Agile Sharepoint development by 21apps and Andrew Woodward MVP

Having worked for more years than I can remember as a consultant I have lived on my laptop,  everything I do is on the laptop including all of the development.   My current, and have to say the best one ever owned, is a Lenovo T61p.  The plus points it has a proper caddy so I run two drives internally (VMs on one, OS on the other) and all of the drivers just seem to work.   I’ve upgraded the drives a few times to squeeze in some more space (now with 500GB available) and for speed 7200K disks,  I plan to upgrade the main drive to use SSD so that I can keep the old friend going a little while longer.

Many of you are likely in the same place and are thinking,  heck that seems like a reasonable bit of kit, what’s the problem?

The problem is the laptop is OK for single machine development,  as in it’s usable but nothing lightening.   I could, as some have done, pave the way and go Windows 2008 server native to do my development but then I loose all the nice things like just closing the lid and it going to sleep.  But the biggest problem is that I want to do my development right!  I want to get my Continuous Integration running,  be able to test in a scaled out environment and also be able to play with all the new software that is being released.   Sadly my laptop is not really going to cut the mustard here.

I was encouraged by the rig that Andrew Connell put together and this really fitted the bill for what I wanted – I knew this was not going to be cheap,  but these are the tools of my trade.  I wouldn’t expect the local mechanic to be using 99p spanners from the market, they buy the tools that are right for the job. 

Decision  #1

At the time the new i7core processors where starting to hit the shelves,  would it be possible to get a ‘workstation’ setup with dual processors using this new chip architecture?   The answer was you could,  you would be bleeding edge and you would pay through the nose.  So the hard choice was made to stick with the tried and tested XEON proc’s.

Decision  #2

I wanted to do a custom build,  I spent ages looking at the components, deciding what case, proc speeds, heat sinks etc.   The problem I had was in the UK I could not source all of these components from a single supplier.   Why a single supplier you ask?  Well in the UK to make it easier for small businesses you have the option to us register for flat rate VAT scheme, this means calculating the VAT payment based on sales rather than reclaim for everything.  Generally it works well,  the problem is that for larger capital purchases the flat rate doesn’t work.  To support this if a single purchase is above a set amount (£2000) then you can claim the VAT back in addition,  the problem is that it has to be on a single invoice.   So if I purchased the components from the various suppliers I was looking at an added cost of 17.5% – a not insignificant amount.

The Rig

Being a fan of Lenovo since owning my T61p I decided to give their D10 workstation a go.   I had read some good reviews and it was designed with one of my biggest criteria in mind,  not sounding like an aeroplane in my office.

Problem #1

I put together my original spec and placed the order with Dabs.com for the D10, an additional processor, RAM upgrade and some additional drives.   I waited,  and waited eventually I received an email from Dabs saying this product was discontinued!  I knew this was not the case so I went to the go direct and ordered from the Lenovo site and purchased the main rig (with all the lovely 4 year next day onsite warranty) and then upgraded this where the cost for the components was cheaper sourced elsewhere. 

Note: About 1 week after buying mine they announced the D20,  If I were buying today this is what I would get.  The problem was I wouldn’t be able to get a D20 for 3-4 months in the UK,  so this was not an option.

The Rig Continued

I did manage to get a good deal on the kit,  making use of an autocad discount code so was not overly concerned about the D20 release.  Working the numbers carefully I got a the following kit from Lenovo to start my build

• D10 Workstation
• Single Xeon 2.5GHz processor (dual capable board)
• 18GB RAM  (2GB with the machine, + 4×4GB additional chips)
• 250GB 7200disk

The big seller for the D10 was the onboard SAS and SATA RAID with the SATA supporting RAID 10

The following additional items were purchased independently from Scan

• 4 x Western Digital Cavier Black drivers
• 1 x Xeon E5420P Quad Core Proc
• 2 x Thermalright HR-01X with 120mm fan and fixings

I also ordered a pair of LG 20” widescreen monitors from PCWorld Business,  was trying to justify 24” but my room is just not that big .   The monitors had good reviews,  support portrait mode and were surprisingly cheap.

The additional kit from Scan arrived first,  followed by the additional memory, I was counting down the days for the workstation.   Finally I had an excuse to convert the spare room into a spare room/office

It arrived

Lenovo Delivered

First thoughts,  this really is going to be a DOA,  I like to treat my machines with a little bit of respect, it looked like the couriers had used this for target practice!  A few photos later just in case I needed to send it back and I had the rig out.   One thing to remember with these Workstations, they are significantly bigger than a standard desktop or tower.

Lenovo Size

Shown here with my 15” T61p laptop you get a sense of the size.

First Impressions

The D10 is really well put together,  hit had survived the crash testing by the couriers and everything was all still connected and working.  This bodes well for the build quality of the machine.

When I first turned on the machine it sounded like a 747 was spinning up ready for take off..  thankfully after a few seconds the fans all slowed down to a give a near silent machine

Once I proved the machine was working and had it booted up into the shipped software I was ready to start customising.  I decided to get the base OS replaced before doing any hardware updates,  my Virtualisation of choice was Hyper V so I started with the latest Windows 2008 release,  I decided not to go R2 beta as I wasn’t planning to refresh the base OS anytime soon.  My long term plan was to have the OS on a RAID 0 SSD setup,  for now it would just run on the shipped drive.

Inside the rig is fairly spacious and it is really well put together.  No need for screwdrivers for adding disks and the ductings allows the air to flow where it’s needed.  There is a lot of fans in the box but it really does not make much noise at all, in fact the loudest part if the noise from the hard disks.

Inside the Thinkstation

I really like the memory configuration with the rear fan pulling all of the air over the RAM to keep it cool. I see that Andrew Connell was having issues on his rig and had to install memory cooling.

Memory Ducting

The machine runs at a reasonable temp I think,  and not had any issues even during the really hot spell we had in the UK and like almost every UK home I do not have aircon so when it’s hot outside it’s hot inside

Internal Temps

The OS install Part 1 & 2

First install of Windows 2008 went very smoothly,  the software went on really quickly and I had a base setup working with.  Next lets get some ram and hard disks into this baby and see how it flies.

Added the 4 x 1TB drives,  these are SATA drives and were configured very easily using the built in configuration to provide a single 2 TB RAID 10 configuration.   RAID 1 is used for resiliance, every thing is copied to both drives,  RAID 0 is for performance where the information is written across the drives giving you much better I/O.   Doing this in a RAID 10 gives you the best of both worlds.

More Info on Raid configurations

I moved the OS drive over to the SAS controller which also support SATA drives as I had now used up all of the ports on the SATA controller (one was still used for the DVD).

Rebooting the machine and nothing,  it was unable to recognize the boot drive on the SAS controller .   I tried various configurations, boot options and changes all to no avail.   Eventually I decided that i would do another install,  this time using the RAID 10 configuration for booting as well.   (Looking back this is probably the right thing to do as I now have resilience for the OS as well).

OS installation Part 2 when smoothly and it was amazing how much quicker the install was now it was writing to the stripped and mirrored drives – ‘I’m going to like this’

Disk Config

Here you can see the basic disk configuration,  226GB for the OS and 1.6TB for the virtual machines.  I also used the 250GB drive as a file share for ISOs and software.

Hyper-V and Graphics Drivers

Still running on one monitor I setup Hyper-V  and installed my first VM,  WOW = this thing flies,  why oh why have I waited so long to get something like this and move away from the about bearable performance from my laptop.

Time came to get my dual display hooked up and working,   i added the additional monitor – the NVidia Quadro FX1700 shipped with the box supports 2 direct connections and I think can support upto four monitors.   I added two in and looked for config options.   Sadly the currently installed SVGA driver that ships with Windows 2008 can’t support external monitors, or dual monitors or anything other than a very very simple single monitor with limited resolution options.   A quick dig into the drivers on lenovo and I was installing the 64bit Vista XP driver,  install went well everything seemed to be working fine.

Note: I’m not planning to do any video editing or normal day to day stuff on here,  this is really a dev rig.  My laptop is still my main day to day machine/

I fired up the VM and connected,  there was a noticeable delay – it was only a few seconds but compared to the lightening speed before this seemed odd.   On researching this more I found that this is a known issues (but not an issue Microsoft have as at today anything planned to resolve).   I switched back to SVGA and lightening fast,  NVidia and a delay in connecting,  the thing to note is the delay is only really when connecting via the Hyper V manager,  if you remote desktop to the VM then it works just fine.

I’m not thrilled about the delay, but the value of the duel monitor and better resolution is more valuable to me.   I had considered perhaps using VMWare workstation at one point but really wanted to get to know Hyper V so kept the VMWare for the laptop.

A few posts on the subject,  I am hoping that Microsoft do solve this

http://blogs.msdn.com/virtual_pc_guy/archive/2009/01/07/bad-performance-with-high-end-graphics-and-hyper-v.aspx

http://social.technet.microsoft.com/Forums/en-US/winserverhyperv/thread/4e1c53f5-0400-4ca9-8819-f942c10881c1/

Adding a second CPU

Now that my machine was running,  VM installed and ready for some serious work I though it the right time to add the second processor and put in those nice Thermalright coolers.

A quick review of the Lenovo site and I had instructions, in fact a fairly simple process for installing the new CPU and adding the heat sinks. As you can see these heat sinks are beasts compared to the chip that they are there to cool.

Heat sink to processor

After some careful work removing the original heat sink,  adding the new on,  adding the additional CPU and it’s heat sink I marvelled at just how good it was to build you own rig.

Problem #2

Next step,  lets get the cover on and give it a whirl.   I had removed some of the internal air ducting as this didn’t fit with the heat sinks but I was more than annoyed to find that I couldn’t put the case back on!!!   The case has a nice handle and a key lock that you can see from above,  the result is that these monster heat sinks will not fit!.   After a few choice swear words I decided to test it anyway, with the side off.  I could look for smaller heat sinks later.

Problem #3

Started the machine and bang!  The famous BSOD!

I knew this was something to do with the new CPU,  but had to start investigating it to see what it was.   Some more fiddling with the machine,  removed the original CPU and heat sink started the machine and it booted up and worked fine.  So there was nothing wrong with the CPU,  was it to do with the motherboard?

A thing to note, taking CPUs and Heat sinks in and out of the workstation is not something I recommend,  and ensure you have some spare heat paste as you will need it.

Time to turn to the internet,  the benefit of buying a branded workstation was that it is easy to narrow down searches.  I was lucky to find a whole support forum dedicated to the Lenovo Workstations and it was here I found the reason for my problem.

http://forums.lenovo.com/lnv/board/message?board.id=ThinkStation&thread.id=755&view=by_date_ascending&page=1

It appears that two Xeon E5420 processors are not always the same.  They rely on both processors having the same stepping code and have a slightly different revision code SLBBR and SLANV.   The problem is that there is no way through any of the suppliers I have seen to be able to order a specific processor revision.

I phoned Scan with my request and a desire to replace the chip,  they were not willing to do this as I had obviously opened the box not expecting to have this problem.   To be fair the stepping code is shown on the box.  I asked the guy at Scan so can I order this particular stepping model to which he said no, sorry we don’t stock them as different items so all you can do is order it and if its the wrong one send it back.

At this point I’m really hitting a low point in my affections with my dev rig,   realising I now have over £350 worth of upgrades that are worthless is not a way to make me happy.

Getting the 2nd CPU

I am not a big user of Ebay,  but found that this was the only way I could get a CPU with the exact stepping code I wanted.   I eventually got one for around £200, was able to install it and it worked first time.  +1 for Ebay

I still have a few left over components I need to shift,  including the Xeon Processor and some heat sinks.   Time to get back and make use of Ebay again.

Living with my Dev Rig

I have found the snapshots and vm revision tree in Hyper V to be really cool and very very quick.   It was a bit confusing at first understanding that you select a snapshot and choose apply to go to that.  I have missed the idea of a Team in VMWare workstation and also the cloning of images using export/import is less intuitive.

I have recently ventured into using System Centre Virtual Machine Manager 2008 R2 to see what value add this can give me.  I know it’s targeted at bigger setups but I think this will be a great investment as I start to automate more of my development and test environments. 

SCVMM does rely on AD so I am now in the strange position where my AD and SCVMM machines are running as Virtual Machines on my Dev Rig and the SCVMM is managing the Hyper-v as a perimeter  setup which includes itself.  I’m pretty sure this is not a Best Practice 

Going forward I can actually see my Dev Rig taking on much more of a server role,  in fact this is really where it is now with very little installed on the host OS.

What I do like is that I now have the power and storage to really start ramping up on all the new cool software that is coming down the line.

I did however notice that my dev rig has caught the attention of someone else in the family.

Someone else using my dev rig

At least he’s working on something important,  School Project: The History of Chelsea Football Club.

Scrum as an agile approach is great, it allows teams to get up to speed quickly with agile and also enable the business to understand what being agile means.  The problem is that people often just do Scrum.  Doing Scrum is not enough,  it is a great place to start and will help you with your projects but your team still need to get the engineering disciplines in place, they still need to work on doing the right thing the right way and with level of quality demanded by the business.

Jack Milunsky posted a nice article that talks about why Measuring Velocity is not enough.

Like burndown charts, velocity is just another metric which the team can use to reach what I believe is the ultimate goal – sustainable throughput. Velocity in my opinion, is not a metric for determining productivity.

What I really liked, and is something that I am going to be covering a lot more over the coming months is the conclusion

.in order to enhance productivity there are many things one can do:

* You have to get the requirements down pat. The best way to do this is to work with the customer to define the user stories and get their intent up-front.
* Ensure you are capturing the acceptance test criteria from the customer up-front. This helps solve the greatest risk of all – getting the wrong product.
* Define your definition of "done" up-front so that there’s no argument at the end.
* Ensure you build quality in right from the start.

The conclusion can be split into two areas

1. Understanding the Requirements
2. Getting your process right

Getting your process right is something the development teams control,  and there is a lot of information out there today about techniques like Test Driven Development, Continuous Integration and the like that all focus on the development process and quality.  

Understanding the the requirements on the other hand and getting the business to provide you with valid and testable user stories is an area of agile development that I feel has been neglected.  There are great books that have been written on how to write user stories and how to ensure they are testable.   There are other books aimed at the business analysts about how to gather requirements.  However these have never come together, agile development teams don’t look much be the user stories. 

This pushing back beyond the user stories and into the way requirements are gathered and the business is understood is something I have been working on for a while.  And with this push back beyond the user story it is clear the way to get business and customer buy in to the agile development approach is to really change the way businesses approach the complex problems they are trying to solve.

I have some really exciting things to announce over the coming weeks that will really take this discussion forward, working with some of the most visionary guys in the field.

Rest assured there will be a very big SharePoint focus.

I recently wrote about my experience of doing agile development in a team of one,  where I questioned the idea that ‘If your the only member of the team is it really worth going through the same agile development process that you would in a team of eight?’

One interesting test for the project was that it was not going to be developed in one go,  the days spent on this project would need to fit around other work and I want to see how using agile techniques could help to keep the momentum and also retain the knowledge from the previous activities in a format that would be easy to pick up.

Sprint 1 – Day 1 was 15th May 2009

Sprint 1 – Day 2 was 4th June 2009 a full 20 days gap!

Daily Scrum

Always start the day with a scrum and ensure that you come prepared.  This meant that I needed to be able to answer the following questions

• What did you achieve yesterday?
• What are you going to achieve today?
• Any blockers preventing you from doing this?

Take a minute,  try and think back 20 days and remember exactly what you had achieved?   Difficult isn’t it!  This is where using a process really helped,  I started by looking at the my white board to see where I left the project.

Interestingly I had not ‘Done’ anything (refer to Day 1 for the definition of Done and how I structure my white board), my Burndown for Day 1 was looking very sad so I won’t even show it.  What I could however see was that I had completed the work on one story,  the second story was in progress and that I had not started the 3rd.  The observant will also notice I have a couple of new stories in the backlog.

If I had not been working on the project for the past 20 days how come there are new stories?

Adding to the Backlog

One of the tenets of being agile is that you work at a sustainable pace,  you don’t do more work than is possible and you allow your team to go home, see the family and have a beer or two.  What happens when you team is having that beer?  They think about work, and have ideas about ways to do things or new requirements and then write these down as new stories.  The same is true for the business, whilst the team is busy working on the current iteration the business is thinking about what they need, reacting to change and then feeding this into the backlog as new stories.  They may even take away some stories that are no longer needed.

What does this mean?

Well for Sprint 1 nothing,  we have agreed a short period of work where we can focus on the deliverables.   The new stories will be considered for Sprint 2 and for any removed stories we will have the warm feeling that we had not spent too long on in big up front design tasks and documentation.

The Rest of Day 2

After the scrum I reviewed the solution I had put together,  and compared this to the stories.  At the end of Day 1 I had a WSP  and some features that meant I could see what I had done and actually use it.  The ability to use the solution very early in the process is another of the big benefits to being agile.  It allows people to see it for real, to make comment and also to make changes if it is not how they envisaged it.

Needless to say I made some changes,  there were bits that I had added that on reflection were not required,  I removed the extra items and did some refactoring of the solution to make it more compact and understandable.

The rest of the day went pretty well,  and at the end of Sprint 1 I had not actually written any C# code or created any Unit Tests.  The solution at the end of the sprint I had completed the three stories I had planned.  I have to confess that I did do a few extra hours work at the end of this day (Sprint) in order to complete the stories. This can be a common scenario where the a stress levels and work will increase towards the end of the Sprint.  The same if true of Waterfall projects but at the end of these projects there is just so much more that is expected to be done the stress levels go off the scale.

The final tasks for the end of Sprint 1 was to do a “Show and Tell” to the business.  To make this more real,  and to be able to provide a summary once the project is released I record a Screencast demonstrating what had been done.   It may sound silly doing the screen cast or doing a show and tell to yourself, however what it does do is make you bring together everything and ensure that what you said was done, actually was done.

Sprint 1 Retrospective

In order to improve our processes, to produce better quality solutions and to make ourselves more productive we need to look at what happened in the Sprint, the good things as well as the challenging ones and also look for ways to improve.

Things that went well

• Adopted great agile techniques
• Was able to pickup after a break
• Good use of tools like WSPBuilder and hosted source control

Things to improve on

• Testing was manual – need more automation of integration and acceptance tests
• Testing was on single machine – need to test on multi server farm configuration
• Project break was too long – final delivery will be too late at current pace

If you want to learn how to be better at retrospectives I recommend getting a copy of Agile Retrospectives, Making Good Teams Great by Esther Derby and Diana Larsen.

Summary

Based on the experience so far I can say that adopting agile techniques in a team of one really is something that everyone should do.   I agree just cranking out code some developers could be a fair bit more productive up front than I have been. However if the project takes off and you need to bring in more help or you have a long break the project will eventually fail without these process in place.

And the biggest benefit to doing this is that you learn the benefits of being agile,  you understand why it works and you will be able to add this to your CV and Life experience and go on to become a much better developer.  You will be able to challenge the traditional practices in your organisation based on experience and not just hearsay.

If your the only member of the team is it really worth going through the same agile development process that you would in a team of eight?

Isn’t it quicker to just crack open Visual Studio and start coding?   After all your the customer so you know what you want right?   You not going to have to deal with any of the social complexities that conspire against you. You’re going have the hottest of communication with the dev team,  the dev and test roles will have an interment knowledge and the scrum master won’t really have to do anything as there is not going to be any blockers on this project.

Right?

So I started down this path and before I knew it I had some weird project created with bits of code and SharePoint artefacts but not really seeming to make any progress.   I was easily distracted by other things and didn’t really have a feel for how long it would take and when I could say I was done.

STOP!

What was I doing!  I spend my time trying to help people develop better, to take small steps and sometimes big steps to help produce better software..   why was I not practicing what I preached?

I stopped!

Deep breath, deleted all the rubbish I can previously created and got back to basics.

The Team

I have a team,  the fact that some (in this case all) of the roles were done by the same person was ok; that happens often in agile teams.

Design

Yes, you heard me right here.  Eric Shupps would be proud of me .   I’m not going to jump into Visual Studio and crank out code starting [TestFixture] I need to do some design work first, even though this application is going to be small I still need to put in the time on the upfront design.  I need to think through the solution at a high level.

Here I used the opportunity to try out the great Balsamiq application,  which I stumped up the $79 license fee for and download the desktop version.

I’ve deliberately blurred the image but you get the idea of what the mock ups look like, and these take only a matter of minutes to produce.  Definately something you should consider if your currently trying to do this in Visio or heaven forbid cranking out HTML mockups.

The few UI based mockups actually helped me to understand how the application would work for the end user, which in turn provided a clear vision for the way the solution would be developed.

User Stories

I didn’t need to do any remote collaboration on the User Stories so I was able to get back to basics and use one of the white boards in my office as the task board.   I wrote the user stories out on post-it notes and added them to the board along with rough estimates for each one.

Here you can see I’m using a very common step of status on my task board

Backlog - This is where I add all stories that came out of the design,  they are all user focused with nothing specific about the implementation.  No stories like ‘Use jQuery to make it nice’  these are user stories.  If I think of something while i’m coding or testing, or even having a beer I can just write it to a post-it and stick it on the backlog.

Sprint – I like Scrum so adopt a Scrum Esq. style to my process.  The sprint will be where I put the stories I plan to do next.  The amount is based on the estimate and also the sprint length.   In this project I’m doing stupily short sprints of 2 days each.

In Progress – When I pick off a story to work on it goes in here,  if you’ve ever looked at the Kanban approaches your looking to keep this ‘In Progress’ to a defined level.  I like to try and move things through to ‘Done’ before picking off new stories.

Verify – When I think it’s been coded properly and dev tested I move the story here.   Verify is where I would validate the code on another environment – i.e. a non-dev box,  one with multiple web front ends.  At this point I also look at introducing some form of automation for both the build/deployment and also the acceptance testing.

Done – This is when it is really really done,  there is nothing left to do, no documentation, no tests or tweaks,  the code is production ready.

Sprint 1 – Day 1

Having selected the user stories for the sprint it was time to setup my dev project and get some of the foundations in place.   If you follow me on twitter you will know that I previously spent time producing a decent SharePoint development Virtual Machine so it was very quick for me to roll out the few things I needed to get going.   A SharePoint team site in which I could spike and test my code,  a VS2008 solution and class library project in which to start cranking out some code.

At the end of Day 1 my board looked like this:

One of the stories I felt was ready for verification and another I was making good progress on.   At this point I had nothing ‘Done’ my burndown in the morning would look a bit sad,  but that’s OK that is often the case at the start of a project or sprint.  I also don’t at this point have a verification setup so this is going to slow me down a bit on Day 2.

Day 1 has been good,  I have the basic project structure defined.   I was able to build a WSP with the SharePoint features I needed to complete the first story and was able to deploy and test this on my dev machine.

People that have read my blog or seen me talk at conferences and user groups know that I am very keen on doing things right; I like to see Unit Tests and better still I want to see people at least trying Test Driven Development.   However at the end of Day 1 I was yet to write a single line of C# code, neither in test form or production code.   So did I really do development on day one?

Well I think I did, only the language used was XML and a little bit of CAML and for these items I do not know of any way, that makes sense, to unit test.  I often think that the XML we define as part of SharePoint projects are more inline with doing configuration,  we are just setting up SharePoint using a specific configuration and the testing for this should naturally falls into the integration space and will be picked up during verification.   That is not to say as a developer I do not validate it,  I need to ensure that it works as expected in my environment before I can move it over to the Verify column and it’ is the job of the tester to make sure this happens.  For this project the tester (me) was adamant that it all worked as they didn’t want to have to waste time on silly errors.

Interesting Test

One thing that I will be having to do with this project is fit it around real work, the sort that helps to keep food on the table sort of work.   This means that the sprints although 2 days in length the elapsed time may be significantly longer.  I know that this will impact on the velocity as I will lose some of the momentum during these breaks, however the benefit I have from adopting a more defined process is that it will be significantly easier for me to pick up the project.   Imaging if I had to go back to the mess of code I had originally created after a weeks break!

I’m really interested to see if my agile development with a team of one really does make a significant impact on the way I develop the code, on the time it takes to complete it and on the quality of the solution created at the end.

Find out what happens on Day 2.

How many times have you added an additional parameter or bit of logic to your code because you thought it could be useful and it’s easier to add whilst your already changing the code?   If you are your introducing Future Creep!  and should stop doing it.

What does Future Creep mean?

If your from from an XP (Extreme Programming) background you would know this better as YAGNI,  You Aren’t Gonna Need It!   I actually prefer the term Future Creep I first saw on the post, beware of future creep by Jamis of 37 signal.   Everyone understands the terms Scope Creep and Feature Creep and its easy to spot these, although often ignored.  Future creep is the adding of additional code by the developer because they can see a possible future requirements.

DONT DO IT!

You may think, hey it’s only gonna take me a couple of minutes to add this additional parameter which I can see being really useful in the future.  You add the code and then move on,  from that point forward that additional code needs to be maintained, every time a developer interacts with your code they need to read it and understand it – often taking more time as that bit of the code is never actually used, which leads to uncertainty about the code.

It happened to me

The reason I wrote this blog post was an example of future creep on a recent project that had a measurable impact on time and effort. The future creep was was simple, the developer was writing a light weight logging library and decided as part of the Logging.Log() method that if the code had a HTTPContext they would also initialise the current web using SPContext.Current.   None of the code within the internal methods made use of this it was just setup and passed into the method.  This in itself does not sound to bad and for most scenarios the code did run with a valid SPContext h0wever this future creep caused a number of issues which forced developers trying work around it

1. The logging library could not be used in code that had elevated privileges as SPContext.Current throws an error if accessed under a different security context
2. The logging library could not be used where a HTTPContext existed but no SPContext, as was the case for some of the Exchange integration code

Having battled with solutions to the second scenario and getting comments about not wanting to make changes to the Logging Library because the dev was not comfortable with the way it had been implemented I challenged the original author of the code as to the requirement for SPContext.

The answer:  Future Creep

A few minutes later the extra code was refactored out, and the library could be used.

Lessons Learnt and how to prevent it?

Don’t do it,  resist the temptation, remember:

The best way to implement code quickly is to implement less of it.

The best way to have fewer bugs is to implement less code.

And adopting techniques like Test Driven Development force you down a path that doesn’t allow it.

A short and to the point post by Dror Helper at Typemock looks at industry findings on the cost benefits of doing TDD.

The Cost of Test Driven Development

..the research proved two points:

1. Using TDD reduce the amount of bugs in the code significantly
2. Using TDD takes more time then not using TDD

This is in line with what was expected, the question is does the extra time doing TDD reduce the quality or is it the unit testing.  Could doing Test Last unit testing give similar results?

There is also the unquantifiable part to the numbers in the quality of the developers in each team.  This has a significant impact on the code quality.  Using processes like TDD help to raise the bar for all of the developers and the cost benefit will improve over time.  It would be a very interesting exercise to see how these numbers work out in a typical SharePoint project.

On the same lines Phil Haack (great name) has a post on Bug Driven Development where he quotes Robert Glass and Steve McConnell on the costs not doing some form of automated testing which reinforces the value in getting the right processes in place.

As development practices around the SharePoint platform mature the demands for better quality and better performing solutions are being made.  Gone are the days when Enterprises will accept that the Roll-Up Web Part they purchased will not scale beyond a few site collections.  A knowledge of the platform and defensive coding alone will not cut the mustard, the need for a holistic approach to your development practices will be the only way to achieve success.  In this blog post I am focusing on one specific technique, code profiling, and using the excellent Red Gate ANTS Profiler.

Disclaimer:  Red Gate ANTS Profiler is a paid for application.  There are other products on the market that do code profiling including Visual Studio.  The reason for using ANTS in this post is due to the ease with which you can get up an running and the quality of the user experience.  I recommend you set aside a day to look at the profiling in Visual Studio and then a few hours to get that much further with ANTS.

Software Performance Analysis – Wikipedia

In software engineering, performance analysis, more commonly today known as profiling, is the investigation of a program’s behavior using information gathered as the program executes (i.e. it is a form of dynamic program analysis, as opposed to static code analysis). The usual goal of performance analysis is to determine which sections of a program to optimize – usually either to increase its speed or decrease its memory requirement (or sometimes both).

As the quote says this technique is specifically aimed at the optimization of code; the tools enable the developer to quickly target the areas of code that need optimization and lead to an improved return for the effort. (seems strange saying ROI when talking about dev techniques, but this is really what it is).

Getting started

Red Gate provide a 14 day trial, so you can try before you buy.  The latest release, version 4.3 as at the time of writing, has made some great improvements over the 4.1 version – most notably the removal of the need to stop all the IIS web sites listening on port 80 .

As you would expect this post will focus on profiling SharePoint code, I’m going to use the Magic 8 ball application developed for my TDD session at the Best Practices SharePoint Conference in San Diego.

More details on Test Driven SharePoint development can be in the Agile section of this site.

Environment

The environment that I will do the testing is a typical SharePoint developers Virtual Machine.

1. Windows 2008 Server (32bit)
2. Visual Studio 2008
3. Microsoft Office SharePoint Server 2007
4. SQL Server 2005

As this is running on Windows 2008 I will be testing against IIS7.   The requirements are identical for Windows 2003 and IIS 6.0

Installation

Download ANTS Profiler and install, not too many options to worry about here.

Luckily for my first run through the team have provided a technical paper on installing, using and getting ANTS Profiler to work with SharePoint, How to profile a SharePoint 2007 Site Collection.  There were a few things in here that I prefer not to do,  I will document what I do below and will pass this back – I’m not criticizing, there’s no way I would even attempt to write a Profiler so mucho respect from me to the Red Gate team

Configuration

We are looking for the ability to profile any site collection within a Web Application.  This involves getting the account used for the associated application pool, giving it some additional rights and then configuring the ANTS services to use this account.

Web Application

Depending on your development machine configuration you will have a number of web applications configured in SharePoint, the configuration below should be reasonably familiar to most of you.

The key here is to identify the web application (IIS Website) as we need to get the account name from the associated Application Pool.

Load IIS Manager, select the web application and chose Basic Settings from the Actions menu on the right.  This will show the associated application pool as below.

Select the Application Pools from the connections menu, choose the application pool identified in the previous step and click advanced settings.  This will give you a properties screen showing the account we need to work with.

In this case 21apps\svc-intranet-ap – if you have a local installation this may be set to Local System in which case the steps below may not be required.   I recommend that dev’s don’t have a default install as you really need to understand how SharePoint is setup to undertake any large scale developments.

Web.Config

In IIS 6 days you would need to edit the web.config file to set the application into debug mode.   In IIS 7 the common ASP.NET configuration options have been promoted to the IIS console.  With the web application still selected click on the .NET Compilation option and change the Debug behaviour to True and click Apply.

Like all good SharePoint developers you will be working with WSS_Minimal trust settings or a custom policy file.  To do profiling however you will need to set the Trust level to Full.  Again IIS 7 provides a UI or you can edit it manually.

It is possible to run the code without setting a Full trust level, but you need to remove the .pdb files and will lose the ability to profile the lines of code.  Which is sort of the point of profiling!  If you don’t set Full trust you will get an error like the one below when loading a web part

Web Part Error: An error has occurred.

Show Error Details
Hide Error Details 

[TargetInvocationException: Exception has been thrown by the target of an invocation.]
  at System.RuntimeTypeHandle.CreateInstance(RuntimeType type, Boolean publicOnly, Boolean noCheck, Boolean& canBeCached, RuntimeMethodHandle& ctor, Boolean& bNeedSecurityCheck)
  at System.RuntimeType.CreateInstanceSlow(Boolean publicOnly, Boolean fillCache)
  at System.RuntimeType.CreateInstanceImpl(Boolean publicOnly, Boolean skipVisibilityChecks, Boolean fillCache)
  at System.RuntimeType.CreateInstanceImpl(Boolean publicOnly)
  at System.Activator.CreateInstance(Type type, Boolean nonPublic)
  at System.Activator.CreateInstance(Type type)
  at Microsoft.SharePoint.WebPartPages.SPWebPartSerializer.get_DefaultControl()
  at Microsoft.SharePoint.WebPartPages.BinaryWebPartSerializer.SerializedWebPart..ctor(SPWebPartManager webPartManager, XmlNamespaceManager xmlnsManager, Byte[] sharedData, Byte[] userData, String[] links, Type type, SPWeb spWeb)
  at Microsoft.SharePoint.WebPartPages.BinaryWebPartSerializer.Deserialize(SPWebPartManager webPartManager, XmlNamespaceManager xmlnsManager, Byte[] userData, Byte[] sharedData, String[] links, Type type, SPWeb spWeb)
  at Microsoft.SharePoint.WebPartPages.SPWebPartManager.CreateWebPartsFromRowSetData(Boolean onlyInitializeClosedWebParts)

[VerificationException: Operation could destabilize the runtime.]

Services

ANTS runs as two services

1. ANTS Memory Profiler 4 Service
2. ANTS Performance Profiler 4 Service

Change the accounts on these two services to the same accounts used in the web application we identified above.

If this is the first time the account has been associated with a service it will be automatically granted the ‘Log On As A Service’ right.

Local Security Policy

From the Administrators menu choose Local Security Policy and grant the application pool account the following rights

1. Act as Part of the Operating System
2. Impersonate Client after authentication

Open a command prompt and run gpupdate /force.

Profiling your code

Your environment is now configured, we can start profiling.  As this is SharePoint I tend to just hit the bit i’m planning to profile first to make sure this is working before I start profiling.  Many hours can be lost trying to solve a configuration issue with a profiler that is not at fault.

Start Red Gate ANTS Performance Profiler.  It knows we’re running on a system a system that has UAC, it offers to make our life easy and always run as an Administrator.  I normally click yes here.

ANTS loads the Profile Settings window is automatically.

There are only a few things to set.

1. Choose the ASP.Net web application (hosted in IIS)
2. The URL to the SharePoint site
3. Use the original port

Further profiling counters can be set, but we just want to get our first run going for now.  Click on Start Profiling.  (In this example I have added a long running process to show how ANTS can identify this as a hotspot).

Profile Results

Profiling will start and a new browser will open at the location specified.  As this is SharePoint the profiler restarts the service so you will always have a small delay while SharePoint is JIT compiled. The profile timeline will show this impact on the processor.

Once the processor has levelled out its time to start profiling your code.   Navigate and interact with the bit of the application you are interested in.  If it is a  web part or navigation load you are interested in then you can refresh the page.

Once you are happy you have performed enough tasks,  I recommend keeping these focused, stop the profiling. ANTS will present you with the HOT code, the bit of code that is the worst performing. You can focus the analysis on any part of the profile graph to show poor performing code around a specific spike in the processor or memory.

In the example above I am focused on the code that runs (including my long running process) when I ask the magic 8 ball a question.

Viewing source code when you have enabled Line level profiling shows every line, the hit count and the % of time taken to perform that operation.

The really cool thing is you can click through the methods to follow the slow running process through the code!

Showing the method grid allows you to open up a call graph for any method to see how the processing is broken down in a dynamic tree view.  This really is just brilliant!  The image below does not do it justice you really do need to see this for yourself to appreciate it.

Conclusion

The long running process I added for the demonstration was picked up and allowed me to drill down into the lines of code that were the slowest.  It even showed how the internal calls within the base classes performed.

Red Gate ANTS Profiler is easy to setup and proves to be an invaluable tool in your software development tool kit.  I suspect profiling will not be something that you do every day, however with the speed and ease of use I would like to think that you will consider the inclusion of profiling a must have rather than a nice to have.

In contrast to the Agile SharePoint development with Scrum session this was very light on the slides.  In good Agile tradition I adopted a Pair Programming approach with me at the keyboard and the audience been my code buddy.  Everyone was provided with a copy of Uncle Bobs Three rules of TDD to help with during the pairing process.  The rules state:

1. You are not allowed to write any production code unless it is to make a failing unit test pass.
2. You are not allowed to write any more of a unit test than is sufficient to fail; and compilation failures are failures.
3. You are not allowed to write any more production code than is sufficient to pass the one failing unit test.

I did have some slides to set the scene (see below), to give some background to TDD and raise the issues around the use of the word TEST.  We also looked at 10 reasons TDD sucks, with a slight SharePoint slant,  and quickly got into Visual Studio where the real learning was to happen.

The user story was to develop a Magic 8 Ball web part that could have custom answers defined in a SharePoint list.  The finished code is available on CodePlex and it is planned to extend this solution over time to expand on the ideas.  My main issue with just getting the finished code is the learning is in the process and it can be hard to see how the design was driven by the code.   I am planning to do a web cast on this to show this process.

CodePlex project now available  http://www.codeplex.com/sp8ball

Winners

The guys at Typemock we very generous in donating 3 copies of Isolator for SharePoint which is an essential component for anyone doing unit testing in a SharePoint environment.  The lucky winners were

Mark Freeman
Shahar Tamari
Kennedy Duncan

Each of you will receive a free license for Isolator for SharePoint,  for those that did not win Typemock are running a special discount.

As promised the slide deck from my Agile SharePoint Development with Scrum.

Abstract

Provide an introduction to Agile development using Scrum and discuss how the iterative approach to development helps the customer to get the solution they want. Look at how this approach works when applied to SharePoint projects, how it helps leverage more of the core platform and focuses effort on the biggest value areas. We will look at the challenges this brings to your development team by doing early integration, dealing with upgrades and changes and understand how addressing the hard things early is the right approach. We will also discuss how Scrum gives visibility of the project and brings both good and bad news. How getting customer engagement is the primary challenge and how the flexible approach is often at odds with the way work is contracted.

Agile SharePoint Development with Scrum

The session was spilt into two parts, an introduction to Scrum and a discussion about how this can be applied to SharePoint projects.  The audience was a good mix of Developers, Managers, ITPros, and business people with a range of exposure to Agile practices and Scrum.  The session provided everyone with an understanding of the terminology and process of doing Scrum.  There was a focus on ‘What is done’ and a good discussion around how you architect solutions.  For those that attended and anyone else interested I recommend you listen to Scott Hanselman’s podcast with Ken Schwaber where Ken covers these topics.  Also looking at the question ‘when do you architect solutions?’ I really like Kens take on how, using the waterfall approach, you have to spend time getting it right upfront. The ability to change later in the process is limited and the costs can be prohibitive.   Using agile techniques the architecture will emerge over time,  you know that you only have the architecture in place to support the business requirements developed.  He poses the question,  if you had defined the architecture up front what happens if it is changed mid way through the project?   With waterfall there is a lot of unpicking to be done and wasted effort.  With agile you don’t have to do the unpicking and the rigours of the process, ensuring things are done, means your code base is easier to change.

Being Agile with SharePoint

Here the focus was on facing the tough challenges early, understanding how to package your solutions, deployments to multiple environments, automation of processes and the eclectic mix of tasks you need to do to deal with the upgrade story.  There was a significant synergy with Ben Robb’s session on automated builds using Hyper-V and Robert Bogue’s coverage of the SharePoint Patterns and Practices coverage of upgrading.

I plan to do a screen cast of this session to provide some background to the slides and also the animations that really took me a long time to do.

This post is part of a series on Test Driven Development – Using Dependency Injection

In this 5 part series we refactored the solution developed in our white paper Unit Testing SharePoint – Getting into the Object Model to make the code more testable, and enable us to work within the available features of TypeMock Isolator for SharePoint.

We started with a review of the code and the specific line of test code that was not supported,  reviewed the options available to us including upgrading to the Full version of TypeMock but decided to refactor the code and introduce a pattern called Dependency Injection.

Before digging into the code we introduced some of the common TDD vocabulary including Loosely Coupled, Dependency Inversion, Dependency Injection and Inversion of Control.   The aim not being to scare people off with fancy terms but to help understand what these mean and the affect they have on your code.

The refactoring of the code began in Step 1. where we started to implement Dependency Inversion to give more control to our calling code, in this case our tests.  We introduced abstract classes and the use of an Interface to support this.

Step 2 Implemented dependency into the GetRandomMessage method and also the concept of creating Fake classes that can be passed in to support the tests we wish to perform.  At the end of this step we were able to refactor our initial failing test and all the tests passed.

The code for this refactor can be downloaded from here.

This blog series proved a valid exercise in refactoring existing code to make it more testable and also demonstrated that the reduced features of Isolator for SharePoint, at a reduced price point, do allow you to do TDD in SharePoint.

Whats next?

Andrew Woodward, follow him on twitter, will be speaking at the Best Practices SharePoint Conference in San Diego on in Feb 2009 looking at agile SharePoint development with Scrum and Test Driven SharePoint Development.