Tech Broadway

[rpanchangan]

Setting up Tool driven Code Review  process Part-2

Going forward in order to overcome the shortcomings described in part one of the blog post it was decided to implement a  tool based review process which supported both pre & post review . Some of the pre-requisites for this process could be that it should be simple to use ,informative, scalable & support our existing tech stack. A number of review tools were identified for this process but two of then were short listed for next level 

a) Review Board - The tool was selected after evaluating various open source tools like code stricker, rietveld, code collaborator, mondarin etc.

b) Crucible: A very famous paid tool

After DAR(Decision and analysis report) report Review board came out as winner with 2.772 * points and Crucible with 2.847 * points. These two tools were evaluated on 80 criteria. The item with minimum points won. The DAR report has been published below:

Old Process

1)The review typically start at the end of development cycle and is usually performed in one go.

2)The review is always Post commit.

3) The Team lead sends mail to the whole team  informing them about the start of Review cycle. Does a bit of administrative work like creating the review document in VSS for that release. (Lot of discussions are held on mails & verbally on issues related to if modules are indeed ready for review)

4) The developers in turn manually create scopes of the code they want to get reviewed and mail them to : 1) Allocated Peers (In case of Peer review ) 2) Technical Consultant (In case of 2nd level review)

5) The developers then wait for their review comments.

6) The reviewers then manually specify the start of review via email.

7) The reviewers go through the code & fill in the review sheet with their comments .Any doubts & discussions are handled verbally or via email.

8) The reviewers send mail to the developers once the code review is complete.

9)The developers then manually go through the comments (Description , severity , Line no ) & start fixing those comments.

10)The Review sheet is further updated with the developers comments (whether the comment is incorporated or not accepted).

11)A mail is send by the team lead to the respective reviewers when their comments are incorporated.

12)The reviewers then go through the  sheet & manually check the code to see if their comments are indeed implemented.

13) The review end with the reviewers marking the comments as fixed with a entry in comments column.

New Process ( after introduction of Review board)

Pre Commit Process 1) The review workflow has two roles: Author and Reviewer of the code. 2) The author creates some changes in his local checkout, computes a diff using svn diff > patch.diff, and creates a new review request. 3) The reviewer opens the review request, examines the changes, adds comments, and optionally marks it as good enough to commit "Ship it!". 4) If the code was marked as "Ship it!", the author can now commit his changes. 5) If the code needs some more changes, the author updates the code, creates a new patch with svn diff > patch.diff, and updates the review request with his patch. 6) The reviewer then starts at point 2 again.

Note: Post-commit reviews are reviews made up of changes which have already been committed to source control manager. Pre-commit reviews are reviews made up of changes which are present in author’s workspace and are not yet committed to source control manager.

Post Commit Process : Real world implementation in production involving Quality center has been described below:

Workflow enforcement

In case of pre commit review the developer is also forced to mention the review id while fixing USD & QC defects thereby solving our workflow enforcement issue.

[rpanchangan]

Setting up Tool driven Code Review  process Part-1

Problem Statement To improve upon the Code review (both Pre & Post commit )process in the project.

Current Working The purpose of review in project is to : 1) Make sure that the code is consistent with the projects coding style  & is of good quality. 2) To spot weak points (functionally & technically) using a second pair of eyes . 3) To prevent the issues from reoccurring (using lesson learnt exercise ).

The same has been implemented on a high level using a two tier process: Tier 1:Peer review Tier2: Technical Consultant review

These two types of review are done  on a implementation level using a combination of Over the shoulder review process (for pre commit ) & email pass around & tool assisted excel based  (for post commit ) .

Though it is simple & light weight process it has many short comings  as specified below: a) Scalability: There are lot of new members in the team &  they are working on diverse modules. There are few gate keepers & it is desired for the members to get the code reviewed by gate keepers before committing it. The no of people & modules participating in review process are growing & the current process cannot handle such level of scalability.

b) Enforcement : How do you automatically enforce review for the code written by a developer from a work package responsible or SME? How do you prevent people from fixing defects in USD & QC with out getting their code reviewed?

c) Administration : How could one manually administer  review requests coming from different members & be structured & efficient at the same time. Currently a substantial amount of the time is spend on administrative work (logging defects, toggling back & forth for code line numbers, copy pasting huge blocks of code etc..).

d) Lack of automation :The current process is manual & thus prone to errors. There is no provision for raising automatic review comments to developers for not adhering to sonar guidelines .

e) Pre commit : Currently there is no process for handling pre commit review requests. They are normally taken on shoulder & many of the discussion in this kind of review are verbal & non committal. 

f) Non available Review Discussions: All the discussions on a review item are either handled verbally or via emails. This valuable data is lost for future reference once the review is completed.

g) Review  Metrics : There is no provision for generation review metrics . Even if we do think of generating them it would take substantial amount of time generating them without any tool support.

h) Lack of flexibility : Code is  reviewed at the end of development cycle in a whole chunk which is never desirable but our existing process is not flexible enough to accommodate continuous reviews.

i) Maintainability of  different process for different kind of reviews: We have different review process for code , requirement, prototype, script, document , process etc . Its very difficult to maintain all of them. 

j) Reporting: It is very difficult to do smart reporting for management using current process . 

k) Lack of IDE Integration The current process is not integrated with IDE so the user have to toggle between the IDE screens & excel sheet in order to log one's review comments. This in turn leads to waste of time & efforts. 

The same were highlighted during my stint as a Code reviewer in subsequent releases:

a) Lot of the time spend on administrative work (logging defects, toggling back & forth for code line numbers, copy pasting huge blocks of code). In total 2 weeks of continuous review 80+ hrs , Still lot left to be desired.

b) Lot of discussions held during code review were of verbal nature & were not documented. 

c) Many comments were raised due to developers not adhering to sonar guidelines around 10 to 15 %

d) Work flow was not enforced so process was not structured . It was my prerogative to send mail to developers asking them not to commit code into subversion when review is going on.

e) Code was reviewed at the end in a whole chunk which is never desirable but our existing process is not flexible enough to accommodate continuous reviews. As you can see from the diagram below  defect density drastically reduces after reviewing 400 lines of code in one go.

f) It was very difficult to revisit my comments & view them if they were fixed properly. It would be much more difficult to analyze the comments in code if we have to do a lesson learnt exercise.

These experiences of mine coupled with scalability issues led us to seek a new process which could streamline & improve upon our current review process.

Questions which may raise on this?

Do we require a new tool? Let us continue with our excel approach Answer : The approach seems good when there are fewer than 5 developers. But as the team size grows it would be very difficult to maintain review comments. 1)  Using excel would restrict people to work on a single document & will have collaborations issues. 2) The administrative cost involved in maintaining excel would be huge & would supersede the main purpose of this process –Review. 3) There would be no workflow enforcement & the process will not be structured. 4) There would be loss of review discussions & it would be very difficult to refer to any older discussions. 5 There would be no automatic generation of Review metrics. 6) An excel would not support Pre commit review , This kind of review will be more often in a team with new resources. 7) f would be difficult for making an excel support say Prototype reviews

Why not give review comments in the code itself?         Answer: Though it sounds very practical & simple it has it own set of disadvantages: 1)How do you perform a Pre commit review here? You can argue that by using patches you can perform it. But what about the comments raised during pre commit review? Where would they be logged & tracked? 2) Can u enforce workflow in this kind of review? 3) How will different stakeholder be informed about the state of the review?? There are no interactive dashboards, pending   queues. 4) What about the discussions performed during review. How would they be taken care of ??? 5) Can there be automatic generation of review metrics?

Why would developers spend lot of time in preparing for a review when the fix is very small? Answer: Just to give an impressions it takes less than a minute to schedule a post commit review in Review board. Its simple drag & drop of the  committed version in Subversion to create an review.

Why not use HP QC for code review tracking: Answer: Quality center is not a code review tool & has lot of limitations, when it comes to automation our code review process. It can only be used to enforce review process using workflows. Rest all our requirements cannot be met by this tool. The three most important requirements of 1)Automatic code review generation 2) Generating & Tracking issues using inline code commenting 3) Administrative support & IDE integration are not supported by this tool.

Next part of the post would include discussion on the proposed solution to tackle the above limitations of the existing process.

[rpanchangan]

Performance testing (environment & infrastructure setup ) -2

Setting up environment

As per the production system architectural we would have decide our environment In our case we decided to have dedicated machines for the following:

1 Linux  server running instance of x App server. (Ideally configured to real world scenario)

1 Linux server running instance of  (Oracle database). (Ideally configured to real world scenario)

1 machine should be configured to handle Controller (Load runner ) or any Load initiation agents for other tools.

X no of machines to generate load ,where x depends on the no of machine calculated depending upon the  of real users generated through  user profiling & transactional profiling. The no of machine required for load generation depends on virtual user footprint & is different for different tools.

Calculating test workload  How many virtual users??  

Suppose the system analyst have identified the X module as the most frequent and resource intensive user action. We can try to calculate test work load for that module.

In the stress test mode, as soon as a virtual user receives a response, it immediately submits the next request. In other words, the virtual user think time equals zero i.e. (Z = 0). In defining the test workload, one has to specify the number of concurrent virtual users.

This number can be estimated as follows: Let R denote the average application server response time, X0 the server throughput, Nr the number of simultaneous real customers, and Nv the number of virtual users. Considering that the throughput and the response time should be the same during the tests and the live operation, one can write the following: X0 = Nr/(R + Z) X0 = Nv/(R + 0). From the two above equations, we have Nv/Nr = R/(R + Z). Let us assume that we have set R = 2 seconds as the average response time goal and Z = 20 seconds for actual users. Then, Nv = Nr × 2/(2 + 20) = 1, 000 × 1/11 = 90.9  91.

Virtual User Memory Footprint

Memory footprint for a virtual user can be called as a function that is dependent on virtual user type (Web, CITRIX, SAP GUI).if your Web/HTTP virtual user takes 5MB when running as a process, 50 virtual user would take 250 MB.

The amount of maximum load  depends on the CPU and memory on a given hardware machine. Execution scenario (varies from product to product)

Virtual users are running on memory instead of disk

Load generators encounter bottlenecks first in memory and then in the processor.

If no of logged in users in X project is: 4000 

A module :  2000 users Clinical suite :  2000 to 2500   we have to find the real users ????

Load Runner virtual memory foot print (report provided by HP)

How many load generators?

Calculations : Memory usage

Process based initiation:  

1 virtual user requires 25.3 MB of RAM 2500 virtual users require 63250 MB RAM = 61 GB RAM 42 virtual users / GB of RAM i.e. 15 systems with 4 GB RAM each handling = 165 virtual users.

Thread based initiation: 

1 virtual user requires 0.76 MB of RAM 2500 virtual users require 1900MB RAM= 1.8 GB RAM 1388 virtual users / GB of RAM i.e. 2 systems with 1 GB RAM  would suffice. CPU utilization

Windows 2000 server   1 CPU , 1 GB RAM:  1 virtual user takes about 0.75 % CPU utilization . Windows 2003 server   1 CPU , 1 GB RAM:  1 virtual user takes about 0.52 % CPU utilization .

Calculation based on  footprint assumption for Load runner on  different machines. Recommended CPU speed for load generation agent machines is 2.33 GHz or greater.

What if memory footprint report is not available ? First we can calculate the memory footprint for 5 to 10  virtual users on our Load testing set up. For that we would be using a simple configuration machine which you would ideally have. ( 3gb RAM or 4gb RAM and single core processor speed) if 5 virtual users take about X MB of RAM. check how many virtual users would take about 60-70% of the total RAM & then check how many machines are required for 2500 users. Based on this calculation you can calculate the number of machines we would need.

How many load generators?

Depends on :

System under test

Protocol used

Test tool

Scripting

Load generation hardware

Q : How much load can be put on a load generator without impacting performance?

Monitor CPU & Memory usage. 

Monitor bandwidth usage .

Monitor usage by script & load test tool. i.e. monitor & load test load testing environment.

Steps: Test the load testing environment. Start by having a few virtual users from a single generator system. Monitor the CPM & memory usage for the same. When you  have the meta data extrapolate the no of users based on the findings. Trial & error method to get the perfect load on a specific instance of a generator.

Note : Use many agent machines with low-end configurations (in terms of number of CPU's and amount of memory) and have these machines generate load. If you use fewer agent machines with high-end configurations, then the operating system is limited by the total number of available sockets, processing parallelism, and limits on multi threading. Therefore, use many agent machines with low-end configurations that are horizontally scaled out for performance testing, rather than vertically scaling agents on a few high-end machines

Check List

Number of Servers: Number of physical and virtual servers

Load Balancing Strategy: The type of load balancing mechanism is in use

Hardware Resources: CPUs count and type, RAM capacity, Number and type of NICs

Software Resources: Standard application build apart from components of the AUT

Application Components: Application components description which needs to be deployed on the server

External Links: Links to third party application and other internal system components.

Some of the parameters which need to be checked on Windows load generation machines MaxUserPort     65534  (Controls the range of dynamically allocated ports) default is 5000 TcpTimedWaitDelay 30 The Windows TCP/IP layer has a registry setting TcpTimedWaitDelay that controls the time a connection is in TIME_WAIT state  )  default is 240 MaxFreeTcbs 72000 MaxHashTableSize 65535 TcpWindowSize 65535 EnableDynamicBacklog 1 MinimumDynamicBacklog 20 MaximumDynamicBacklog 1000 DynamicBacklogGrowthDelta 10 TcpAckFrequency 1

[rpanchangan]

Innovation for Delight

Innovation is the life blood of any organization. Innovation cannot be forced but has to be nurtured from ground up. Recently we had a discussion on how to  improve on the current practices in out account to encourage innovation. As part of that exercise I headed a team to figure out the best way to achieve our objective. This presentation was the end product of that exercise.The presentation was titled "Innovation for Delight". The whole exercise involved participation from 6 team members and a whole lot of exercises & workshops. This exercises also led to the development of a concept of innovation cycle where a innovation could be send to its logical conclusion.     

Primary focus Areas: 1) Member involvement 2) Triggering Innovation thought process 3) Simple yet practical ,result oriented process

Member involvement:

Innovation Questionnaire to measure Innovation quotient. Purpose : Member willingness to participate. Key Observations : New members more keen on participating in Innovation process.

Weekly meeting

Discussions on Lessons learnt & Pain areas.

Triggering Innovation thought process: Workshops conducted for the same

Brain storming

Idea Circle

Spot the trend

Faces of Innovation

Edison Quote

Bell the say no Cat

Simple yet practical Result Oriented process

Primary Idea

Process Planning

Research

Implementation

Result

Each participating member will have to run his/her selected idea through this Innovation cycle which will be tracked by the moderator.

Idea circle:  Ideas give rise to Ideas , Workshop conducted on key focus areas identified :CompetenceTeam CommunicationKnowledge Dissipation/TransferProcess Improvement Some  numbers : No of Participants 6,  No of hrs participated:2 hrs No of Ideas generated : 91

Highlights / Ideas that can be considered for next phase

Most Active Participant : Every body participated with equal enthusiasm& zeal. This gives us a good impression about the future( of this initiative)that with right & systematic  approach we can achieve success.  

Most important learning: 1)Innovation cannot be forced . It can only be nurtured with careful planning.   2) Innovation needs Innovative ideas .    Brainstorming can take you to a level after that people need to think out of the box to proceed. Workshops defined above helps a lot in this area. Numbers speak for that. 3) Innovation should be fun   :) not part of appraisal.

Most important  : People do come up with great ideas when they are given right environment  & push. Scope of improvement Innovation should be a ongoing process not one time affair. Focus should be on innovative thinking not monetary savings.  You tend to loose many ideas when you think about money first. People should be rewarded for ideas  :)

[rpanchangan]

Performance testing (environment & infrastructure setup ) -1

This is the first article in a two part series describing the approach for setting up performance testing environment. It contains some of my experience in setting up the same in my project.

The most important part while creating an environment for performance testing is to study the existing production environment & create the exact replica of the same or a proportional toned down version of the same. This is undoubtedly the most important part of the whole testing activity as even the minor differences between the test environment and production system can completely invalidate the test results. 

Things we need to look out for: Servers Infrastructure (Across application tiers): Cost & Complexity challenge Network Infrastructure: Bandwidth & Connectivity issues Application Tiers: Should have exact no as in production environment to achieve accurate results Database size: Size impacts results. Should be same as production (in proportion if cost exceeds ). A database with different size than production will not be able to generate accurate results.

Remember (For accurate performance testing environment setup): 1) Test environment isolation  (No processes are running during  text execution) 2) Network isolation  (Sufficient bandwidth availability) 3) Load injectors Requirements: a) Sufficient hardware resources to support virtual users b) Amount of load generated depends on (RAM, CPU, Disk), network bandwidth, script complexity and think time) 4) Test data generators requirements (data sets should be same in real or in proportion) 5) Remove proxy server  from network path

Environment Setup options / approaches

There are four approaches for setting up environment:

a)Testing setup on Production Environment

Advantages: 

a) Readily available production data       b) reduces infra costs & time c) application setup complexity readily available   

          Disadvantages:

a) Very risky b) might introduce junk data on the system c) Difficult to get downtime.

b)Testing setup on proportionally  Scaled Environment

Advantages:

a) Cost effective easy to set up b) application bottlenecks east to identify in scaled down environment.

Disadvantages:  a) Difficult to find performance bottlenecks beyond scaled environment. i.e. Not that accurate  as above approach.

c) Testing setup over the Cloud

Advantages:      

a) Flexible & cost effective.

Disadvantages:

a) Security & privacy concerns   b) Network connection speed c) Not reliable. (May stop support) d) Not ideal for new framework setup rather if perfect for a matured process.    

d) Testing setup with Service Visualization

(Normally being used in agile development. Emulates certain components for fulfilling requirements) Advantages:     

a) Emulate extreme loads  (realistic performance ) on dependent application with very less cost. b) emulate components yet to be developed.

Disadvantages: 

a) Not reliable b) Cannot emulate all kinds of components

Environment Setup options / approaches

What should we follow?   Since in my project we have the following challenges  :

Privacy & security issues (Clinical data)

Cost, time  & infrastructure constrains

Waterfall model

Negligible downtime available

The best approach to follow was to use the scaled environment approach . We should perform the load testing on the scaled down version of production & then extrapolate the results to reach out an conclusion.

“Analyzing the system under stress”

Identify system components

Systems architecture diagram

Specify configuration of each component

Hardware (CPU , )

OS

Memory (RAM)

Specify the user profile

Type of users Common tasks Task distribution

Why do we require user Profiles?

To calculate real users for mapping them to virtual users to be used for replication load )

Questions we need to ask : (To calculate )

Who are the users ??? 

How many users???

What tasks do they do ???  

How often they do them??

Transactional Profile : We need to calculate transaction profile as well to target those modules which are used most . In our case we have selected three basic modules for simplicity purpose.

Ideal scenario states that we need to have both user & transaction profile available to take decisions regarding no of virtual users to be used for implementing load testing. But in our case ,due to unavailability of this information  we are making an assumption using the no of concurrent users working on the system.

the next article will include design decisions on how the actual environment was set up.

[rpanchangan]

MS word as WYSIWYG  editor for Wiki

While creating wiki pages for  my project I came across this neat plugin which converts MS word into WYSIWYG  editor for Wiki. This plugin helped me in reusing the existing content I had in MS doc files. Using these plugin has considerable increased the acceptance of wiki as a one stop destination for information. The plugin can be downloaded from the following URL:  http://www.microsoft.com/en-us/download/details.aspx?id=12298. This plugin add a new format to MS word called Media Wiki. The only thing user need to do is to save his word  documents into this format .

save it to MediaWiki format

The content in this txt file then needs to be copied directly to your new wiki page.

after saving the wiki page you can see the word table is exactly replicated in the wiki page .

[rpanchangan]

WWDC 2015 is starting tomorrow....

Just a week after Google had their developer I/O conference we are now getting ready for Apple developer conference. These are exiting times  as we can now see tech companies coming up with innovative solutions and ideas to be part of your daily life. 

Google started with project Brillo an operating system which will help your  home appliances to talk to each other and to connect with web.

& Apple is now coming up with Homekit a kind of home automation platform. 

In coming times we may see a wide variety of apps specifically designed to cater to Home automation and based on these frameworks.

[rpanchangan]

Configuring JSHint in IntelliJ Idea

JSHint is a great static analysis tool for Java script. It helps in detecting errors , potential problems & standardizing Java script coding guidelines in a project.

IntelliJ idea comes pre bundled with JSHint. The users can activate JShint by enabling JSHint in JavaScript -->Code Quality Tools --> JsHint

Now the question arises what if a project wants to use latest version of JSHint but is using a older version of IntellJ idea? Is there a way to manually install the latest JSHint configuration in IntelliJ idea.Yes there is...

User can specify the latest version of JSHint to be used by IntelliJ idea by updating the versions.json file present in IntelliJ ideas system folder created for each specific user.

The file can be found at the following path: C:\users\ your user name\.IntelliJIdea version no\system\javascript\jshint\version.json

Users need to updated this file by adding new entry of the version desired by the project.

As soon as the version is updated IntelliJ idea automatically downloads the specified version and places the same the same the JSHint folder.

The specified version is also now available in the version drop down in IntelliJ idea JSHint configuration

In upcoming series I will talk about How to setup a static code analysis for process web app projects involving angular js.

[rpanchangan]

What motivates us?

Now that's a big one. Psychologists have been trying to answer that question for many years. There have been many experiments & case studies as well. Each study ,proves one thing at the end of the day “Motivated people aka employees are a great asset to an organization”.

So in short this question effects every organization whether it is a start up or a multi billion dollar behemoth.

I recently attended Scrum training where the trainer shared a great video titled

The surprising truth about what motivates us  by RSA Animate. Now generally every one equates motivation with rewards. If you get rewarded you are motivated. But this video proves other wise.

This video beautifully relates 1) autonomy 2) mastery 3) purpose as three major factors for performance & personnel satisfaction.

A stepping stone for high performance Scrum teams

[rpanchangan]

How to configure tumblr with go Daddy ?

Tumblr is a great place to blog and has become an important platform for bloggers all around the world. It allows people to create their own blog or to share multimedia & other content. One interesting feature in Tumblr is it allows you to map your blog to your custom domain.The post below list the steps required to configure your tumble blog with goDaddy account:

1) First the changes in your go Daddy account--> log in to your GoDaddy account

2) Click on Manage My Domains link

3) Click on the Domain name you have registered with go Daddy

4) Click on the DNS zone file

5) There are two things you need to do in this file

a) Create a new entry in A(host) section and point A-record (IP address)  to 66.6.44.4

b) Point CNAME record to "domains.tumblr.com"

6) You are done with your go daddy changes, now Log in to you Tumblr account

7) Click on the Settings (the gear icon) at the top of your Dashboard.

8) Select the blog that you’d like to update on the right side of the page.

9)Click the pencil to the right of the username section and enable Use a custom domain.

10) Enter your domain name and click on “Test domain”. This test should be positive for you to be sure that your configuration is done properly.

11) When “Test domain” is successful Save your configurations.

Note: It would take at least 72 hrs for the configurations to take effect, during that period you might see tumbler error if you try to access your domain. Once the sync is complete you will be redirected to your Tumblr blog via your custom domain.

[rpanchangan]

Welcome to Tech Broadway