Logging with multiple loggers/handlers

[Jeff Shannon]

I'm having some difficulty getting my logging configuration set
correctly. I'm using a config file (copied at end of post), with the
intent of setting several loggers which write to a combination of a
file, stderr (for debugging), and the NT Eventlog, but I don't seem to
be getting the right combination of handlers called.

The desired combinations are as follows:

root - Eventlog, stderr
log02 - file, stderr
log02.log03 - file, stderr
log02.log04 - Eventlog, file, stderr

Currently, logging to stderr is working for all four loggers, but only
log02 is being written to the logfile. Also, log03 is writing to the
Eventlog even though it shouldn't be reaching that handler at all.

Any suggestions as to how I could change my logconfig.ini to get the
desired effect? (I'm considering creating an 'init' logger off of the
root, which would handle the Eventlog records, thus allowing me to
propagate all the stderr records to root, but this doesn't seem like
it'd directly affect the parts I'm having problems with...)

For reference, I'm using Python 2.2.1 and v0.4.9.2 of the logging
package. (Upgrading to a newer version of Python is not preferred,
though it could be done if it'll resolve the problem, but I've gathered
that the logging package *should* work fine with Py2.2.)

Jeff Shannon
Technician/Programmer
Credit International

----- logconf.ini -------------------------
[loggers]
keys=root,log02,log03,log04

[handlers]
keys=hand01,hand02,hand03

[formatters]
keys=form01

[logger_root]
level=NOTSET
propagate=1
channel=
parent=
qualname=(root)
handlers=hand01,hand03

[logger_log02]
level=DEBUG
propagate=0
channel=log02
parent=(root)
qualname=log02
handlers=hand02,hand03

[logger_log03]
level=DEBUG
propagate=1
channel=log03
parent=log02
qualname=log02.log03
handlers=

[logger_log04]
level=DEBUG
propagate=1
channel=log04
parent=log02
qualname=log02.log04
handlers=hand01

[handler_hand01]
class=handlers.NTEventLogHandler
level=DEBUG
formatter=form01
appname=LogTest
dllname=
logtype=Application
args=('LogTest', '', 'Application')

[handler_hand02]
class=FileHandler
level=DEBUG
formatter=form01
filename=Logs\python.log
mode=a
args=('Logs\python.log', 'a')

[handler_hand03]
class=StreamHandler
level=DEBUG
formatter=form01
stream=sys.stderr
args=(sys.stderr,)

[formatter_form01]
format=%(asctime)s %(name)s %(levelname)s %(message)s
datefmt=

[Jeff Shannon]

To follow up to my own post -- I've found my problem, and it appears to
have been between my ears. ;P

I somehow managed to misinterpret the spec for logging.getLogger().
From the current docs:

getLogger([name]):
Return a logger with the specified name or, if no name is specified,
return
a logger which is the root logger of the hierarchy.

All calls to this function with a given name return the same logger
instance.
This means that logger instances never need to be passed between
different
parts of an application.

Combined with this bit from PEP 282:
Loggers are never instantiated directly. Instead, a module-level
function is used:

def getLogger(name=None): ...

If no name is specified, the root logger is returned. Otherwise,
if a logger with that name exists, it is returned. If not, a new
logger is initialized and returned. Here, "name" is synonymous
with "channel name".

And the descriptions of the config file
(http://www.red-dove.com/python_logging.html#config):

#The channel value indicates the lowest portion of the channel name
of the
#logger. For a logger called "a.b.c", this value would be "c".

All of this combined to give me the impression that the name that's
passed to getLogger() should be the "lowest portion" channel name,
rather than the fully qualified name -- i.e., that I should be using
"log03" instead of "log02.log03". As it turns out, my impression was
wrong -- it's the fully qualified name that should be used. My attempts
to use the unqualified name were creating new, unconfigured loggers
(with no handlers attached) that could only propagate records up to the
root logger. I realized this only after inspecting the log manager's
loggerDict and seeing the unconfigured "extra" loggers.

While it's fairly likely that my interpretation of the docs is not
something that many people will come up with, I think it might be
worthwhile to expand the example sections to actually show how to use
multiple heirarchical loggers. Had there been an example to follow, I
certainly wouldn't have made this mistake.

Jeff Shannon
Technician/Programmer
Credit International

[Vinay Sajip]

> Loggers are never instantiated directly. Instead, a module-level

function is used:

def getLogger(name=None): ...

If no name is specified, the root logger is returned. Otherwise,
if a logger with that name exists, it is returned. If not, a new
logger is initialized and returned. Here, "name" is synonymous
with "channel name".

And the descriptions of the config file
(http://www.red-dove.com/python_logging.html#config):

#The channel value indicates the lowest portion of the channel name
of the
#logger. For a logger called "a.b.c", this value would be "c".

All of this combined to give me the impression that the name that's
passed to getLogger() should be the "lowest portion" channel name,
rather than the fully qualified name -- i.e., that I should be using
"log03" instead of "log02.log03". As it turns out, my impression was
wrong -- it's the fully qualified name that should be used. While it's fairly likely that my interpretation of the docs is not
something that many people will come up with, I think it might be
worthwhile to expand the example sections to actually show how to use
multiple heirarchical loggers. Had there been an example to follow, I
certainly wouldn't have made this mistake.

The example in http://www.red-dove.com/python_logging.html#config does
show in bold the items used by the configuration API, and says that
the others are just used by the GUI configurator. Perhaps "channel"
was the wrong name to use for a private element in the configurator,
but it's done now.

[Jeff Shannon]

Vinay Sajip wrote:

While it's fairly likely that my interpretation of the docs is not
something that many people will come up with, I think it might be
worthwhile to expand the example sections to actually show how to use
multiple heirarchical loggers. Had there been an example to follow, I
certainly wouldn't have made this mistake.

The example in http://www.red-dove.com/python_logging.html#config does
show in bold the items used by the configuration API, and says that
the others are just used by the GUI configurator. Perhaps "channel"
was the wrong name to use for a private element in the configurator,
but it's done now.

May I suggest changing the wording of the getLogger() docs? I think
that simply adding a few words to a single sentence will clarify things
considerably.

- Here, "name" is synonymous with "channel name".
+ Here, "name" is synonymous with "fully qualified channel name".

Jeff Shannon
Technician/Programmer
Credit International