Logging, StreamHandler and standard streams

The following script, log1.py:

import logging, sys

class SingleLevelFilter(logging.Filter):
    def __init__(self, passlevel, reject):
        self.passlevel = passlevel
        self.reject = reject

    def filter(self, record):
        if self.reject:
            return (record.levelno != self.passlevel)
        else:
            return (record.levelno == self.passlevel)

h1 = logging.StreamHandler(sys.stdout)
f1 = SingleLevelFilter(logging.INFO, False)
h1.addFilter(f1)
rootLogger = logging.getLogger()
rootLogger.addHandler(h1)
h2 = logging.StreamHandler(sys.stderr)
f2 = SingleLevelFilter(logging.INFO, True)
h2.addFilter(f2)
rootLogger.addHandler(h2)
logger = logging.getLogger("my.logger")
logger.setLevel(logging.DEBUG)
logger.debug("A DEBUG message")
logger.info("An INFO message")
logger.warning("A WARNING message")
logger.error("An ERROR message")
logger.critical("A CRITICAL message")

when run, produces the following results.

C:\temp>log1.py
A DEBUG message
An INFO message
A WARNING message
An ERROR message
A CRITICAL message

As you'd expect, since on a terminal sys.stdout and sys.stderr are the same. Now, let's redirect stdout to a file, tmp:

C:\temp>log1.py >tmp
A DEBUG message
A WARNING message
An ERROR message
A CRITICAL message

So the INFO message has not been printed to the terminal - but the messages directed to sys.stderr have been printed. Let's look at what's in tmp:

C:\temp>type tmp
An INFO message

So that approach appears to do what you want.

Generally, I think it makes sense to redirect messages lower than WARNING to stdout, instead of only INFO messages.

Based on Vinay Sajip's excellent answer, I came up with this:

class MaxLevelFilter(Filter):
    '''Filters (lets through) all messages with level < LEVEL'''
    def __init__(self, level):
        self.level = level

    def filter(self, record):
        return record.levelno < self.level # "<" instead of "<=": since logger.setLevel is inclusive, this should be exclusive


MIN_LEVEL= DEBUG
#...
stdout_hdlr = StreamHandler(sys.stdout)
stderr_hdlr = StreamHandler(sys.stderr)
lower_than_warning= MaxLevelFilter(WARNING)
stdout_hdlr.addFilter( lower_than_warning )     #messages lower than WARNING go to stdout
stdout_hdlr.setLevel( MIN_LEVEL )
stderr_hdlr.setLevel( max(MIN_LEVEL, WARNING) ) #messages >= WARNING ( and >= STDOUT_LOG_LEVEL ) go to stderr
#...

Since my edit was rejected, here's my answer. @goncalopp's answer is good but doesn't stand alone or work out of the box. Here's my improved version:

import sys, logging


class LogFilter(logging.Filter):
    """Filters (lets through) all messages with level < LEVEL"""
    # http://stackoverflow.com/a/24956305/408556
    def __init__(self, level):
        self.level = level

    def filter(self, record):
        # "<" instead of "<=": since logger.setLevel is inclusive, this should
        # be exclusive
        return record.levelno < self.level

MIN_LEVEL = logging.DEBUG
stdout_hdlr = logging.StreamHandler(sys.stdout)
stderr_hdlr = logging.StreamHandler(sys.stderr)
log_filter = LogFilter(logging.WARNING)
stdout_hdlr.addFilter(log_filter)
stdout_hdlr.setLevel(MIN_LEVEL)
stderr_hdlr.setLevel(max(MIN_LEVEL, logging.WARNING))
# messages lower than WARNING go to stdout
# messages >= WARNING (and >= STDOUT_LOG_LEVEL) go to stderr

rootLogger = logging.getLogger()
rootLogger.addHandler(stdout_hdlr)
rootLogger.addHandler(stderr_hdlr)
logger = logging.getLogger(__name__)
logger.setLevel(logging.DEBUG)

# Example Usage
>>> logger.debug("A DEBUG message")
>>> logger.info("An INFO message")
>>> logger.warning("A WARNING message")
>>> logger.error("An ERROR message")
>>> logger.critical("A CRITICAL message")