On Wed, Oct 30, 2013 at 10:42:48AM +0000, Santhosh Edukulla wrote: > > ConfigManager is as synonymous to any other configuration facility > provided for products. Currently, we provide configuration as part > of some file say json\cfg and provide the command line option to use > that. Now, there are places where the hard coding specific to > setup\data is done inside of test data. I have given one simple > example related to ldap search and replace. In order to replace > these issues, we can use the ConfigManager facility added. We can > now extend it to some more usecases if required. It has now some > provision to read the config, abstract to users and obtain to test > features as easy to use dictionary.
Sure - I understand the intent. This cleanup is good. Can you show how the ldap test changes with the introduction of ConfigManager? I guess that will make it clear how to use this. It should also serve as a test for your changes. > > The new facility added handles all that mentioned in doc strings. It > has a simple interface to retrieve the configuration provided and > users can use it easily inside the test features as a dictionary. > With this, all configuration pertaining to run will be at one single > place and we provide a simple usage of this through dictionary. It > doesn't have any setters for that matter. With this, all hard coding > will be moved to configuration and users use the dynamic dictionary > created based upon config and is more driven through config. We can > remove current not a cleaner way of search and replace with this. I guess I don't understand how the config that we inject into the test right now is any different from the ConfigManager interface proposed here. Example, to reference the pod information of a deployment for example one would do config.zones[0].pods[0] now. This config attribute is part of the test already. I'm sure this can be more intuitive to access using a dictionary approach. But the structure and depth of the keys of that dictionary is going to be in the hands of the user? > > Users mention all >parameters related to their test in config file, and they have a dictionary of >the parsed config inside of their tests to use this data. The user don't need >to hard code this. Of a test as we are doing currently. On a given setup they >can use a different config and it works to work with those ips set up >infrastructure dynamically. If I have a ldap server ip different from other, >the code still works, we just need to change the config. And we already do this: the deployment config within Citrix labs is vastly different from the one used on jenkins.bacd.org. So how is this more automatic when using the configmanager? One would still need to go change the json config manually, correct? Sounds like we are moving external elements into the config file from the test? which is how it was always intended. May be we just have a poorly written test? > > The features, where this hard coding is done will now be replaced with this > usage of moving those hard coded strings to configuration file and replace > those with config dictionary values, which will be filled in dynamically. > > What else you see is missing? Let me know and I can enhance it to add. > So the ldap example if you could change using the configManager would be useful to see how the change would be helpful. > > Santhosh > > -----Original Message----- > From: Prasanna Santhanam [mailto:t...@apache.org] > Sent: Wednesday, October 30, 2013 3:42 PM > To: Santhosh Edukulla > Cc: Prasanna Santhanam; cloudstack > Subject: Re: Review Request 15021: Fixed Bug: 4899 : Added Configuration > Support to Marvin. > > I've given some inline comments in the patch. The ConfigManager looks too > simple perhaps because it's a WIP? At least the goals as you've written in > your docstring seem to talk about more than what's provided in the module. > > On Tue, Oct 29, 2013 at 06:00:14PM -0000, Santhosh Edukulla wrote: > > ------- > > > > Added Configuration Support to Marvin. > > > > 1. It provides the basic configuration facilities to marvin. > > > How will this be different from the current configGenerator module? > The configGenerator could read and write a structured deployment > configuration. Is the goal of this configGenerator different? > > > 2. User can just add configuration files for his tests, deployment > > etc, under one config folder before running their tests. > > cs/tools/marvin/marvin/config. > > This seems restrictive because previously one could have their configs placed > anywhere. what would be the advantage of placing all configs in > tools/marvin/marvin/config? When packaged and deployed this path is going to > change to /usr/local/lib/python/site-packages/marvin/config. > Do we want that? > > > They can remove all hard coded values from code and > > separate > > it out as config at this location. > > How would the ConfigParser know where to find a specific section and at what > depth? Is there a structure? Or do you leave that up to the test author? I'd > rather not do the latter as things go wild with too many different ways to do > the same thing. > > > Either add this to the existing setup.cfg as separate > > section > > or add new configuration. > > 3. This will thus removes hard coded tests and separate > > data from tests. > > Really? I think a lot of data is not specific to deployment configuration > today in tests. If we unify all data into the deployment json it's going to > become a mess to manage, edit and control when running tests. > > > > > 4. This API is provided as an additional facility under > > cloudstackTestClient and users can get the > > configuration object as similar to apiclient,dbconnection > > etc to drive their test. > > marvinPlugin.py,L#158 - the config is already injected into each test. > I like it that it's part of the testClient, wil you be providing setter > properties to it if it's part of the testClient via ConfigManager. Test case > authors might be tempted to change configuration on the fly which might > affect downstream tests in the same suite? > > > > > 5. They just add their configuration for a test, > > setup etc,at one single place under configuration dir > > and use "getConfigParser" API of cloudstackTestClient > > It will give them "configObj".They can either pass their > > own > > config file for parsing to "getConfig" or it will use > > default config file @ config/setup.cfg. > > 6. They will then get the dictionary of parsed > > configuration and can use it further to drive their tests > > or > > config drive > > 7. Test features, can drive their setups thus removing hard coded > > values. Configuration default file will be under config and as > > setup.cfg. > > 8. Users can use their own configuration file passed to > > "getConfig" API,once configObj is returned. > > > > Another such case where we are using sed or bash script is in between a > > build run for replacing hard coded ldap ip for region\setup specific. We > > can now change all parameters before run under configuration, the test > > features will use configuration object and thus values, rather hard coded > > strings which avoids replacement through shell script. > > > > Also, did few naming convention changes. Its better to follow uniform > > naming conventions. Currently, wherever iam seeing a specific notation not > > followed. We are incorporating those changes. > > > > ToDo: > > clean up of current config at places, making configuration required for > > marvin\tests unified and available at one place and clean up of files\code > > related to it. > > > > > > Diffs > > ----- > > > > tools/marvin/marvin/cloudstackTestClient.py be93f35 > > tools/marvin/marvin/config/setup.cfg PRE-CREATION > > tools/marvin/marvin/configGenerator.py 0cfad30 > > tools/marvin/marvin/integration/lib/utils.py 7d662af > > > > Diff: https://reviews.apache.org/r/15021/diff/ > > > > > > Testing > > ------- > > > > Basic check to retrieve configuration values is done. > > > > > > Thanks, > > > > Santhosh Edukulla > > > > -- > Prasanna., > > ------------------------ > Powered by BigRock.com -- Prasanna., ------------------------ Powered by BigRock.com
