Information on Samba QA process, Regression testsuites

Martin Pool mbp at samba.org
Mon Mar 10 02:59:56 GMT 2003


On  9 Mar 2003, Scott Prive <scottprive at earthlink.net> wrote:

> Thanks Tim. So far, most of the tests I've seen in CVS are unit- or
> small-tests, but as a software tester I'm more interested in integration
> tests... which it sounds like you and Martin are working on.

Our framework is intended to eventually cover everything up to about
the system-test level, where you will run Samba tests against a number
of NT or Samba servers to check network interoperation.

A more-developed test suite using the same framework and design
principles is used in distcc:

  http://cvs.samba.org/cgi-bin/cvsweb/distcc/test/

It is partially inspired by the very good Subversion test suite.

Any concrete practical suggestions you have on the design would be
welcome.

> If it's less than 3MB, please email what you have.

(Sent by private mail)

This is in a private samba cvs directory; we'll probably make it
public in the future.

The design notes (not completely up-to-date) are attached.

-- 
Martin 
-------------- next part --------------
Notes on Samba Testing Framework for Unittests
----------------------------------------------

This is to be read after reading the notes.txt from comfychair.  I'm
proposing a slightly more concrete description of what's described
there.

The model of having tests require named resources looks useful for
incorporation into a framework that can be run by many people in
widely different environments.

Some possible environments for running the test framework in are:

    - Casual downloader of Samba compiling from source and just wants
      to run 'make check'.  May only have one Unix machine and a
      handful of clients.

    - Samba team member with access to a small number of other
      machines or VMware sessions.

    - PSA developer who may not have intimate knowledge of Samba
      internals and is only interested in testing against the PSA.

    - Non-team hacker wanting to run test suite after making small
      hacks. 

    - Build farm environment (loaner machine with no physical access
      or root privilege).

    - HP BAT.

Developers in most of these environments are also potential test case
authors.  It should be easy for people unfamiliar with the framework
to write new tests and have them work.  We should provide examples and
the existing tests should well written and understandable.

Different types of tests:

    - Tests that check Samba internals and link against
      libbigballofmud.so.  For example:

        - Upper/lowercase string functions
        - user_in_list() for large lists

    - Tests that use the Samba Python extensions.

    - Tests that execute Samba command line programs, for example
      smbpasswd. 

    - Tests that require other resources on the network such as domain
      controllers or PSAs.

    - Tests that are performed on the documentation or the source code
      such as:

        - grep for common spelling mistakes made by abartlet (-:
        - grep for company copyright (IBM, HP)
                         
    - Link to other existing testing frameworks (smbtorture,
      abartlet's bash based build farm tests)

I propose a TestResourceManager which would be instantiated by a test
case.  The test case would require("resourcename") as part of its
constructor and raise a comfychair.NotRun exception if the resource
was not present.  A TestResource class could be defined which could
read a configuration file or examine a environment variable and
register a resource only if some condition was satisfied.

It would be nice to be able to completely separate the PSA testing
from the test framework.  This would entail being able to define test
resources dynamically, possibly with a plugin type system.

class TestResourceManager:
    def __init__(self, name):
        self.resources = {}

    def register(self, resource):
        name = resource.name()
        if self.resources.has_key(name):
            raise "Test manager already has resource %s" % name
        self.resources[name] = resource

    def require(self, resource_name):
        if not self.resources.has_key(resource_name):
            raise "Test manager does not have resources %s" % resource_name

class TestResource:
    def __init__(self, name):
        self.name = name

    def name(self):
        return self.name

import os

trm = TestResourceManager()

if os.getuid() == 0:
    trm.register(TestResource("root"))

A config-o-matic Python module can take a list of machines and
administrator%password entries and classify them by operating system
version and service pack.  These resources would be registered with
the TestResourceManager.

Some random thoughts about named resources for network servers:

require("nt4.sp3")
require("nt4.domaincontroller")
require("psa")

Some kind of format for location of passwords, libraries:

require("exec(smbpasswd)")
require("lib(bigballofmud)")

maybe require("exec.smbpasswd") looks nicer...

The require() function could return a dictionary of configuration
information or some handle to fetch dynamic information on.  We may
need to create and destroy extra users or print queues.  How to manage
cleanup of dynamic resources?

Requirements for running stf:

    - Python, obviously
    - Samba python extensions
-------------- next part --------------
		-*- indented-text -*-

(set lotus no)



Notes on using comfychair with Samba (samba testing framework units):

The tests need to rely on some external resources, such as 

If suitable resources are not available, need to skip particular
tests.  Must include a message indicating what resources would be
needed to run that test.  (e.g. must be root.)

We want to be able to select and run particular subsets of tests, such
as "all winbind tests".

We want to keep the number of configurable parameters down as much as
possible, to make it easy on people running the tests.

Wherever possible, the tests should set up their preconditions, but a
few basic resources need to be provided by the people running the
tests.  So for example, rather than asking the user for the name of a
non-root user, we should give the tests the administrator name and
password, and it can create a new user to use.

This makes it simpler to get the tests running, and possible also
makes them more reproducible.

In the future, rather than using NT machines provided by the test
person, we might have a way to drive VMWare non-persistent sessions,
to make tests even more tightly controlled.


Another design question is how to communicate this information to the
tests.  If there's a lot of settings, then it might need to be stored
in a configuration file.  

However, if we succeed in cutting down the number of parameters, then
it might be straightforward to pass the information on the command
line or in an environment variable.  

Environment variables are probably better because they can't be seen
by other users, and they are more easily passed down through an
invocation of "make check".




More information about the samba-technical mailing list