Heroku founder Adam Wiggins outlines their vision of next-generation application deployment architectures. He describes the challenges and choices Heroku has made and the upcoming technologies and capabilities that are exciting in their space.
Jason Huggins, Al Sargent, and Steve Hazel from Sauce Labs describe their Selenium-in-the-cloud testing framework, including integration with CI systems like CruiseControl.
- Functional tests running slowly after upgrade to Rails 2.2
A project upgraded to Rails 2.2 which was previously able to complete its functional suite in under 300 seconds now takes a leisurely 6000 seconds to finish. Speculation about possible causes is welcomed. Meanwhile, I suspect ghosts.
UPDATE: apparently the functional tests were running slow due to memory swappage (the memory swappage, in turn, caused by ghosts, no doubt).
- Rails 2.2 may break Selenium tests
- FixtureScenarios and slow test suites
A pivot installed FixtureScenarios on a project and found that the addition of the plugin alone (with no configured fixtures) doubled the time the project’s test suites took to complete. FixtureScenarios are now dead to Adam. Dead!
- No Pivotal Breakfast tomorrow
Karen Tsui and Ling-Wen Chang tirelessly prepare the world-famous Pivotal breakfast each day and generally make the San Francisco office an awesome place to work. They’re finally taking a much deserved break tomorrow, so now is a wonderful time to publicly recognize their much appreciated work. Thanks, and happy holidays!
- Helper tests failing after Rails 2.2 upgrade.
A project was upgraded to run on Rails 2.2 and now has a handful of hate for helper tests. The tests in question use a stub Controller to generate a template and end up getting some nil variables. cache_classes is off. Suggestions welcome.
- Pivotal Tracker Team Strength Reminder
If team members are out sick or on vacation, you can adjust your team strength settings by clicking on the blue icon at the end of the current iterations release marker. This will recalculate your current iteration velocity accordingly.
- ‘flash’ Partials
It is bad form to name your partials ‘flash’, because they attempt to populate a flash variable which has the nasty side effect of blowing away your notification array. Naming a partial ‘phlash’ produces no adverse effects outside of looking a little silly. ‘shazzam’ and ‘booyakkasha’ are also acceptable although somewhat less descriptive.
- Capistrano and Git
A default deploy with capistrano and git appears to prefer using the local working set instead of the configured repo (in this case, github). This may end up surprising you a great deal if you do a deploy from a machine that isn’t an exact copy of HEAD. It was suggested that best practice is to deploy tags from a dedicated CI box rather than a development workstation.
- Ruby On Rails Noteworthy Applications
http://rubyonrails.org has a new design, and with it came a page of noteworthy apps running RoR. Are you a Pivotal client who’d like to see your site up there? Get in touch!
Here’s a good, quick read. It got its start about 10 years ago:
- Modular Design
- Tolerance (”Be liberal in what you require but conservative in what you do”)
- Test of Independent Invention (”If someone else had already invented your system, would theirs work with yours?”)
- Principle of Least Power
When you’re heads-down doing Agile or OOP sometimes you find yourself accidentally assuming that certain useful general principles are special to what you practice – when the truth is they’re probably not even unique to your discipline, and some Greeks wrote them down around 500 BC.
One project reported a problem with the enumeration mixin when upgrading to Rails 2.2. The problem was with the all method.
The enumerations mixin allows you to treat instances of your ActiveRecord models as though they were an enumeration of values.
The proposed solution is to switch to in memory enumeration instead of using ActiveRecord.
Great article comparing Pivotal Tracker to Mingle, as an example of a tool that simplifies things, rather than trying to manage complexity (and getting in the way):
Thanks for the positive feedback, Matthias!
Ask for Help
A new project would like to find a useful deploy.rb for Engine Yard. Preferably using git. There were a few suggestions from other projects to view theirs.
Do not create records in a
The transactional wrapper around the tests don’t apply to the before(:all).
Ask for Help
“How can asset_host and retardation_inhibitor could work together nicely?”
Tells your Rails application to generate external resources’ links pointing to the host of your choosing.
The general opinion was that any sophistication should be handled by asset_host configurations in the appropriate rails environment files.
This lead to the sticky problem of our CI process running the Selenium tests in the development environment rather than the test environment. The asset_host using the development environment configuration rather than the test environment.
Everyone agreed that this wasn’t the best way to run the selenium tests.
A desperate request to rename the Pivotal project “retardation_inhibitor” met with no resistance, much to the delight of the Pivot responsible for explaining it’s usefulness.
“New Relic & Desert?”
Ruby & Rails can sometimes appear to be moving so quickly that not advancing feels like standing still. A project found problems with the latest versions of New Relic and Desert.
Every page request resulted in a stack trace in the log.
The general consensus was that the latest version of New Relic would solve this problem. Of course the team had already tried this along with several other versions.
Some Pivots in New York had apparently encountered and solved similar problems. Hopefully they can help out.
Another Pivotal project had similar problems and had to upgrade to 2.7.3. They used these update steps.
They originally tried to just install the latest without following the steps outlined in the link above and still had problems.
Look out for a public Pivotal Pulse announcement soon.
The world has moved on, and so has the state of the art in remote pairing. I work remote 75% of the time, so this is an important topic for me. The setup we have now is working pretty good, so I wanted to describe it for the benefit of other remoties. Also, I’m only going to describe one specific setup – the one that is currently working best for me. So, here it is:
Computer and OS
Macintoshes running OS X 10.5 (Leopard) and maxed out on ram (at least 2 gig+), with a second monitor, ideally a 24″.
Pivotal uses 24″ iMacs exclusively. This lets you have your remote pair’s screen up on the second monitor, while still having the primary iMac screen available for local work (configuration, looking stuff up, temporary soloing, etc). You must be very disciplined and up-front – you should be explicit about when you are paying attention to your pair and when you are not, but that’s a whole other topic.
Also, at my home office, I have an iMac which I use for screen sharing, but I usually run the Skype audio/video on my MacBook. This takes the CPU load off of the iMac, which is important, because Skype is a CPU hog (more on that later).
Screen Sharing Software
Apple Screen Sharing which comes with Leopard.
Apple has done a nice job making the screen sharing app perform well, and the most important feature for any remote pairing screen sharing tool is performance. In my opinion, it performs better than any other VNC client on any platform, assuming you have a good connection (possibly excluding some windows native-video-hook solutions like UltraVNC or Remote Desktop, but there’s no way I would ever use Windows for development). All VNC clients which use the standard RFB Protocol can only be tweaked so much, and will only give you mediocre performance. However, there are several annoying bugs and gotchas with Apple Screen Sharing:
- It is hard to find. Look under /System/Library/CoreServices/Screen Sharing.app. It is easiest to use Spotlight/Quicksilver or drag it to your dock. Supposedly you can start it with iChat but this never worked for me, I had to run the app directly.
- Most of the useful features are disabled by default with no way to access them via menu or toolbar buttons. This is an amazingly annoying decision by Apple, but it is fairly easy to hack the toolbar buttons back into existence. Here is a Macworld tutorial showing two options.
In general, it seems that the best remote-control tools are those with some sort of native/low-level GUI integration: Leopard Screen Sharing, NoMachine NX, UltraVNC, Windows Remote Desktop, etc. Higher-level platform agnostic tools (like standard VNC/RFB protocol) just don’t perform as well – no matter how much you tweak the available color/depth/etc settings.
Plantronics GamePro USB Headset
This is a great headset, and you need a really good headset if you are going to wear it all day, every day. Cloth earpieces, mic cover, very long cord, and I believe it also has some echo cancellation built in (there’s a huge box inline in the cord that does something). Unfortunately, I don’t see the exact model on the Plantronics website anymore. It may be replaced by the “GameCom” model, but I haven’t tried this.
Built-in iMac Microphone/Speaker
The built-in microphone and speaker on iMacs is really good. If you want to talk to a group of people remotely (for example, project standup), this is the way to do it. However, if the ambient noise gets too much, you can switch back to the headset.
You can even combine the two. For example, if you want to hear the surrounding conversations, but your pair is having trouble hearing you over the noise, then can wear the headset, but still keep the input set to the built-in iMac mic.
Sometimes you will need to adjust the input/output levels to reduce echo, and the remote pair should handle this themselves – they know what it sounds like.
Built-in iMac camera
Just like the built-in iMac mic and speaker, the iSight is a great camera. A detached iSight is just as good, if you want to be able to move it around or aim it without moving the computer.
Skype is the best I’ve found. It does have drawbacks: it crashes rather frequently, it sucks a lot of CPU, it can do bad things to your network if you become a supernode, and it doesn’t support video in conferences.
However, it has great echo cancellation, it is free, and easy to use. The echo cancellation is really the most important thing – all other audio conferencing tools I’ve tried seem to have much more issues with echoes – even when you are using echo-cancelling hardware devices or speakerphones.
Some people seem to like iChat, but I have not had good luck with it. It takes longer than Skype to connect, the echo cancellation is not as good (sometimes it is, sometimes not), and most annoyingly, it doesn’t always close. I often end up having to force quit it – which is even more annoying when it is stuck on a freeze-frame of me making a stupid face or scratching my nose. Skype never does this – video always goes away when you shut your video or kill the call.
iChat has video conferencing, though, which is a benefit. You can sort of work around this by putting up the video preview in iChat, and having multiple remote people connect to view it via screen sharing, if you only want to see video for one of the participants (e.g. a couple of remote people calling in to a company meeting).
Network, Network, Network
This is the last but most important component to usable remote pairing. A fast, low-latency network connection is critical. I don’t have any numbers, but I believe that low latency is at least as important as high bandwidth. I also (without proof) believe that ping is not necessarily a good indicator of latency – I bet it is possible to have a good ping (ICMP) but still have issues with TCP/UDP latency. Who knows what’s going on in the tubes between you and your pair? Any data, tools, or insight on this would be very welcome.
As empirical evidence, for the first year or two at Pivotal I had DSL, which was pretty fast with low ping, but had continual problems with performance. Then, I switched to corporate-grade cable with a significantly higher bandwidth limit. My experience improved dramatically and my problems decreased greatly. This was about the same time I switched to Leopard screen sharing, so I think that had something to do with it, but the better connection definitely made a huge difference. Again, sorry I don’t have more concrete numbers, but I will guarantee that the better your connection, the better your experience will be.
Also, if you are in a corporate network, this may cause you problems. Even if there is a big pipe to your location, there may be saturation on your local LANs or intranet. Again, no hard data, but this is backed up by experiences of having consistently better performance when connecting to another remote at-home pair with a good connection as opposed to connecting to the Pivotal office which has a much larger pipe.
Remote Pairing Presentation at RailsConf 2008
- At RailsConf 2008, Michael Buffington and Joe O’Brien did a good presentation on Remote Pairing. This is a very good presentation which covered many important aspects of remote pairing, as well as presenting some innovative ideas. Unfortunately, I don’t see a link to the preso, please post one in the comments if you have it.
This isn’t meant to be the be-all, end-all set of recommendations, it’s just what is working pretty well for me now. By “pretty well”, I mean that I can be an efficient pair, even when I’m driving the remote machine.
However, I’ve learned to cope with a lot, and adapted my work habits. It has forced me to become much better at communication, and describing what, why, and how I am programming. In general, though, I believe that remote pairing is physically, emotionally, and intellectually taxing. Regardless, I personally deal with it because Pivotal is such an awesome place to work and Pivots are such incredible developers. Most importantly, I come out in person for a week every month, attend retrospectives and brownbags, have some beers, and generally stay “entangled” with the rest of the team in person. If I was 100% remote, I don’t think I could handle it long-term.
So, I hope this helps out all the other remoties out there. Please let me know what you think, your experiences, and what works well for you.
Recently, one of the Pivots was called out for that civil duty of jurification. That is, he was called to sit and decide on trial.
Now, my only interaction with juries thus far has been to watch depictions of them in media, but while many seem to speak about trying to get out of this duty, I think I’d consider it an honor. It seems to me that being the victim or accused in a crime is a time of extreme need – we know that it has happened that the wrongly accused have been found guilty, and likewise, that guilty have gone free – and while no system is perfect, one of the few protections against these outcomes is the presence of a discerning, unbiased jury.
As such, I’m glad to know that a Pivot was out there, doing their civic duty, and I’ll be happy to do the same if the occasion arises. But the interesting thing which came out of this were his requests at Standup one morning: that there’s a duty non-jurors have to jurors over the time of the trial. Which is,
not to ask about the trial: Jurors are asked not to speak about a trial to protect the decision-making process from undue influence. There are strict rules about what evidence is admissible in a court-room, to protect one side or the other from presenting faulty evidence such as hear-say. Discussing the trial introduces the risk of improper influence, and it’s best to respect the process by leaving the subject aside.
not to joke about the trial: Trials range from the frivolous to the truly weighty, and jurors are sometimes asked to shoulder an intimate knowledge of event not discussed in polite company. Also, the consequences of a jurors decision are serious and not to be belittled. It’s best to avoid making light of the situation.
So keep these in mind, the next time a friend or colleague serves on a jury; it’s a serious act for which they deserve your deference and respect.
How my outlook on coding in Rails has changed over the past few months!
When I made the switch from Java to Rails a few years back, I, like many of my fellow Pivots making that same well-chronicled transition, delighted in the ease with which we could suddenly knock out a web app. How we cheered when our object-relational mapping took zero lines of code! How we applauded when we declared our model object validations in near-English! How we roared with laughter when convention viciously slapped the face of configuration! And how we shook our heads in dismay when we realized that our new development environment appeared to be from the mid-, if not early, nineties.
For, while we had arrived in a brave new world of minimalist declarative meta-programming, rapid prototyping, and an new-found sense of productivity that made even the most nimble forms of Java development look like wading through a morass of slimy boilerplate code and endless XML, we soon realized that the IDE situation was less than awesome. Our productivity was overall much improved, but we had taken a huge step backwards when it came to the act of writing – and especially changing – code and tests. Overnight, we went from living it up in a paradise of automated refactorings, seamlessly inbuilt test runners and powerful debuggers to roughing it with a text editor that, to our spoilt eyes, appeared to offer barely more than code highlighting and support for homemade macros.
Not only didn’t our favorite Java IDE, IntelliJ, not function well with Ruby, but nor did Eclipse, and nor did NetBeans. Early on, each had some nominal support for Ruby, it’s true, but it was mostly just code highlighting and some basic navigation. If we wanted to run a test, we had to (horror of horrors) leave our IDE, go to a shell and run a command (mapping a key to an “external tool” was cold comfort, it seemed to me). If we wanted to rename a variable, we had to do it manually. And if we wanted to debug something, well, the only option was to use a tool that appeared so prehistoric that we simply didn’t do it.
In fact, things were so bad that if we were honest with ourselves, TextMate sometimes looked like a better option that the IDE we knew and loved. We pleaded with Pivots who didn’t come from the world of Java to use IntelliJ, if only (for them) for its awesome global search, and if only (for us) to make ourselves feel comfortable in the new world of Rails. But IntelliJ had been rendered so impotent by its Ruby Kryptonite that it felt at times as if we were simply dragging around a comfort pillow that had lost its stuffing. All the progress that had been made in the world of Java IDEs for so many years seemed to have been lost. Woe was certainly us.
But then came some signs of life. NetBeans and Eclipse (in its various forms) were making progress in Ruby-land: they had test runner integration and debuggers that worked. Big steps forward indeed. While I suspected that I would probably miss IntelliJ, I nevertheless gallantly tried to commit to one of them: if they were going to offer me something even close to the power I used to have in my Java IDE, I was willing to put up with a lot. But despite my best attempts to be patient, I was disappointed time and time again. Eclipse still felt clunky compared to IntelliJ. NetBeans still suffered from a problematic global search and less-than-perfect VCS integration. They are gallant attempts, and they deserve credit. They’re both great products and they both serve communities that no doubt find them invaluable. But in that very unquantifiable, personal, emotional way, they weren’t what I really wanted. I wanted that IntelliJ feel.
So where was JetBrains on this? There was the Ruby plugin for IntelliJ, which was making some great strides forward, but that always felt like a bolt-on solution at best. What was more, it often didn’t work with new IntelliJ updates.
Imagine my surprise and delight, then, when Christmas came several months early this year. Suddenly, in late 2008, we are quietly presented with RubyMine. IntelliJ for Ruby? Surely not! I dared not hope for too much when with some trepidation I downloaded an early pre-EAP candidate a few months ago. But, joy of joys, it turned out that it was, or rather, is gradually becoming, (almost) all I was hoping for. Built-in test runner, debugger, pretty good refactorings – all things that other IDEs provide, but in that special JetBrains way that is so much more intuitive. Great success!
Now, I realize that all this probably sounds like a gush of praise that is not yet due, and perhaps that’s true. It’s certainly true that there’s a long way to go: RubyMine is lacking in many ways. But the signs are there that things are moving hastily in the right direction. Recently, each build that has come out has been noticeably better to me than the last (additional useful features, stability, performance). But more importantly, I feel like JetBrains is on the right trajectory with RubyMine. We now have a path towards a great IDE for Rails. I think we’re back in the land of the living, and I’m hopeful that between Rails and RubyMine our productivity will soon jump higher than we could have imagined just a few years ago.
OK, start the “All you need is Emacs”, “VIM is the best thing ever”, and “TextMate rocks, what’s this IDE nonsense?” comments below :)