Monday, January 27, 2014

Milwaukee Agile Meetup - Continuous Integration Panel Discussion Summary

Last week I had the pleasure of being on a panel discussion about Continuous Integration at the Milwaukee Agile Meetup. My fellow panelists were Carl Schrammel, Adam Black, Jason Duff, James Carpenter and Adam Johnson.

The discussion grew out of the group's growing interest in this area and challenges that individuals and teams attending the Meetup had expressed recently. I know several people who said they had wanted to attend, so I thought I would post a summary of the discussion.

Q: What does Continuous Integration mean to you? When were you first introduced to the concept and how did it change your perspective?
A: After the initial text book definition was given ("Continuous integration (CI) is the practice, in software engineering, of merging all developer working copies with a shared mainline several times a day." - Wikipedia), the group continued to express their take on CI.

  • CI is a powerful feedback tool for both engineering as well as project management - reduces surprises due to integration and deployments and makes the ones which occur more manageable.
  • The benefits of CI are best reaped via automation, but even manual CI can bring great improvement.
  • CI can help build confidence in the product and the team - one always knows the state of the code at the end of the day.
The distinction between Continuous Integration and Continuous Delivery (CD) was also drawn... with Continuous Delivery being an advanced practice where a team is able to deploy changes to production with high frequency. 

Q: What tools are being used?
A: The standard list of revision repositories, build servers, testing frameworks, continuous integration servers, deployment automation were quickly run through - basically if you are interested in CI there is likely a set of tools readily available for your language/tech stack of choice. It was noticed, however, that there are software products and systems which are not necessarily easy to integrate automation or other CI tools too. Particularly these tend to be configuration based systems, or closed systems which perhaps try to manage revision control internally - preventing access or control by external ci/build/test setups. This led to the discussion that aside from tools, the other aspect of CI is the practices and behaviors. These behaviors are often harder to implement than the tools. However, if the behaviors are embraced by a team, even the most simplistic of CI tooling can become powerful. The further consensus of the panel was that it is best to start simply - implementing behaviors and supporting tooling gradually. In this way, each advancement should be lauded as a positive win in the right direction rather than disappointment at not having achieved an elusive ideal CI or CD state in one try.

Q: How to prove the value of CI to developers? to business/product owners? how soon can one expect to see value?
A: The answers to this echoed some of the previous discussion. The general feeling was to take each instance in a case-by-case basis and look for opportunities to introduce behaviors and tools pragmatically to provide greater insight into specific areas where a team may be experiencing pain. Examples: recurring bugs, painful massive code integrations, lost code, misunderstood and mis-implemented user requirements, painful coordination/integration with other systems/teams on deployment, out of sync versions of code between environments (Prod getting ahead of Dev or Test), not being able to do certain types of testing. All of these were brought up as pain points which some degree of CI could have alleviated. 

Since CI sits on the front lines of ever changing code, the positive impact of CI can be seen very quickly (as soon as one build cycle in some cases). The key is to communicate this to the parties who would most appreciate the benefit. This is where a CI server can help by providing dashboards and automated messaging, but even utilizing the low tech approach of holding stake-holder meetings before and after a CI 'event' can shed light on the benefit of the activity.

The panel agreed that in order to get the buy in to do more advanced levels of CI requires the building of trust, something familiar to Agile Coaches and leaders everywhere. However, even the most rudimentary CI practices can provide excellent data and feedback which, if communicated properly, can facilitate the building of that trust. Due to the fact that CI can generate data about the health of the deliverable every time a change is made, it can be one of the most powerful and prolific sources of data for information radiators. The trick as with all data is to mine it for the most relevant information to the message you wish to communicate.

Q: Who owns CI? Who sponsors/drives for CI?
A: The panelists seemed to generally agree that in most large project teams, there was a dedicated Operations team in charge of maintaining the CI tools (build servers, repositories, deployments and deployment automation, ci servers, etc). However, many expressed how this could lead to the negative effects of siloing. Panelists shared their experiences with teams who used various means to combat this. Rotation of members into the ops team, rotation of ops duties among teams, the use of embedded testers to serve as the bridge between dev and ops, and use of separate scrum-of-scrums to coordinate dev teams with ops.

We did not get too much into the sponsors for CI. I would suggest, as with trying to get buy in, any individuals or groups who feel they would get a benefit out of the use of CI could sponsor it. I have seen CI started as an engineering initiative and I have also seen it sponsored by Project Management and even Product Owners.

Q: Of those doing CD, has anyone automated the signoff process in tightly controlled organization/industry?
A: No one had seen full blown automated sign-off at the level of the typical product owner, upper management deployment committee. However, several had seen groups who had automated sign-off at lower levels, such as moderate UAT sign-off. This can help pave the way to make the manual sign-off at the executive level easier once trust has been established through a record of reliable test/dev/pass/sign-off cycles.

The evening flew by and that was about all we were able to get to. Prior to the discussion, David Neuman from Redpoint Technologies sent out a list of questions to the panelists that would serve to seed the discussion. I think we hit most of the items tangentially. If anyone would like to discuss them further, leave a comment or email me.

The last question on the list asked if there was one recommendation we have for successfully implementing and leveraging CI. 

Build Automated Tests. Even if you still have to trigger your builds or other steps manually, having automated test suites that you can run against your built code is one of the most powerful tools to start getting an ROI on your CI efforts. Furthermore, many of the behaviors and practices of CI which are difficult to master initially revolve around disciplines of TDD. If you have automated tests, it is easier to start to instill these disciplines in your team. As some of the other panelists and I were discussion after the meetup, automated TDD is fun... it makes coding like a game with almost instant gratification to our problem/puzzle solving efforts. Making something fun can really make developing good habits to support it easier.

Wednesday, January 22, 2014

Agile Martial Arts to Prevent The Project K.O.

Recently, I have been having a number of discussions with colleagues about our projects and I noticed that there was a similarity in how we were discussing using the communication and feedback tools of Agile Development with concepts of Martial Arts.

When training in martial arts, one comes to learn that one of the most powerful tools one has in dealing with an adversary is to control distance and time. One can either increase distance in order to give one more time to analyze and deal with an adversary's attack or one can reduce distance and/or time to limit your adversary's options and ability to develop an attack.

Similarly on a project, surprises are seldom a good thing. However one cannot know everything which is going to happen. Change management strategies have evolved to try to deal with this, however, traditionally these have been largely reactive... like a novice at Close distance flailing wildly to block each attack. So how do you control time/distance to problem fallout on a project? 

This is done via the use of feedback loops of varying frequency. Feedback loops of lower frequency can give you the time to "step back" and assess changes before rushing in. This can called a Distance/Time giving action since they give you more time to assess a developing problem, but that means they may also allow the problem to develop further. Feedback loops of higher frequency allow you to address issues immediately, preventing them from ballooning into larger problems or even implementing proactive solutions to stop the problem in its tracks. These are Distance/Time takers, since they shorten the time allowed for a problem to develop. The number and roles of individuals who receive and deal with the feedback initially also plays a part in whether a technique is more giving or taking in nature. Generally, lower number of individuals whose roles are closer to the source of the problem will be a taking action, while the converse is more of a giving action.

Some Distance/Time giving techniques (lower frequency loops):

  • Daily scrums/standups
  • Iterative development/delivery
  • Regular demos
  • Regular retrospectives
  • Pairing

Taking a step back to gain distance and time is one of the simplest defense moves. Appreciation of the different threat levels each distance presents and the ability to back-out to a safe distance is something which is taught to beginners. Similarly, the techniques for gaining Distance/Time in Agile Development are simple enough that any team can begin to use them right away. 

For example, a team is following a more traditional project management approach with few touch points across the team. Perhaps the PM only reports to a board  the status of things as dates for delivery of milestones occur. At this point, it is usually too late and if a problem has occurred, the PM is delivering the surprise bad-news. Even worse, a milestone is delivered only to have the product owners report all the surprise deficiencies. Obviously, both of these are uncomfortable for all involved and panic may often set in. By establishing short, frequent and regular meetings with the team including all stakeholders, a team can get the opportunity to see the problems coming and properly deal with them. They are effectively giving themselves the time to deal with the problems rather than rushing headlong into the fray. 

A recent client complained that historically, every time they tried to do a deployment the tools the development team used seemed to create surprise errors. Moreover, items were added to the new environment without the client's prior knowledge. These would not necessarily break anything, but they did add confusion and wasted time. In order to address this we established regular meetings with tech leads, the deployment lead, the clients and testing. At the meeting, proposed pending changes were discussed. A draft-deploy was reviewed for surprise dependencies which might be pulled in by the deploy tool. Team members who previously never spoke to one another now communicate regularly around a problem they were all experiencing from different perspectives. Out of this, the team has developed a much more rigorous deployment practice, surprises have been eliminated, and all parties have gained a much better understanding of their deployment tools and an appreciation for each others difficulties in deployment. It is not as much a typical Change Management/Approval board as it is an opportunity for the interested parties to communicate. Meetings never last more than half an hour and usually are about fifteen minutes, but we meet at least once a week. While not one of the traditional Agile Team Meetings, it follows the spirit of frequent cross-communication which is so vital to the Agile Approach. 

Some Distance/Time taking techniques (high frequency loops):

  • Automated Testing
  • Test Driven Development
  • Acceptance Criteria/Tests
  • Test First Development
  • Pairing
  • Continuous Integration
  • Daily scrums/standups

Distance/Time taking techniques are usually more advanced in Martial arts as they require a bit more training and discipline to do safely. So too in software development. Most of the ones I listed here obviously employ development practices which software engineers will need to hone if they have not done them before. However, once proficiency is obtained in them, these practices can be life savers and even keep at bay the dreaded pager. Utilized well, these operate as a means of clinching problems by stopping them well before they develop into much larger problems in a more devastating environment. Use of Automated testing with a modern CI server can not only identify problems in with new code early on, it can also help make sure that once a bug is fixed, it stays fixed. One of the mantras on a team I was on was "Find a Bug, Write a Test." Basically, whenever a bug was reported, the developer wrote a test for how the solution was supposed to behave. Of course, until the bug was fixed, the test would continue to break. However, the bug would likely never resurface again since our test coverage of our code had now improved. This of course requires the developer discipline of running tests before committing, as well as a Test-First mentality. Without those disciplines the team is back to reactive flailing.

These techniques do not always stop problems from occurring, but like the clinch, they can prevent the problem from developing into a more lethal one. A team with good CI, TDD, and Acceptance testing practices for instance, can deal with and minimize the impact of major architectural changes or even severe problems introduced by changes in third-party software. They can tackle the severe problem in manageable chunks at a level of relative safety. 

I listed daily scrums and pairing under both because depending upon the situation these techniques can act both to give time to deal with potentially developing problems as well as to stop problems from starting in the first place through open communication. The latter is usually the case where someone either states what they are working on or having difficulty with and someone else on the team either gives them a solution or can help them avoid pitfalls they have experienced before. As it happens at such a frequent schedule, the time for problems to develop is often completely eliminated - effectively clinching the problems. Scrums and Pairing as time gaining operate in the sense that by approaching problems as a team tends to prevent us from making rash decisions.

A final note: you will notice that I do not say Pair Programming... while I like Pair Programming, many of my colleagues have noted that pairing is a powerful tool for all team members. I have seen PM's, BA's testers, even Product Owners pair on deliverables with great results.