How to do SMART Regression
Test?
Regression
Test is a very critical test in any software development phase, which can’t be
skipped or overlooked if you want to deliver high-quality products to your
customers. After multiple new enhancements and bug fixes, only through
Regression Test one can ascertain that the application is indeed stable and
good enough to release into production. To ensure there are no
side-effects/impacts due to new code development, one must carefully craft the
regression test that can be done with minimal effort and maximum coverage. That
is the primary reason on why your regression test has to be SMART.
“~50%
of production defects leaked by QA team are due to inadequate regression tests”
Do we know
that, out of all the production defects leaked by QA team, ~50% of production defects
are due to inadequate regression test coverage? It may surprise you, but if you
do your due diligence in doing the root cause analysis for all the defects
leaked by QA team then you will know that it is true. If it is true in your QA
team, then how do we build the right regression test? Building the right
regression test is very challenging as there are multiple factors that needs to
be considered in planning a regression test with adequate test coverage.
Doing a
regression test by running as much as test cases possibly through automation is
not the SMART way of doing regression test. In today’s Agile delivery model,
finding the luxury of time for doing thorough regression is nearly impossible
and challenging. To address these challenges, I am going to explain key
principles and factors that one needs to consider in building a SMART
Regression Test.
“Smart
Regression is NOT running extensive tests as much as through automation”
10 Key factors for
performing Smart Regression Test:
1. Coverage:
Test
coverage of the regression test is very important factor in deciding the
quality of your regression test. To ensure the adequate coverage of your
regression test, follow the below guidelines
Coverage
by Functionality/Feature – Does your regression test covers all the functionalities/features in
the application? Do you ensure adequate regression tests are added/modified in
the regression suite when a functionality is newly added/modified? List down
the functionalities of the application and map the regression test cases to
appropriate functionality to ensure coverage of regression test for each
functionality in the application.
Coverage
by critical user behaviors/paths/flows – Do you have the understanding of the
critical behaviors/paths in each functionality? Does your regression test
covers all the critical paths/flows in the application? From the System Test
cases designed for a functionality, identify the critical path scenarios and do
ensure to add the critical path scenarios into the regression test suite
Coverage
by critical data combinations – Do you have regression tests that covers the critical data
combinations to validate the business logic implemented in each functionality?
It is important to ensure that the test data combinations used for regression
test is generated using scientific methods (ex: Pair-wise test combination
method). By this you will be able to validate the stability of the application
by running minimal tests with maximum coverage. There are multiple tools in the
market that can help you to implement the right test data combinations without
redundancy. Few tools of my choice: Hexawise (Licensed), Microsoft PICT (Open
Source).
Identifying
the right regression test from the existing System Test cases is a skill and
has to be done thoughtfully by considering the above 3 factors (functional
coverage, path coverage and data coverage). Don’t just dump your System test
cases into Regression Test suite, it will lead to inadequate or redundant
coverage. We should follow proper balance in adding new test cases into
regression test suite. Once should also consider, can we modify the existing
regression test case to cover the new enhancement or not. Always adding lots of
new test cases into the regression test suite and blowing off the test count
will not be the smart way to manage your regression test.
Quote for thought: “Measure the quality of your regression test by coverage and not by volume of test case count”
2. Classification:
Having
right classification in your regression test is very important practice to
ensure proper identification/selection and reporting of your regression test.
Few guidelines on how you can classify your regression test suite
By
Application, by Functionality/Feature, by Priority (P1/P2/P3), by
Sub-functionality (if any), by business criticality of functionality (High,
Medium, Low), by Execution mode (Manual/Auto)
As mentioned on the 1st factor (Coverage), once you have identified the critical & non-critical paths in your application/functionality, do ensure that your regression tests are classified by test priority (P1/P2/P3).
Example
for Classification of regression tests:
|
Reg. Test Case ID |
Application |
Functionality |
Business Criticality |
Test Priority |
Exec. Mode |
# Defects identified (in Past Reg.
Test) |
|
TC-01 |
Member |
Login |
High |
P1 |
Automation |
0 |
|
TC-02 |
Member |
Login |
High |
P2 |
Automation |
3 |
|
TC-03 |
Client |
Registration |
Medium |
P1 |
Manual |
5 |
Quote for thought: “A well classified
regression test is always easy to manage & maintain”
3. Impacts
Impact
assessment is the most important factor in identifying smart regression test. Impact
assessment can’t be done by QA team in silos, it needs collaboration with
Business and Dev teams to ensure thorough assessment of the various impacted
areas in the application.
I will
classify the application impact areas into two categories.
Functional
impact – What are
the different existing application functionalities that have direct/in-direct
impact because of the introduction of a new/enhanced functionality? To do this
impact assessment, it is important for QA to have ‘Functional Interaction
Matrix’ or ‘Impact Assessment Matrix’ for the application in scope. What is
‘Functional Interaction Matrix’? It is a simple 2-dimensional matrix which
shows the interaction between different functionalities in the application.
Note:
* As the above example – Interaction Matrix can also be written between two different applications that has dependency/related to each other
Interpretation of above Functional Interaction Matrix:
- Functionality 2 interacts with Functionality 1
- Functionality 3 interacts with Functionality 2
- Functionality 4 interacts with Functionality 3
By using the above interaction matrix one can understand the impacted
areas and identify the regression suite based on the impact of new/enhanced
functionalities introduced into the application.
In the above approach, QA team has to review the identified areas with Business and Dev team to ensure that the impact assessment is done accurately and there are no gaps in identifying the impact areas which needs regression test.
Technical Impact – To understand the impacts created by technical changes in the underlying code/configuration, QA team has to work closely with Dev team to understand the impacted functionalities. For example, any enhancements/bug fixes in the critical component of the code structure will have direct/in-direct impact on certain functionalities. So, QA team needs to engage with Dev team to get more insight on the technical impacts and identify those functionalities that needs regression test.
Once we have clear understanding of the impacted functionalities that
needs regression test, then you regression will be more focused and targeted
regression than a generic regression. This is a key step in defining smart
regression test.
Quote for thought: “If you know the
impact well then your regression is right on target”
4. Historical
Getting insight from the historical data is an important factor in identifying smart regression test. Historical data could be multiple sources as listed below.
Past Regression defects – Do you track the defects of past regression tests? Functional areas that are error prone in the past are good candidates for regression test. Historical regression test results should help to identify risky/brittle functionalities. Based on the distribution of historical defects in your application, classify the features/functionalities into High-Risk/Medium-Risk/Low-Risk areas. This will play as key input in identifying areas which are not directly impacted by the introduction of new functionalities.
Past Production leaked defects – Do you track the production defects leaked by QA team? This list will carry vital information on why that defect escaped the quality gates of QA team. Do ensure that you track the production defects leaked by QA team and do thorough RCA (Root Cause Analysis) to find the gaps in QA process. Based on this information, we should include specific tests in the regression test that can simulate the leaked production defect scenarios.
Defects from ST (System Test) phase – It is also important to look into functionalities where more defects were uncovered during the ST phase and has gone through multiple defect fix cycles. This could be potential risk areas which can fail in the regression test due code breaks. Do consider this factor for your regression test scope.
Defects from UAT (User Acceptance Test) phase – This is another good area to focus and identify the gaps in QA test and improve the regression test. In most of the teams, UAT will be done by Business users who are functional experts. If UAT team finds any unique defects which are not found by QA in System Testing, then we need to do RCA of those defects to identify the gaps and include the tests in our regression test.
Changes introduced in past few releases – It is also important to track the
changes introduced in the past few releases as it could be a potential unstable
functionality because it is relatively new to the product and hasn’t been
tested multiple times in the past regression. New functionalities released in
recent past releases could be a high-risk candidate for regression test.
Quote
for thought: “You may forget the history, but always remember &
carry the key lessons from the history”
5. New Vs Existing
Does the
regression test can consist only of the tests for existing functionality, or it
can contain the tests for new functionality?
According
to me, Regression test should include the test for existing functionality &
new functionalities (that are ready for go-live/in done state). We need to
remember that the goal of the regression test is to guarantee the stability of
the application. In most of the cases, we run regression test towards the end
(after completing the Sprint testing/completing the System testing/completing
the testing for a release scope). So while identifying the regression scope, we
have to identify the scope from existing/established functionalities + we need
to identify the scope from the new User Stories/new enhancements for which
system testing is completed (waiting for go-live). This approach of including
New & Existing functionality validation in regression test will good
confidence of the overall stability of the application.
Quote
for thought: “Don’t overlook the new when you are too focused on the
old/existing ones”
6. Execution Mode
Once we
have decided on the regression scope, we need to plan the execution mode
(Manual or Automation) for the regression test. Regression test are always the
first choice for implementing the test automation and one should consider
running the maximum regression through automated execution. But at the same
time, we need to have some check and balance to see the outcome of automated
execution. It is a very common notion in most of the QA teams, that automation
test finds no/only few defects. The problem is not with automation or the tool
used, but we need to evaluate all the above factors explained in this article
to ensure there are not gaps in regression test identification.
At the same
time, my recommendation will be also do a limited manual testing (more of an
exploratory test/out of the box scenarios) to check if nothing has escaped the
quality gates of regression test.
Quote
for thought: “Speed is great, but sometimes walking will help your
find hidden mysteries”
7. Iterate
Like the
iterate process in Agile methodology one has to constantly evolve the product
based on the frequent feedbacks. Similarly, we need to constantly evolve our
regression test based on the changing application requirements, new features
introduced and historical data, to ensure that the regression test meets all
the coverage parameters, and it is on par with the application changes.
Regression Test suite should never be stagnant, there should be continuous
effort on a monthly/release basis to enhance regression test bed. This is a key
factor in sustaining a smart regression test.
Quote
for thought: “Don’t be stagnant for long time, it will stink later”
8. Phasing
Running the
regression test in right phase of software development is very important to
have smart regression test. Usually, regression tests are done only in System
Test phase, which may seem like a good idea as we have all the code available
for test will complete system integration. But we need to understand that
finding defects late in the software development cycle is very costly and it
will lead to lot of re-work and potential delay for release timeline. So, to do
smart regression test, we need to shift-left our regression test approach to
the early phases of development to find defects early in the cycle. We can
identify sub-set of our regression (Priority 1 test) and classify them as
Sanity/Smoke test, and we can run it in the development phases to validate the
stability as they are incrementally delivering new code into the system. Automation
will be key player in the shift-left regression approach. This will give quick
feedback to the developers on the impacts created by the new code and will
enable them to fix it early in the development cycle.
Quote
for thought: “Start early as possible, so that you can finish well”
9. Timing
As
important as phasing the regression, it is important to time the regression. Most
of the teams plan the regression test only at the end of System Testing or
before the release. But as mentioned earlier it is going to create lot of pain
and re-work, if we find lots of regression defects at the end of system test
phase. My recommendation is to run-continuous regression throughout the System
test in a parallel mode. If the regression test is automated, it is going to be
easy to adopt this approach.
Key recommendations on timing:
Run Sanity/Smoke test (Subset of Regression [P1 tests]) – Immediately after major code deployments. This can be done alone with your In-Sprint testing in each Sprint.
Run parallel regression test during System Test phase - Run minimal targeted regression test based on the impact assessment of new code deployments. This can be done alone with your In-Sprint testing in each Sprint.
Run extended regression at end of System Test phase – This could be after Sprint +1 or before release. Based on the availability of time, one can plan for extended regression (full scope) or targeted regression (focused on impacted areas)
Quote
for thought: “There is no good time/bad time, running all time will
not get you surprises”
10. Governance
Finally, governance is a very important factor in driving the success of smart regression test. One need to consider the below recommended process to ensure the governance of the regression test
- Regular checkpoints to validate the maintenance of Regression Suite – Master Tracker (tracker with details of all the test cases part of regression scope, with all the appropriate classifications, with clear summary of the regression pack)
- Process to review the regression suite maintenance activity and ensure that the desired test coverage is achieved
- Process to publish the Regression Test dashboard (metrics on TC development/maintenance, Automation, Execution results, defects identified etc.,)
- Process to review the impact assessment process with Business & Dev teams
- Process to review and sign-off the regression scope identified for each sprint/release
Quote
for thought: “Process will never let you down, it will keep your in
momentum and push you further”
Summary
Coverage:
Do ensure that the regression test meets the test coverage by 3 important areas. Coverage by Functionality, coverage by critical paths and coverage by data combination.
Classification:
Classify the regression test by application, functionality, sub-functionality (if any), Test priority (P1/P2/P3), Business Criticality (High/Medium/Low), Execution Mode (Manual/Automated).
Impacts:
Do impact assessment of new/enhanced functionalities introduced based on the functional impact and technical impact.
Historical:
Do consider the data from past production defects leakage, past regression defects, defects from SIT & UAT to identify the high-risk areas and enhance the regression test to validate the stability of the high-risk functionalities
New Vs Existing:
Have balance in the regression test to consider not just the regression tests of the existing functionality but also include the regression test for the new functionalities that are waiting for go-live.
Execution mode:
Focus to achieve maximum automation of the regression test and do consider the inclusion of manual regression test to cover some exploratory/out-of-the-box scenarios.
Iterate:
Regression test needs to be continuously evolved based on the changing application requirements, new features introduced and historical data, to ensure that the regression test meets all the coverage parameters, and it is on par with the application changes.
Phasing:
Planning regression test only at the end of System Test phase could be costlier due to the cost of defect fix and re-work, but instead plan to run regression right from the development phase to get early feedback on the system stability and to act quickly on the issues identified. Think about shift-left regression test (running targeted regression in early phases).
Timing:
Plan to do regression test progressively during the System Test phase (Run Sanity/Smoke test in the beginning of code drops, run targeted regression in parallel with System test, Run extended regression at the end of system test).
Governance:
Establish
governance process to validate maintenance of regression suite, measure the
outcome of regression test and process to include key stakeholders for review
& sign-off on regression scope.
Finally,
you have reached the end of this article, thank-you very much for your patience
and curiosity. Hope this article has helped you to learn and implement few new
areas in your regression test. TEST SMARTER
AND REGRESS SMARTER...
No comments:
Post a Comment