Making Use of Your Fail Pics' Newfound Freedom

The benefit of letting your fail pics go is that it provides special flexibility in observing and analyzing test failures as they occur. As mentioned in the "extra bonus benefit" section of our last post, we have been having some trouble with our test execution lately. Between an unsettling number of illegitimate test failures and mysterious blank snapshots, we have found ourselves in need of a tool to help troubleshoot our test execution.

Real Time? How About Node.js...

We wanted a tool that would allow us to monitor snapshots as they occur in real time by listening to our snapshots directory and pushing newly added photos to a web client. At first, Node.js seemed like the perfect tool for the job. In fact, with a combination of Node.js, Express, Socket.IO and a directory watcher module, we had a simple implementation up and running on my local machine in little time. Unfortunately, we soon discovered that, while it worked well for watching directories on my local machine, the Node.js event based file watcher does not work with network drives. As a result, we turned back to our good friend Ruby.

Back to Ruby...

Not only would we be able to watch the network drive with Ruby, it has the added benefit of being the lingua franca of our testing team. This means that all members of the testing team can feel more comfortable reading and contributing to the project's code base in the future. The application was built using Ruby, Sinatra, and a directory watcher gem, making the server side code simple enough to place in a single server.rb file:

On the client side, we long poll the server for new events, passing in the time stamp of our latest result. As you can see, the server then hands over all events with a time stamp later than the one we passed in. The client takes that response and adds each event to the page. Now, we are able to visit the "Quality Center Snapshot Feed" web page and watch the fail pics roll in.

Future Plans

The current implementation works, but leaves plenty of room for refactoring and improvement. In the future, we plan to utilize web sockets via Event Machine, rather than long polling. In addition, we would like the application to associate each snapshot with the test machine it occurred on and which team owns the test.

Free Your QTP Test Results!

While building our web site to speed our analysis of our daily regression tests, we needed to access result data quickly.  Our failed error messages and snapshots of crashes were trapped inside QTP result logs.  As anyone who has looked into these logs knows, they were not designed for casual deciphering.

Rather than fight the HP logs, file structures, compression methods, and database, we went around it.  We started collecting our own copy of key information -- free and easy for us to use.  We are able to build this site with crash snapshots and error messages:

We developed shadow methods for collecting information.  When a test step fails or when something crashes, QTP writes information to its log, and we write a copy of the same information to our database or file server.

To support this, we created a database table that stores error messages and is keyed off cycle and run ids and wrote a ShadowFailMessage method for our QTP function library:

To make this work, we added a reference to the ShadowFailMessage subroutine after each fail (it would have been more elegant to have opened the Reporter class and added the additional functionality, but this is VBScript, not Ruby L).  This takes the error message content and writes it to a database table along with other information needed to associate the message with the right test.  The pattern looks like this:

Reporter.ReportEvent micFail, title_of_error, “Expected: “ & expected_value & “ – Actual: “ & actual_value

ShadowFailMessage title_of_error, actual_value, expected_value

Here is the code for the subroutine we referenced:

Sub ShadowFailMessage(nameOfDataValue, actual, expected)
    actual = Replace(actual,"'","''")'escape single quotes that will break the SQL query
    expected = Replace(expected,"'","''")
    On Error Resume Next
    cycle_id = QCUtil.CurrentTestSet.ID
    run_id = QCUtil.CurrentRun.ID
    If expected ="" Then
        fail_message =     Left((nameOfDataValue & ": " & "Error message: " & actual), 500)
    Else
        fail_message =     Left((nameOfDataValue & ": " & "Expected: " & expected & "; " & "Actual: " & actual), 500)
    End If
    'Write to the cfc_SNAPSHOTS table. set sn_type to 1 to indicate error message
    strSQL = "INSERT INTO test_snapshots (sn_cycle_id, sn_run_id, sn_snapshot_url, sn_type) " &_
              "VALUES (" & cycle_id &"," & run_id & ",'" & fail_message & "', 1)"
    objRecordSet = GetRecordSet(strSQL, "QC")
    On Error GoTo 0
End Sub

We also have Shadow subroutines for crash snapshots and actual/expected QTP datatables (for verifying reports and tables).

We no longer complain about the limitations of our existing tools – we own the solutions to our own problems.  In our case, we used open-course tools like Ruby and Sinatra libraries to develop our own applications for using our test results.  I have even started to like HP's problems.  They have become excuses for some fun coding -- too bad we are paying for the problems though.

What Have We Been Up To?

It has been a while since we updated this blog.  Fortunately for us, the reason we have not been writing is not because we are out of ideas, but because we have been implementing new ideas!

One problem that plagued us is how long it took us to analyze our BPT/QTP test results.  HP’s QC does some things well, but it is not exactly nimble. 

·         You have to use Internet Explorer (I die a little inside when I use IE, and there is no way that other developers will use IE),
·         Updating QC results in QC is sooo slow  (in this case, I do not die a little – instead, I want to kill a little.  Why does it take so long to update things in QC and why does it hang my browser while it makes updates?)
·         It takes too long to find the important information in the QTP result viewer
·         Licensing requirements to view results through QC

Anyway, it is clear that we cannot dramatically increase the speed of our daily work by using the QC tools as they come from HP.

We developed our own tool for analyzing BPT/QTP results.  In this view, we can see almost everything we need to understand why a test failed – at one quick glance, maybe even from my iPhone – and we can complete our analysis (without any of the HP “slows”).

There are a couple cool stories/technical solutions related to this tool:

·         Super-fast updating of QC results – even setting tests to rerun
·         Creation of shadow results – pictures and fail messages
·         Super support of analysis of tabular results
·         Prioritization of results
·         Browser plug in for QTP results viewer
·         How good/reliable is this test
·         Easily sharing test results with others on the development team

Expect detailed postings on each of these topics (for real, this time).

Our Current GUI Framework – Love and Hate

I have started a series of blog posts on how to write a test framework.  I am excited to share our story (and soon share tools that we developed).  Before getting too far into that, I would like to share some details on a significant influence on how we see things:  the dreaded bad relationship.

I have a passionate love/hate relationship with our current GUI test framework.  We use HP’s Business Process Testing (BPT) framework which uses HP’s QuickTest Pro (QTP) as the execution engine.

It doesn’t take a Googling genius to find complaints about HP tools – I have written some of those complaints.  I also know that many of the complaints are bogus.  I am sure that many are written by people who have never learned how to use the tools well and have never worked with a talented team to implement them  (sometime, I will write a proper defense of QTP) .

From my perspective, HP tools are the best thing out there for GUI testing.  And I can’t wait to move far away from them (Sorry HP)

What makes BPT so great?

• Great abstraction.  All test frameworks have some level of abstraction. BPT has several. From lowest to highest: low-level function libraries, high-level function libraries, object repository (a great xref between the properties of a GUI object and a logical name), ability to use keywords, functions, and custom scripts), business components to package all the bits into something with business user meaning, and finally the test where components and test data are assembled into tests.  As I look back over the list, I know that it is probably not so meaningful to those not on my team, but I think BPT adds better abstraction than other frameworks I have seen.
• Super easy to read tests.  If you understand our business, you will understand our tests.
• Clear separation between automation development and test building. Perhaps not unique to BPT, but it does it well.  Depending on your needs, you can build automation without understanding or caring about the use case, build tests without any automation knowledge, or do it all.
• Ideal for scaling and growing teams. Several years ago, our team was small and we used regular QTP (scripted tests with shared actions – regular stuff).  It was hard to scale beyond a couple automators – we couldn’t agree on anything.  BPT’s structure and our implementation enabled us to grow from 5 to over 20 and one office to four offices in different countries. BTW, the implementation  was no thanks to HP.  Much of what we do is not standard HP implementation (because there is no standard HP implementation!)

Where is it suckish? (i.e., what are design features for our own frameworks that our experience with HP  have influenced)

• Sucks for integration with development teams. Between licensing issues and the QC interface, this is not a tool for the full dev team.  No one outside of the test team wants anything to do with the HP tools.  For us, HP tools interfere with our goals of sharing ownership of tests.
• Execution is slow as hell. We run our full suite every night and it takes every bit of it. As an advocate of BPT, it embarrasses me to no end when someone new watches one of these dinosaur runs. As we move to new tools (and look for ways to enhance our current tools), the new goal is to regression test our apps fully in under 20 minutes (we have to trim about 9 hours from current).  HP tools are not going to help us there.  Our new approaches are and will be fully scalable and capable of cloud busts.
• Support is a nightmare. There was a period after a new HP release that I was sure I worked for HP as a tester.  The release was a nightmare, every issue I reported was a homework assignment for me, and I was helpless to solve my own problems. No more test tools that I cannot fix myself.
• “A good used Toyota” – this is what you could buy for the price of a single floating QPT license.The core parts of a good framework (for web apps) are open sourced and there for the taking. Free out of pocket. I know that time is not free, but it is much more satisfying to spend time writing your own tools rather than burning time with a closed system that you can't directly influence.

Love and hate.  Not a perfect relationship, but we learned a lot from it and it’s time to move on.  Our HP experience has and will and will guide us as we move on with services testing and GUI testing frameworks.

Automatic Rerun: QualityCenter Workflow

For those of you who haven't been keeping up with the View into the Automation Lab story,  we have testers who are spending way too much time babysitting tests to make sure they run.  Combine that with the fact that we can't see what is actually running and where it's running at, we end up having a huge gap in machine utilization, meaning we are not only wasting people time, but also machine time.

The Automatic Rerun

Our attempt to automate the rerun process through QualityCenter had to start with some backend processing.  First we needed a database table to store the rerun information.  I'm including the database here since it has to be done and it's pretty simple.  We implemented a single table with 5 columns added to the QualityCenter database.  The columns are id, test_id, test_name, run_position, host_group so that we can order the tests and run them on the host machine they are supposed to run on.  Next, we decided to use QualityCenter workflow to drive the work that needed to be done by writing to this table.  Once that is complete, we can process this information the way we want.

Ideal Workflow

1. Test is identified as "Rerun"
2. Tester changes the status of the run to "Rerun"
3. QualityCenter would take over and insert a table row with the test run information
4. QualityCenter would change the status from "Rerun" to "Queued" to show that the test has been picked up
5. Separate process would chug through the table and run tests as they were popped onto the table
6. This process would also be multi-threaded based on the number of available machines
7. Web page would show current test status (tests in the queue, order of tests in the queue, tests currently running and which machine they were running on)
8. Web page would allow for reordering tests in the queue
9. Web page would also show the current machine status

I haven't worked a lot with QualityCenter workflow and even less in the workflow scripting.  I knew the code I needed to put in...but WHERE?  I started my quest in the script editor after reading a few descriptions of what could be done I decided that the TestSetTests and TestSet were two that made the most sense and I needed either FieldChange or FieldCanChange.  , so add a few message boxes in and start changing test status to "Rerun" to see which function was being hit.  After a little back and forth, I was able to identify the correct Function as *** Insert the function here ***  Now when a test status is set to "Rerun" the workflow updates the status to "Queued" so that we know the test information has been sent to the next step in the process to run the test by way of writing to a database table (Steps 1-4 above)

Heavy Lifting (aka Backend Code)

Most of what we need here is setup and connection to QualityCenter because we already have the database and simple SQL gets everything we need.
Step 1: Read the first row of the table to get the Test ID and the Host
Step 2: Check the host to see if it's a group or a specific machine.

• Specific machine: we have everything we need, just send that machine name to QC
• Host Group: we need to follow a few extra steps to get an available machine for running the test
• Get the first machine name from the host group

• As long as you have something returned, put the data into an array so you can easily loop through each machine to see what's available to run on, something like this:

• Finally, the magic SQL to tell whether a machine is available to use or not:

Step 3: Connect to QualityCenter (this is all standard QC API stuff)

Step 4: Return the test set folder object of the test you are trying to run

Step 5: Return the test set object of the test you are trying to run

Step 6: Start the scheduler and run the test (don't forget to remove the row from the table)

Step 7: Finish up by monitoring the execution object so that you can close out all of the objects when done

Step 8: Clean up after yourself and do it all over again

That's it!

View Into the Automation Lab

The fun of having technical skills on a test team is that you can fix problems that bother you. One  bothersome thing that irked us is babysitting automated test execution.  Starting the execution is easy  (the build system does it), but managing the run to completion (rerunning failed tests) eats time.

Here are details about our situation:

• Our automation environment has a limited number of test machines (11-ish to be exact), these machines are locked away in a room because the desktop cannot be locked to run the automated tests, and they are shared by teams across three continents (Europe, North and South America).  
• Today we have a team of testers that monitor the automated regression run and make sure that all of the tests have run.  When tests are identified for rerun this team will then take a group of tests and manually kick them off.
• Once the tests are running, someone on the team then monitors this set to make sure the tests complete and then kicks off another set of reruns as they are identified.
• If anyone from the other offices wants to kick off a test on those machines, he has to coordinate it with the regression team.  There is no good visibility.
• The end result here is a lot of time spent watching tests run and potentially creating large gaps in machine usage.

Cool Tech to Solve Problems

“Why not build a web based application that can be accessed by anybody in the company needing to know a current status of their tests?”  Right away the ideas started to fly around and the solution, while all relatively easy to accomplish, all of a sudden became further reaching than just automatically rerunning tests.  It would also be nice to have some sort of a dashboard that shows the status of the test machines.  Oh, and how about the ability to change which machine a test is running on.  While we’re at it, since certain projects may have a higher priority let’s add the ability to reorder tests in the queue so those can run first....the ideas were now flowing.

The first step of this solution was to build a function to monitor the test machines.  The rerun app will need to know this.  It would also be cool for people to know this too, so we built a web based dashboard that shows the current status of each machine.  It’s very simple, but provides a huge amount of information that is helpful when evaluating test runs.

This part is taken care of through Ruby and some SQL queries against the QualityCenter database.  Here's how QualityCenter database tells if a machine is available or not:

1: SELECT RN_HOST
2:   FROM td.RUN
3:  WHERE RN_STATUS <> 'Not Completed' 
4:    AND RN_HOST IN ('<your machine name here>') 
5:    AND RN_RUN_ID IN
6:       (SELECT MAX(RN_RUN_ID)
7:          FROM td.RUN
8:         GROUP BY RN_HOST)

With the dashboard in place, the lion’s share of the work is left up to backend code. The rest is display, database updates and other code-type stuff to give it some appeal and not look like a site that a bunch of QA folks came up with. Afterall, we are a technical test team!