(feat) decorator labels

The setup of using @REQUEST_TIME.time() is nice. would it make sense to extend it to support labels so you could have different metrics for different method calls / http calls.

[drfloob]

Reviving this thread … it seems Guido had “a gut feeling” in disallowing the kind of decorator syntax that would allow method chaining https://mail.python.org/pipermail/python-dev/2004-August/046711.html. Mentioned in PEP 318

I hacked a fairly non-pythonic workaround:

engine_process_message_timer = REQUEST_TIME.labels(method="engine_process_message")
@engine_process_message_timer.time()
def my_func(...):
    pass

[luafran]

If you can live with this, a workaround is to instantiate a Histogram object with label values and then call time() on this object. Like this:

REQUEST_TIME = Histogram('http_request_duration_seconds', 'help', ['method', 'endpoint', 'service'])
REQUEST_TIME_GET_TEST = REQUEST_TIME.labels(method='GET', endpoint="test", service='webapp-python')

@app.route('/test/<time_ms>', method='GET')
@REQUEST_TIME_GET_TEST.time()
def get_test(time_ms='0'):
    logger.info('Processing GET /test/'+time_ms)

    time_to_sleep = int(time_ms) / float(1000)
    time.sleep(time_to_sleep)

    retdata = {
        "status": "OK"
    }

    return retdata

The details after a not so deep analysis of the issue

When you create a Histogram (or any other metric) with label names like this:

REQUEST_TIME = Histogram('http_request_duration_seconds', 'help', ['method', 'endpoint', 'service'])

Instead of a Histogram object, what you get is a _LabelWrapper object.

Since you need a Histogram object to call the time() decorator factory, the workaround is to get back a Histogram object calling labels() on the _LabelWrapper object before using the decorator.