems like you can't make multiple calls to logging.fileCon fig().
> If I call this twice with different ini files, it appears that any
> handlers set up in the first call are silently dropped when the second
> call is made. (Strangely, log messages sent to loggers set up in the
> first call seem to just be silently dropped.) Am I diagnosing this
> behavior properly? Is there a way
> to work around it?[/color]

fileConfig() is not meant to do incremental configurations, so it does
its best to disable any previous configuration.I t does this by clearing
out the list of previously defined handlers and disabling any
previously defined loggers. You can find the code easily enough and
comment it out if you wish not to disable old loggers.

Regards,

Vinay Sajip