Table of Contents

Perfmetrics configuration for Zope and Plone

zperfmetrics works like and depends on perfmetrics, additional it:

• offers a ZConfig configuration for the statsd,
• uses the current requests path in the statd dotted path prefix,
• also provides a patch to measure plone.app.theming for diazo compile and diazo transform if Plone is available,

Installation

First get your statsd configured properly. This is out of scope of this module.

Depend on this module in your packages setup.py.

Given you use zc.builodut to set up your Zope2 or Plone, add to your [instance] section or the section with recipe = plone.recipe.zope2instance the lines:

[instance01]
recipe = plone.recipe.zope2instance
...
zope-conf-imports =
    zperfmetrics
zope-conf-additional =
    <perfmetrics>
        uri statsd://localhost:8125
        before MyFancyProject
        hostname on
        virtualhost on
        after ${:_buildout_section_name_}
    </perfmetrics>
...

Given this runs on a host called w-plone1, this will result in a prefix MyFancyProject.w-plone1.instance01.

uri

Full URI of statd.

before

Prefix path before the hostname.

hostname

Get hostname and insert into prefix. (Boolean: on or off)

virtualhost

Get virtualhost and use in ZPerfmetrics after the static prefix. (Boolean: on or off)

after

Prefix path after the hostname.

Usage

You need to decorate your code or use the with statement to measure a code block.

Usage:

from zperfmetrics import ZMetric
from zperfmetrics import zmetric
from zperfmetrics import zmetricmethod

@zmetric
def some_function():
    # whole functions timing and call counts will be measured
    pass

@ZMetric(stat='a_custom_prefix')
def some_other_function():
    # set a custom prefix instead of module path and fucntion name.
    pass

class Foo(object):

    @zmetricmethod
    def some_method(self):
        # whole methods timing and call counts will be measured
        pass

    @ZMetric(method=True, timing=False):
    def some_counted_method(self):
        # whole methods number of calls will be measured, but no times
        pass

    @ZMetric(method=True, count=False):
    def some_timed_method(self):
        # whole methods timing be measured, but no counts
        pass

    def some_method_partly_measured(self):
        # do something here you dont want to measure
        with ZMetric(stat='part_of_other_method_time'):
            # do something to measure
            # call counts and timing will be measured
            pass
        # do something here you dont want to measure

    @ZMetric(method=True, rate=0.5):
    def some_random_recorded_method(self):
        # randomly record 50% of the calls.
        pass

Request Lifecycle Integration

All ZPerfmetrics with a request passed in are considered to be under the request_lifecycle section.

All metrics in here are build like: $PREFIX.request_lifecycle.[$VIRTUAL_HOST].$PATH.*.

Zope

This package provides subscribers to measure the time a request takes, including some points in time between.

These subscribers are loaded via zcml and are logging under publish.*:

publish.traversal

time needed from publication start until traversal is finished.

publish.rendering

time needed from traversal end until before commit begin.

This value is a bit fuzzy and should be taken with a grain of salt, because there can be other subscribers to this event which take their time. Since the order of execution of the subscribers is not defined, processing may happen after this measurement

If plone tranformchain is active, the rendering time is before transforms are starting.

publish.finalize

time needed from rendering end (or transform end if plone.transformchain is active) until database commit is done.

publish_all

whole time needed from publication start until request is completly processed.

Plone

Installing this package in Plone by depending on zperfmetrics[plone] forces usage of plone.transformchain version 1.2 or newer.

First, publish.rendering gets less fuzzy because the expensive transforms (also subscribers to publish.beforecommit) are all done.

Then it introduces new measurements related to plone.transformchain:

publish.transform_all

time needed for all transforms in the plone.transformchain. This usually includes Diazo.

publish_transform_single.${ORDER}-${TRANSFORMNAME}

time needed for a specific single transform. transforms are ordered and named, both are replaced.

This package patches:

diazo.setup metric

plone.app.theming.transform.ThemeTransform.setupTransform is patched as a basic (path-less) perfmetrics Metric. The setup of the transform happens once on startup and is the time needed to create the Diazo xslt from its rules.xml, index.html and related files.

Statd, Graphite & Grafana in Docker

Setting up Statsd, Graphite and Grafana can be complex. For local testing - but also for production environments - firing up some docker containers comes in handy.

A very minimal version of such a Statd, Graphite & Grafana in Docker setup (original) helps getting things initially up and running. Install Docker and install docker-compose (just pip install docker-compose), then just clone the repository and in its docker directory run docker-compose up -d.

Let Zperfmetrics point to uri statsd://localhost:8125 and collect some data. Open Grafana in your browser at http://localhost:3000.

Go first to Grafana Getting Started, the 10 minute video really helps to find the hidden part of its UI.

Source Code

The sources are in a GIT DVCS with its main branches at github.

We'd be happy to see many branches, forks and pull-requests to make zperfmetrics even better.

Contributors

• Jens W. Klein <jens@bluedynamics.com>
• Zalán Somogyváry