dm.zope.profile 1.0

Profiling Zope applications (for Zope >= 4)

The package allows the profiling of Zope applications. It is the successor of Products.ZopeProfiler for Zope>=4. It uses Python’s standard profiling support from its runtime library modules cProfile and pstats and integrates it with Zope. Read the corresponding documentation to learn how to interact with and interpret the collected profiling information.

Installation

You can install this package in the same way as you have installed Zope, i.e. either via pip or via buildout.

Ensure that its configure.zcml gets executed on startup. It defines request subscribers to collect profiling data and the page @@profile to manage profiling or present collected data.

Profiling

When you start a Zope instance, profiling is disabled. You can enable the collection of profiling data via the page @@profile. This page also allows you to configure the profiling and present the collected data in various alternative ways.

The configuration is split into 2 parts: the Profiler part controls the collection of profiling base data, the Statistics part controls how this information is presented.

The Profiler configuration allows you to specify which timer (real time or CPU time) is used for profiling and which requests should be profiled. For the latter task, you can define 2 collections of regular expressions: include and exclude. A request is not profiled if any regular expression in exclude matches; if include is not empty, then a request is only profiled if it matches at least one of those. Matching is always performed in search mode on the request’s PATH_INFO. When you change the timer, the collected profiling data is cleared.

The Statistics configuration essentially provides parameters for Python’s pstats. Especially, you can select various views (stats, caller, callee), sort according to various criteria (e.g. cumulative, internal, …), strip directory information and control the amount of data shown. Read the Python documentation for pstats to get details.

Note that profiling usually does not see thread switches. This is always the case when you use a real time timer and can also be the case with a CPU time timer (e.g. for Python 2 or on Windows). In those cases, time spend in a foreign thread may be wrongly accounted for a profiled request. You may consider configuring Zope to use a single worker thread to reduce this problem.

Note that both configuration and collected profiling data is maintained in process memory; i.e. in a multi instance Zope installation, each instance has its own private data.

Note that profiling reveals important implementation details. You should take appropriate measures to avoid that it can be accessed by a hacker. One option is to make it only available in development installations behind a firewall.

Configuration

dm.zope.profile maintains its configuration in RAM. As a consequence, configuration changes are lost when the process is shut down. You can define the initial configuration on start up via ZCML. For this purpose, the package’s meta.zcml defines the directive config in namespace http://namespaces.zope.org/dm.zope.profile.

The configuration uses options with either simple values (timer, type and strip_dirs) or collection values (include, exclude, sort_by, restrictions). Options with simple values are specified via attributes of the config directive. Individual elements of collection values are specified via element subdirectives. Those subdirectives specify the option they provide a partial value for via their name attribute and their contribution via their value attribute.

example_config.zcml contains an example. It recreates the code defined default conflguration (and thus serves only as an a demontration).

Note: Currently, you cannot omit options in order to inherit them from the the code defined default configuration: if the config directive is used, then the code defined default configuration becomes completely irrelevant. This might change in the future.