~bzr-pqm/bzr/bzr.dev

Profiling

=========

Using profilers

---------------

Bazaar has some built-in support for collecting and saving profiling

information. In the simpliest case, the ``--lsprof`` option can be used as

shown below::

  bzr --lsprof ...

This will dump the profiling information to stdout before exiting.

Alternatively, the ``--lsprof-file`` option can be used to specify a filename

to save the profiling data into to. By default, profiling data saved to a

file is a pickled Python object making it possible to reload the data and

do with it what you will. For convenience though:

* if the filename ends in ``.txt``, it will be dumped in a text format.

* if the filename either starts with ``callgrind.out`` or ends with

  ``.callgrind``, it will be converted to a format loadable by the

  KCacheGrind visualization tool.

Note that KCacheGrind's Open Dialog has a default filter than only shows

files starting with ``callgrind.out`` so the longer filename is usually

preferable. Here is an example of how to use the ``--lsprof-file`` option

in combination with KCacheGrind to visualize what the ``status`` command

is doing::

  bzr --lsprof-file callgrind.out.st001 status

  kcachegrind callgrind.out.st001 &

.. Note:: bzr also has a ``--profile`` option that uses the hotshot profiler

   instead of the lsprof profiler. The hotshot profiler can be useful

   though the lsprof one is generally recommended. See

   http://docs.python.org/lib/node795.html.

Note that to use ``--lsprof`` you must install the lsprof module, which you

can get with::

  svn co http://codespeak.net/svn/user/arigo/hack/misc/lsprof

Profiling locks

---------------

Bazaar can log when locks are taken or released, which can help in

identifying unnecessary lock traffic.  This is activated by the ``-Dlock``

global option.

This writes messages into ``~/.bzr.log``.

At present this only logs actions relating to the on-disk lockdir.  It

doesn't describe actions on in-memory lock counters, or OS locks (which

are used for dirstate.)

Profiling HPSS Requests

-----------------------

When trying to improve network performance, it is often useful to know

what requests are being made, and how long they are taking. The ``-Dhpss``

global option will enable logging smart server requests, including the

time spent in each request.