Re: CSIT test tools
|
Luis Gomez <luis.gomez@...>
In general I think both ways have their pros and cons: Custom libraries are very powerful but I also feel we miss the test definition concept in Robot, i.e. all the test is defined in Python and we need to actually see the python code to understand the test case. Anyway I will check tomorrow with Carol and some other people joining the test tools discussion of every Thursday what are the Robot limitations with regards to our system test and how we can better use CSIT_test_tools scripts to work around the limitations.
BR/Luis
From: integration-dev-bounces@... [mailto:integration-dev-bounces@...]
On Behalf Of Luis Gomez
Sent: Wednesday, November 06, 2013 7:44 PM To: Baohua Yang Cc: integration-dev@... Subject: Re: [integration-dev] CSIT test tools
Hi all,
I have tried simple test case “get topology” with robot request library and it also works pretty neat:
*** Settings *** Library Collections Library RequestsLibrary
*** Testcases *** Get Request ${auth}= Create List admin admin Create Session controller http://10.125.136.52:8080 auth=${auth} ${resp}= Get controller /controller/nb/v2/topology/default Log ${resp} Should Be Equal As Strings ${resp.status_code} 200
However I am doing a simple match on “200 OK” response while Baohua python scripts we are doing much more:
def get_topology(self): """ The name is suggested to match the NB API. Show the topology >>> TopologyManager().get_topology() True """ r = super(self.__class__, self).read() if r: v = r['edgeProperties'] for i in range(0, len(r), 2): nc = v[i]['edge'] if nc[u'tailNodeConnector'] == {u'node': {u'type': u'OF', u'id': u'00:00:00:00:00:00:00:03'}, u'type': u'OF', u'id': u'3'}: if nc[u'headNodeConnector'] != {u'node': {u'type': u'OF', u'id': u'00:00:00:00:00:00:00:01'}, u'type': u'OF', u'id': u'2'}: print False elif nc[u'tailNodeConnector'] == {u'node': {u'type': u'OF', u'id': u'00:00:00:00:00:00:00:02'}, u'type': u'OF', u'id': u'3'}: if nc[u'headNodeConnector'] != {u'node': {u'type': u'OF', u'id': u'00:00:00:00:00:00:00:01'}, u'type': u'OF', u'id': u'1'}: print False elif nc[u'tailNodeConnector'] == {u'node': {u'type': u'OF', u'id': u'00:00:00:00:00:00:00:01'}, u'type': u'OF', u'id': u'1'}: if nc[u'headNodeConnector'] != {u'node': {u'type': u'OF', u'id': u'00:00:00:00:00:00:00:02'}, u'type': u'OF', u'id': u'3'}: print False elif nc[u'tailNodeConnector'] == {u'node': {u'type': u'OF', u'id': u'00:00:00:00:00:00:00:01'}, u'type': u'OF', u'id': u'2'}: if nc[u'headNodeConnector'] != {u'node': {u'type': u'OF', u'id': u'00:00:00:00:00:00:00:03'}, u'type': u'OF', u'id': u'3'}: print False else: print False print True
So I believe this is what Baohua means that we can be more flexible using our own library, right?
BR/Luis
From: Baohua Yang [mailto:yangbaohua@...]
Thanks punal and luis!
It is a good lib that uses the python requests library and provides api for the robot framework. The lib was evaluated before we try to find other solutions, and we think it a nice general packaging.
Before planning writing our tool, we also tried similar ideas but found several difficulties to handle our scenarios using such way. 1) The result validation may not be that straightforward when directly calling the NB API, because lots of responses must be filtered and analyzed . e.g., in switch manager, the list function will return large content including the timestamps, hence we cannot pre-set the standard answer before getting them in real time. Even more actions are involved, however, will reduce the robot scripts' readability. So we want keep more flexibility and scalability here, while keeping robot easily-satisfied. 2) Another problem is the dynamics. Some functionality must be evaluated combing several NB APIs, such as adding an entry and then removing it. And even more complicated scenarios in future. 3) Our further aim is to load the network config dynamically and help setup the environment automatically. So far as we know, there has to be additional code to do this, even with helps of similar libs and the robot tool.
IMHO, an appropriate solution may be keeping robot itself simple and readable, while guaranteeing enough scalability and flexibility for users and future developments.
Btw, one big advantage of the lib we can see is the session support, which can easily be implemented in current CSIT tool. However, we keep this undecided as we are still investigating whether the caching and persistence mechanism by session is necessary for our tests (all can be done in seconds currently), and what is the advantage and disadvantage of using session in the specific controller test? I would appreciate very much if anyone can drop some comments on this!
Anyone has some comments? We would like to welcome more evaluation comments ASAP. Thanks!
On Thu, Nov 7, 2013 at 6:29 AM, Luis Gomez <luis.gomez@...> wrote: Can anyone try this robotframework-requests vs CSIT-test-tool requests and see what are the differences?
I would like to try in the Ericsson Lab today if I have time….
From: Punal Patel [mailto:Punal.Patel@...]
Hi Team,
We can use robotframework-requests
https://github.com/bulkan/robotframework-requests
We are using “requests” right now in CSIT test tools but robotframework-requests will integrate CSIT test tools to robotframework.
Any thoughts, comments?
Thank You, Punal Patel
From:
integration-dev-bounces@... [mailto:integration-dev-bounces@...]
On Behalf Of Baohua Yang
Sure, luis, how about the following?
Integration -- distributions packaging test -- tool CSIT_test robot teston
On Wed, Nov 6, 2013 at 10:31 AM, Luis Gomez <luis.gomez@...> wrote: Hi Baohua,
I have not really thought much on the name but yes we need a folder for all that is test code (python, robot, etc..), so one idea is to have the following structure under integration:
Integration – distributions packaging testcode – tool robot teston
Or maybe have tool, robot and teston directly in the root, we can discuss more tomorrow during the meeting if you or someone else has more ideas.
BR/Luis
From: Baohua Yang [mailto:yangbaohua@...]
Yes, luis. Definitely. What I concern is should we put the code now or wait until the code is more stable? As Deng hui and I are still fixing one little bug. After that, we can see the basic release on the tests of the base edition. Beside, I found our repo here at https://git.opendaylight.org/gerrit/#/admin/projects/integration. Should we put the code into a new directory like "tool"? Or other comment? Thanks!
On Wed, Nov 6, 2013 at 10:12 AM, Luis Gomez <luis.gomez@...> wrote: Hi Moiz,
My only concern was that the existing verify job will build the release vehicles every time someone changes something in the test code, I do not know how much effort/time is required for the build (maybe it is very less) but it is something I would like to optimize in Jenkins like the mail I sent this morning to Andy and the rest of the team.
Said that, yes, sooner or later the scripts will be in the git so please feel free to upload them already. Also Baohua and Dengui, being the contributors for the scripts, are you familiar with pull/push process in OpenDaylight?
Thanks/Luis
From: Moiz Raja [mailto:moraja@...]
When you push code into the repository it will trigger a verify build. However if you do not have your integration tests integrated into the maven build then they will not be run.
Was your concern that pushing the code will automatically trigger a test run? Because it won't.
I suggest that we simply add whatever is in the Github repo to the ODL integration repo. I can do it if you like.
-Moiz
On Nov 5, 2013, at 5:13 PM, Luis Gomez <luis.gomez@...> wrote:
Hi Moiz, just a question: our repo as it is configured now, should not trigger a verify build job whenever someone tries to put something on it (i.e. with git push)? or is there another way to do this?
From: Moiz Raja [mailto:moraja@cisco.com]
How about just adding the python scripts into the integration repo instead of using github. It doesn't hurt.
-Moiz
On Nov 5, 2013, at 4:31 PM, Luis Gomez <luis.gomez@...> wrote:
Hi guys,
Congratulations, I downloaded the python scripts to the test tools VMs, changed the controller IP and run the system test with no issues. I saw you coded very well all the REST requests so this should be a good input for Robot framework.
BR/Luis
From: Baohua Yang [mailto:yangbaohua@gmail.com]
Hi Luis Thanks for your willing to help test the code. Currently, we've finished the tests on main functions of all modules in the base edition. We believe there's still lots of work to do and welcome for any feedback! Denghui and I have discussed a lot on the development plan, but we think there will be more power community-widely. Every member, please do not hesitate to drop lines. Thanks!
On Tue, Nov 5, 2013 at 12:04 PM, Luis Gomez <luis.gomez@...> wrote: Hi Moiz,
See my answers inline:
From: Moiz Raja [mailto:moraja@...]
Hi Guys,
A couple of questions on the System test.
a. Will the System Test be integrated with the build? The system test will not run with the build, at least the one based on Robot/Phyton. The idea is to trigger a job so that the controller VM (separated from the build server) fetches the latest release vehicle from Jenkins and runs it. After this we will trigger the test case execution in Robot.
b. What framework are we going to use to deploy the built artifacts? Is it going to be something like capistrano or custom bash scripts? The test code (Robot or Python) does not need to be built so I do not think we are going to have release artifacts as we have in Java. Instead we will have the test code stored in our git and then Robot will fetch the code from there. c. Will the python/robot tests live in the integration repository? Anything tests that I can look at? Yes, that is the idea although nothing has been uploaded to the repo yet. So far we have 2 things: Python scripts created by China team and stored in an external repo and Robot framework installed in Open Lab at Ericsson. Both are now described (Carol updated the Robot today) in https://wiki.opendaylight.org/view/CrossProject:Integration_Group:Test_Tools
I remind everybody this week is to get familiar with these tools and see how we can better use them so yes you and everybody is invited to take a look.
I will personally try to get the python scripts to work in the Open Lab at Ericsson tomorrow the latest.
BR/Luis
-Moiz
On Nov 3, 2013, at 12:16 AM, Luis Gomez <luis.gomez@...> wrote:
OK, I changed the test case order so that we start with the most basic services and end with most abstracted services :
- Switch Mgr - Topology Mgr - FRM - Statistics Mgr - Configuration - Host Tracker & Simple Forwarding - ARP Handler - Forward Manager - Container Mgr
Note that basic service does not necessarily means simple to test, so maybe I was not very precise when I said simple-to-complex. What I really meant was something like basic-to-extra functions.
I also added the required steps on every area so every test case above is self-complete. Please review the test plan and let me know if you agree with it.
As for your question, yes, we can categorize the services in different ways like for example: basic network functions (Switch Mgr, Topology Mgr, FRM, Stats Mgr), extra network functions (Host Tracker, Simple Forwarding, ARP Handler, Forward Mgr), basic node functions (Configuration, User Mgr, Connection Mgr) and extra node functions (Container Mgr, Cluster Mgr). This is just an idea but it could be more, anyway besides the classification the important is that we do not leave features/modules without test.
BR/Luis
From: Gmail [mailto:yangbaohua@gmail.com]
Sure, Luis. This is a valuable question! IMHO, the simple-to-complex order is good. However, we might also keep test cases independent from each other, i.e., each test case should be self-complete. Because, we sometime may want to test the function of individual module, instead of the entire platform. Besides, we may even categorize the tested modules based on their functions, test complexity, etc. For example, state collection modules, basic forwarding modules, QoS modules.... And each category can be tested separately. How do you think?
_______________________________________________
-- _______________________________________________
--
--
-- |
|
|