Measure (At Least) Twice

20180218_143538 - SmallBear with me here. I promise this is about software – but first I need to talk about building.

I have several uncles and uncles-in-law who are builders, and when I was younger I ended up helping out with minor jobs here and there. One of the most recurrent mantras was this one:

“Measure Twice, Cut Once.”

The idea, if it’s not immediately obvious, is that you double-check your measurements before committing to an action which is costly or hard to undo. Once you’ve cut that beam or board, you can’t un-cut it – the board will stay cut despite your best efforts to turn back time. If you cut it too small, you can no longer use it for what you planned to. If you cut it too big, you have to try again. Measuring is cheap – cutting is expensive. Measure twice, cut once.

I consistently see software engineers fail to even measure once.

I once reviewed a pull request from a developer who had decided to optimise part of our code base which performed intensive weather modelling. He had spent a week on it – and he had spent that time re-writing a large number of string plus operations to use StringBuilders. In some cases this can provide some performance benefit – but the re-written code was in the setup and teardown phases of the modelling. Setup and teardown took only seconds, while the modelling itself could take minutes or hours. Worse, this developer had introduced several bugs into previously-working code; but that would make this an anecdote for a blog post about the importance of testing and safe refactoring, and this is about measuring, so let’s focus on a different “worse”. Not only had he optimised the wrong part of the code base, but because he hadn’t measured performance, he couldn’t demonstrate that he’d actually made anything faster (aside from some vague assertions that StringBuilder is faster than string-plus, which isn’t always true).

I didn’t even need to measure to know that his change wasn’t going to make things significantly faster – the code wasn’t in a loop, and I knew that the overall modelling time was long. Shaving off microseconds at a time in a scenario like that is a losing game. The other reason I knew for certain that his change wasn’t going to pay dividends was that I had already spent quite some time improving the efficiency of that setup process. We had benchmark tests around parts of that code, and we had made significant wins. Most of them were had by fixing the .GetHashCode and .Equals methods of a custom coordinate class we were using. We were wrangling a fairly large amount of data into data structures needed by the modelling code, and set operations like .Contains were taking far too long. It turns out both .GetHashCode and .Equals were creating additional instances of the class to perform certain comparisons, and so we were churning the garbage collector as well as performing far too many operations. Some careful tweaks to those methods – operations performed many times within nested loops – took our setup time down from minutes to seconds. We knew that we’d made things better because we were measuring – but the benefits of measurement go much deeper than just knowing how you did.

All developers fall somewhere on the spectrum between, roughly, terrible and genius. I’m not sure where on that spectrum I fall (I suspect it varies depending on what sort of programming I’m doing), but I am pretty sure that I wasn’t good enough to read through thousands of lines of C# across many different files and realise, through pure mental analysis, that we had a performance problem in our coordinate comparison code. Even if I had noticed that it wasn’t efficient, I wouldn’t have been sure how much the problem was contributing to our run time (as compared to, say, our inefficient string concatenation). Instead, I found out by measuring.

In this case, I used one of the simplest of all code profiling techniques: the pause button in my debugger. Simply run your code, and sometime during the long-running process, hit pause. The line you’ve broken on is extremely likely to be part of the problem. Typically, it will fall into one of two categories:
1. A line which is executing many times. In this situation, something further up the stack trace is calling the code you’re looking at inside a large (or nested) loop.
2. A line which is taking a long time to execute. Perhaps it’s hitting disk or network and waiting a long time before continuing.
There is a third (statisically less-likely) option: you’ve just happened to break execution somewhere that isn’t part of the performance problem. That’s why you measure (at least) twice. Break your long-running operation a few times, and compare the stack traces. If your stack dump consistently looks the same, you’ve found your performance hot-spot.

That’s one of the simplest ways of measuring your code, and it may suffice for tracking down hot-spots, but there is a lot more to measurement than just hitting pause in your debugger. Do you suspect you have a database problem? Don’t go hunting through code looking for your favourite anti-patterns. Instead, measure the number of database round-trips your operation makes. Use your database tools to identify expensive queries – and even once you’ve found them, don’t necessarily just start staring at code. Are you missing an index? There are tools which can tell you. Even if the query is bad, looking at the query plan will tell you exactly which parts of the query are causing the problem, so you don’t need to guess.

Maybe you are hitting multiple services and combining the results? If you’ve just learned about the async keyword, you might be tempted to dive in and start making everything async – but if almost all of the delay is coming from a single call, it might not help much (you’d be better trying to find out why that service is so slow). If you really are dealing with multiple similar-cost calls which you combine after they’re all complete, async may be a good investment of your time – but you won’t know until you’ve measured.

For one weather-forecasting algorithm I measured, I didn’t turn up anything obvious. I measured and measured some more, and eventually I discovered we were utilising the CPU cache poorly – the bulk of our algorithm time was spent fetching data from main memory after cache misses. Updating the algorithm was a very costly measure, but investing in hardware with better memory bandwidth gave us some short-term wins. There are very few developers who can look at an algorithm and spot a cache utilisation problem, but measurement can find it for you.

I think I understand why some developers prefer to dive right into code, instead of measuring the problem. We like to think we’re smart. We know stuff about coding, and we’re good at it, and so we can jump in and spot what the original developer didn’t. We can use the sheer power of our minds to solve the problem. It’s a little like the builder who drives in nails with his forehead. It might be impressive, but it’s not very effective, and you’ll end up with a headache.

The opposite problem happens too: we make the mistake of not valuing our time. “What can it hurt to spend a week re-writing those string-plus operations as StringBuilder operations,” we ask ourselves. It makes our code better (by one measure, I suppose), and it won’t hurt performance. But if we spend a week of our time, was it really worth the investment? Measure up front, so you can make that decision based on facts instead of guesses.

Perhaps some developers just feel like they have great intuition. One of my uncles was actually pretty good at estimation: he’d cut timber a shade too big, and trim it down as he was placing it. He only did that occasionally, though. A lifetime of building told him when it made sense to trust his estimation skills, and when to measure twice, cut once.

So we’re back to builders, and here’s the deep lesson. If a builder will measure twice to avoid wasting a bit of cheap wood, how much more should we measure to avoid wasting precious time? I’ve seen weeks wasted on the wrong problem, simply because someone didn’t bother to measure.

Please, my fellow developers: measure (at least) twice.

Hooking (Hacking?) the ASP.Net FileChangeNotifier

I’m going to tell you a little bit about a package I’ve just put together which lets you exclude certain files from the ASP.Net change notifier (the thing which restarts your IIS app pool whenever you change certain files). On the way, we’re going to dig into some internal framework code, and I’m going to horrify you with some code that should definitely never get run in production! Let’s start with the inspiration for this work.

A Bad Dev Experience

The dev experience on one of our legacy projects had been annoying me for a while: whenever I triggered a Gulp build, my IIS app pool would restart. That sometimes had (annoying) flow-on effects, and we ended up with various work-arounds depending on what we were doing. It was never a priority to fix, because Gulp builds only happened on dev and CI machines, never in deployed environments – but one week it really got in the way of some work I was doing, so I decided to get to the bottom of the problem.

You may know that ASP.Net will restart your application whenever there are changes to your web.config or any files in your /bin folder. What you might not know (I didn’t) is that it can also monitor various other files and folders in your application, and what it monitors doesn’t seem to be well-documented. Our Gulp build dumps the output into a /dist folder, and ASP.Net was restarting our app every time any file in that folder changed. After a little Googling, I discovered a StackOverflow answer which gave a slightly-hacky way to disable the ASP.Net file monitoring, which I did – and it worked. No more app restarts!

Not So Fast…

After a few hours of front-end work, I needed to make some back-end changes. I finished wiring up my API endpoints, ran the project, and expected success! No luck. After some more tinkering with the front-end, I suspected my C# code – but after some checking, it was definitely correct: I had the unit tests to prove it. Eventually I fired up the debugger, and found my problem: the debugger was reporting that the running code didn’t match my file. Disabling file change notifications had also disabled notifications on the /bin folder, so my rebuilt DLL files weren’t being loaded. I wasn’t prepared to accept rebuilds not triggering reloads, so I needed to revisit my solution.

Before I went too far, I wanted to be really sure that I was looking at the right problem. It turns out someone going by the name Shazwazza has already done the hard work to build a list of files and folders ASP.Net is helpfully monitoring for changes, and yes – there is my Gulp output folder.

What I really want is to be able to preserve all of the existing functionality – but skip monitoring for certain paths. I investigated a few different switches and options, but none of them gave me exactly what I wanted. I was going to have to get my hands dirty.

Diving Into Framework Code

Having read Shazwazza’s code, I had already learned that the file monitor is stored on the HttpRuntime singleton, so I started my investigation there. I was hoping it just stored a reference to an interface, so I could just replace it with my own wrapper over the base implementation.

Well, no interface – but I suppose I can just subclass FileChangesMonitor and use reflection to swap the base implementation out for my own.
Oh. It’s a sealed class – so I can’t even subclass it. It was about this point that I stopped trying to solve the problem on work time: this had become a personal mission, and I wasn’t going to give up until I’d made this thing behave the way I wanted to.

First, I needed to get a solid understanding of how the FileChangesMonitor class worked. After a lot of reading, it boils down to this: other classes ask the FileChangesMonitor to monitor a particular path, and it maintains a hash of paths to DirectoryMonitor instances which it creates. In turn, DirectoryMonitor has a hash of filenames to FileMonitor objects it has created, and FileMonitor has a HybridDictionary which stores callbacks. FileChangesMonitor, DirectoryMonitor, and FileMonitor are all sealed classes, so it looks like I’m out of luck – but I refused to be beaten!

The core classes I’m dealing with are all sealed, so there’s definitely no option to change behaviour there – but at each step of the object hierarchy, a parent object is storing instances of these sealed classes in plain old collection types: Hashtables and HybridDictionaries. Those classes aren’t sealed, so that’s going to have to be the seam where I stick my metaphorical crowbar.

A Word of Warning

If the idea of messing with private-member Hashtables makes you nervous, good. I’m about to go messing with the internals of framework classes in a very unsupported, highly-fragile way. This is not production-quality code. This is nasty, hacky code which is probably going to break in the future. If you ever use this code, or this extension technique, please do it in a way which will never, ever, ever run on a production machine.

I’m trying to fix the dev experience here, and so I decided to accept the hacky nature – but I have put plenty of guards around it to keep it clear of our production machines. I’ve even made sure it stays clear of our test environments. It’s a local-dev-environment only thing!

A Note on Reflection

If you’re familiar with the .Net reflection API, you’ll know it can be a little clunky. I’m using the following extension methods to make my code look a little cleaner.

public static class ReflectionHelpers
  public static FieldInfo Field(this Type type, string name) => type.GetField(name, BindingFlags.NonPublic | BindingFlags.Instance);
  public static FieldInfo Field<T>(string name) => typeof(T).GetField(name, BindingFlags.NonPublic | BindingFlags.Instance);
  public static FieldInfo StaticField<T>(string name) => typeof(T).GetField(name, BindingFlags.NonPublic | BindingFlags.Static);

  public static ConstructorInfo Constructor(this Type type, params Type[] parameters)
    => type.GetConstructor(BindingFlags.NonPublic | BindingFlags.Instance, null, parameters, null);

  public static MethodInfo StaticMethod(this Type type, string name, params Type[] parameters)
    => type.GetMethod(name, BindingFlags.Static | BindingFlags.NonPublic, null, parameters, null);

  public static Type WebType(string name) => typeof(HttpRuntime).Assembly.GetType(name);
  public static Object Make(this ConstructorInfo constructor, params Object[] parameters) => constructor.Invoke(parameters);
  public static Object Call(this MethodInfo method, params Object[] parameters) => method.Invoke(null, parameters);

How I Hooked (Hacked) FileChangesMonitor

What I really want to do is to prevent FileMonitor from setting up file watches for specific paths, and my only extension point is the HybridDictionary of watches. When FileMonitor is asked to set up a watch, it checks its dictionary to see if there’s already one there for that specific callback. If it doesn’t already have one, it sets one up. So what I need is a HybridDictionary which always has an entry for any given callback!

Sadly, HybridDictionary doesn’t declare its methods as virtual, so we can’t just go overriding them. Fortunately, it does just use a Hashtable under the covers – and we can use some reflection trickery in the constructor to replace the Hashtable it creates with one of our own design!

/// <summary>
/// Intercepting hybrid dictionary. It always responds to any key request with an actual value
///  - if it didn't exist, it creates it using a FileMonitor factory.
/// HybridDictionary always uses a hashtable if one has been constructed, so we just rely on InterceptingHashTable.
/// This relies on an implementation detail of HybridDictionary but we're way past worrying about that kind of thing at this stage.
/// </summary>
public class MonitorInterceptingHybridDictionary : HybridDictionary
  static Type EventHandlerType = WebType("System.Web.FileChangeEventHandler");
  static ConstructorInfo FileMonitorConstructor = WebType("System.Web.FileMonitorTarget").Constructor(EventHandlerType, typeof(string));
  public MonitorInterceptingHybridDictionary()
    var fakeTable = new InterceptingHashTable(o =>
          return FileMonitorConstructor.Make(null, "test");
        catch (Exception ex)
          Log.Error(ex, "Unable to create a file monitor target pointing at {Target}", o);
          return null;
    Field<HybridDictionary>("hashtable").SetValue(this, fakeTable);

Our new HybridDictionary is using another new class I’m going to re-use shortly – the InterceptingHashTable. Here, we finally catch a break: Hashtable declares a lot of its methods as virtual, so we can override functionality. What we’re going to do here is to create a new subtype of Hashtable which always has the key you’re looking for, even if it hasn’t been added. We pass a factory function in to the constructor, so whenever something asks for the value for a given key, we can construct it on the fly if it hasn’t already been added.

/// <summary>
/// Intercepting hash table. It always responds to any key request with an actual value
///  - if it didn't exist, it creates it with the provided factory.
/// This currently only implements this[key] because that's all the file monitor stuff uses.
/// To be generally useful it should override more methods.
/// </summary>
public class InterceptingHashTable : Hashtable
  private readonly Func<object, object> _factory;
  public InterceptingHashTable(Func<object, object> factory)
    _factory = factory;
  public override Object this[Object key]
      var maybe = base[key];
      if (maybe == null)
        maybe = CreateMonitor(key);
      return maybe;

  public Object CreateMonitor(Object key)
    return _factory(key);

If you take a look back at the MonitorInterceptingHybridDictionary code above, you’ll see that it’s going to pretend it’s already got a FileMonitorTarget anytime something looks – and that means the calling FileMonitor will think it’s already got a watch on the file, and skip creating it. Bingo! The next step is to replace the HybridDictionary on FileMonitor with an instance of our new class.

Of course, it’s a private member, and we’ve currently got no way to hook the creation of FileMonitor instances. What does create FileMonitor instances? The DirectoryMonitor class – and fortunately for us, it keeps a Hashtable of all of the FileMonitors it’s created. That means we can swap out the Hashtable for another of our InterceptingHashTables, and every time the DirectoryMonitor looks to see if it already has a matching FileMonitor, we create it if it doesn’t already exist.

// Create a dodgy hashtable which never has a blank entry - it always responds with a value, and uses the factory to build hacked FileMonitors.
var fakeFiles = new InterceptingHashTable(f =>
      // Build a FileMonitor.
      var path = Path.Combine(o as string, f as string);
      var ffdParams = new object[] {path, null};
      ffd.Invoke(null, ffdParams);
      var ffdData = ffdParams[1];
      object mon;
      if (ffdData != null)
        var fLong = ffdData.GetType().Field("_fileNameLong").GetValue(ffdData);
        var fShort = ffdData.GetType().Field("_fileNameShort").GetValue(ffdData);
        var fad = ffdData.GetType().Field("_fileAttributesData").GetValue(ffdData);
        var dacl = secMethod.Call(path);
        mon = fileMonConstructor.Make(dirMon, fLong, fShort, true, fad, dacl);
        // The file doesn't exist. This preserves the behaviour of the framework code.
        mon = fileMonConstructor.Make(dirMon, f as string, null, false, null, null);
      // Break the FileMonitor by replacing its HybridDictionary with one which pretends it always has a value for any key.
      // When the FileMonitor sees it already has this key, it doesn't bother setting up a new file watch - bingo!
      var fakeHybridDictionary = new MonitorInterceptingHybridDictionary();
      mon.GetType().Field("_targets").SetValue(mon, fakeHybridDictionary);
      // Take this warning away later
      Log.Debug("Provided fake monitoring target for {Filepath}", path);
      return mon;
    catch (Exception ex)
      Log.Error(ex, "Unable to create a file monitor for {Filepath}", f as string);
      return null;

Now, we’re faced with another problem: we have no control over the creation of DirectoryMonitor instances, but we need to replace the private Hashtable it keeps with the fake one we’ve just made.

Luckily for us, the FileChangesMonitor class keeps a Hashtable of DirectoryMonitors!

You might be able to guess where this is going: we need another fake Hashtable which always creates DirectoryMonitor objects on the fly. Stay with me here, we’re approaching the end of this thread we’ve been pulling on.

// This is a factory which produces hacked DirectoryMonitors. DirectoryMonitor is also sealed, so we can't just subclass that, either.
Func<object, object> factory = o => {
    if (!(o as string).Contains(@"\EDS\dist"))
      Log.Debug("Allowing file change monitoring for {Filepath}", o as string);
      return null;
    var dirMon = dirMonCon.Make(o as string, false, (uint) (0x1 | 0x2 | 0x40 | 0x8 | 0x10 | 0x100), fcnMode);
    // Create a dodgy hashtable which never has a blank entry - it always responds with a value,
    // and uses the factory to build hacked FileMonitors.
    // ... the previous code block goes here ...
    // Swap out the hashtable of file monitors with the one which lets us always create them ourselves.
    dirMon.GetType().Field("_fileMons").SetValue(dirMon, fakeFiles);
    return dirMon;
  catch (Exception ex)
    Log.Error(ex, "Unable to create a directory monitor for {Filepath}", o);
    return null;

Finally, the HttpRuntime class holds a reference to a single copy of FileChangesMonitor, and HttpRuntime itself is a singleton!


With a little more reflection, we can therefore swap out the Hashtable in FileChangesMonitor with our hacked Hashtable of DirectoryMonitors, and now we have control of all the instantiation of DirectoryMonitors and FileMonitors, and we can intercept any filenames we don’t want tracked and provide dummy FileMonitorTarget responses to prevent the watches from being set up.

* I am so, so sorry if you ever have to maintain this.
* ASP.Net monitors a pile of folders for changes, and restarts the app pool when anything changes.
* Gulp often triggers this - restarting the app pool unnecessarily, annoying developers. Annoyed developers write code like this.
var httpRuntime = StaticField<HttpRuntime>("_theRuntime").GetValue(null); // Dig out the singleton
// Grab the FileChangesMonitor. It's a sealed class, so we can't just replace it with our own subclass.
var fcm = Field<HttpRuntime>("_fcm").GetValue(httpRuntime);

// Grab a bunch of internal types, methods, and constructors which we're not meant to have access to.
var dirMonType = fcm.GetType().Field("_dirMonSubdirs").FieldType;
var dirMonCon = dirMonType.Constructor(typeof(string), typeof(bool), typeof(uint), typeof(int));
var fileMonType = dirMonType.Field("_anyFileMon").FieldType;
var fadType = fileMonType.Field("_fad").FieldType;
var fileMonConstructor = fileMonType.Constructor(dirMonType, typeof(string), typeof(string), typeof(bool), fadType, typeof(byte[]));
var ffdType = WebType("System.Web.Util.FindFileData");
var ffd = ffdType.StaticMethod("FindFile", typeof(string), ffdType.MakeByRefType());
var secMethod = WebType("System.Web.FileSecurity").StaticMethod("GetDacl", typeof(string));

// ... insert all the above code to create the factory ...

// Swap out the hashtable of directory monitors with the one which lets us always create them ourselves.
var table = new InterceptingHashTable(factory);
fcm.GetType().Field("_dirs").SetValue(fcm, table);

So what’s the up-shot of all of this? Well, firstly, I’ve written some of the nastiest, hackiest code of my career. Secondly, I’ve wrapped it all up into a nuget package so you can easily run this nasty, hacky code (it’s not up quite yet – I’ll share another post when it’s ready to go). And thirdly, I need to warn you again: Do not put this near production! I don’t care how you accomplish it, but please put plenty of gates around the entry point to make sure you don’t trigger this thing in a deployed environment. It really is a dev-only device.

Surely There’s a Better Way?

There is absolutely a better way. I encourage you to find other ways to avoid dev changes constantly churning your app pool. I strongly suspect our problem was serving our Gulp build output via a secondary bundling step: the Asp.Net bundler seems to be one of the things which might register files with the FileChangesMonitor. For various reasons (mostly, that the project was a big, complex, legacy project which is hard to test thoroughly), I strongly preferred dev-only changes to making production-affecting changes. And so here we are.

If you can find a way to avoid using this technique, I strongly recommend you do.

What I talk about when I talk about DevOps

What does ‘DevOps’ mean, anyway?

devops-photo-mWe’re a DevOps consultancy, and we’re hiring DevOps engineers. But the software industry has a crisis right now – and that crisis is all about the word ‘DevOps’ and what it really means. I’m starting to wonder whether putting ‘DevOps’ in our job ads might actually be counter-productive – and at YOW! this week, I finally understood why.

Elabor8 (where I work) had a booth at YOW!, and I gave a couple of lightning talks there. Probably the biggest crowd-pleaser was my talk on resiliency patterns in distributed systems. I covered some difficult topics like the importance of having idempotent rollback steps in compensating transactions and how lessons learned from the ship-building industry help us craft better distributed systems, all presented in 10 minutes in a crowded event space.

Hang on, you may well ask. If I’m a DevOps consultant, why am I talking about atomicity and consistency in distributed systems? Shouldn’t I be talking about cool PowerShell tips and how to set up Jenkins?

As is so common with rhetorical questions, the answer is a resolute ‘No’.

When I talk about DevOps, I talk about Software Engineering.

When I do DevOps work, I’m doing software engineering. When I hire for DevOps roles, I hire software engineers. But I don’t hire just any software engineers: I want the ones who care about the delivery process. They know how to build software, and also how to put it in front of the user – and how to keep on putting that software in front of users, sprint after sprint, story after story, rapidly, efficiently, and without breaking things.

‘DevOps Engineer’ is the next ‘Full-Stack Developer’ – and not because it’s the next hype-cycle in tech hiring. In the same way full-stack developers expanded their scope to cover both the UI and the back-end, DevOps engineers have expanded their scope beyond writing software – and into the realm of how we get that software in front of users, in the fastest and most reliable way possible.

No longer are we content to just build back-end systems and UI layers on top of them: as a profession, we’re coming to understand that software engineering is bigger than just churning out vertically-sliced user stories. Software engineering is about building the right thing and keeping it running – and a DevOps engineer is a software engineer who cares about both steps. You don’t want a team full of DevOps engineers – but you definitely need at least one.

When I talk about DevOps, I talk about Agile.

You start to see the real benefits of automation when you’re deploying to production regularly. For example, if you’re only shipping a few times a year, the overhead doesn’t hurt you that much. If you have a 6-week QA pipeline and a 2-month UAT window, you don’t need DevOps (yet). Once you start trying to deploy regularly – getting your cycle time down, keeping your WIP low, delivering value to the user faster and more frequently – that’s when the overhead starts to hurt.

Once you introduce agility to your process, that’s when you need to pay attention to the DevOps movement. That’s when you need some software engineers who care about automation. Please note though that I’m not saying you need “some DevOps” – there’s no such thing as “some DevOps”, and anyone who tries to sell you “some DevOps” is doing you no more favours than someone who tries to sell you “some Agile”. What you need is some smart software engineers who care about DevOps – and you need to give them the time and resources they need to do their job.

When I talk about DevOps, I talk about teams.

DevOps engineers are software engineers, but that doesn’t mean you should fill your software teams up with DevOps engineers. DevOps engineers tend to be passionate about a bunch of really interesting stuff: resilience patterns, testing, automation, source control, and release management. The great thing about multi-disciplinary, cross-functional teams, however, is that you get a bunch of people with different passions together, and that breadth gives you the ability to do great things. Don’t try to hire DevOps engineers (or worse, to build a DevOps team). Hire software engineers, and when you find ones who are great at DevOps, keep them.

Having cross-functional teams also gives your team members the chance to cross-skill while they work with other team members, which is why…

When I talk about DevOps, I talk about teaching.

Great DevOps engineers have a genuine enthusiasm for quality software engineering and release management, and their enthusiasm is infectious. Great DevOps engineers don’t hoard their knowledge, but help their fellow software engineers to learn more about the DevOps mindset by sharing what they know. They also learn from their colleagues who specialise in other software engineering fields, becoming more well-rounded themselves as they help others do the same.

Finally: When I talk about DevOps, I talk about people.

DevOps is a branch of software engineering – and whatever you might hear, software engineering is all about people. It’s about the people who use our software, and the people who build it. DevOps is the intersection of those two groups: users and developers. Our users are not just end-users, who enjoy higher-quality software, but also our fellow engineers, who rely on our automation to work more efficiently. They’re our managers, who rely on the insights we give them into the release process. Our users are the junior software engineers who may one day specialise in DevOps engineering – or who might use what they can learn from us to be better at some other branch of software engineering.

And as always, we have a bunch of work on, and we’re hiring.

So, I have two things to ask. Firstly, if you have feedback on our job ad, please reach out! I’m always trying to improve these things.

Secondly, if you’re a software engineer who cares about the difference between GitFlow and GithubFlow; who has made calls to the Octopus API; or who loves showing other developers how to write a better test or add more context to a log message; please talk to me about joining Elabor8. We’re a bunch of software engineers who care about DevOps – and our users aren’t just one team, but a range of teams across some really great clients. If you want to take the engineering you care about and have a broader impact on more people, you want to join us. Get in touch.

Anatomy of a Job Ad

Wanted: Senior DBA!

  • Minimum 10 years experience managing SQL Server 2015 or newer

It’s a classic joke – the job ad which wants you to have experience in a specific technology for longer than it’s existed. Sadly, it’s funny because there’s a grain of truth in it: job ads are terrible.

Wanted: Junior Developer

  • 1-2 years experience writing software
  • Strong knowledge of SQL, git, Python, Haskell, and Perl
  • Good UI skills, excellent API design skills, knowledge of ESBs and other distributed software patterns
  • Excellent grounding in the principles of good software design and Agile project delivery
  • Experience with [you usually find a list of specific platforms and libraries here]
  • Go-getter, always-be-learning attitude
  • Highly regarded: strong maths and stats skills, C++ experience
  • Highly regarded: exposure to data science, experience with Big Data platforms such as Hadoop
  • Excellent communication skills are a must!

I hate seeing job ads like this one. You want someone with 1-2 years’ experience to know all of that – and be confident enough to tell a recruiter that, and back it up in an interview? You’re not selecting for capability here. You’re selecting for over-confidence, and the coincidence of having a first job which involved just the right mix of technologies.

Technology Leadership Position!

Are you a dynamic leader with top-notch management skills and brilliant technical ability? Do you have world-class knowledge of big data platforms and machine learning? Are you just as comfortable writing a PhD dissertation as you are selling a group of executives on a new company strategy? Do you have a passion for creating a dynamic company culture and mentoring a team of brilliant engineers, all while maintaining an unwavering focus on creating an unmatched user experience? Do we have a job for you!

If you’re posting a job ad like this one, you’d better have a remuneration package to match; but you probably don’t. The technology industry is full of people with imposter syndrome – and if the perfect applicant (who would have been great for your team) didn’t have it before they read this job ad, they will afterward. If the candidate who does meet that brief exists, they’re definitely not reading job ads on Seek, anyway.

These are, of course, caricatures, but they’re not so different from real job ads I’ve seen. How about a real example from my own life?

Checkbox Syndrome

Many years ago, I landed a job in the United States (I’m based in Australia, so this would have been a big move for me). The recruiter explained that the job had been open for nearly two years, and they were thrilled to have found me! I was their first suitable candidate, and they were keen to rush me through the hiring process.

This puzzled me, because I’m not that special. How was I the first suitable candidate in two years?

Well, it turns out that they’d written a very specific candidate description, and someone in HR had been handed the job of finding someone that matched. They were after someone who listed optimising high-throughput transactional systems on their LinkedIn profile, and who had completed at least two years of tertiary study in Latin or Ancient Greek.

They were building some language-processing software, and someone had decided that an engineer with a background in linguistics would be good to have on the team. That was somehow translated to requiring some tertiary education in a classical language – and bam! Two years of turning down candidates who probably would have been highly successful in the role. All they actually wanted me to do was to come up with a way to load-test the system, and then find hot-spots in the code and optimise them.

I never did end up moving to Houston: my visa application fell through (thanks, Global Financial Crisis). It was probably for the best: I wouldn’t have been a person to this company, or even an engineer; just a series of checkboxes.

What is a job ad for, anyway?

When you’re writing a job ad, please don’t forget what the ad is for. Your goal is to attract suitable applicants to apply, and discourage unsuitable applicants. Anything which doesn’t accomplish one of those two things is wasteful. Anything which discourages suitable applicants is a net loss.

As a specific example of this, please leave “Good Communication Skills” out of your ad. You might think that poor communication skills will disqualify a candidate – but putting that in the ad isn’t going to stop unsuitable candidates from applying, and it’s not going to encourage suitable candidates to apply. It’s noise. Leave it out.

Sell the job – genuinely

It all starts with empathy. Try to imagine yourself as your target engineer, tester, analyst, or whoever you’re trying to hire. Your goal is to sell them on the job, but also keep enough important criteria in there to discourage unsuitable applicants. The selling part is important: if there aren’t many great candidates out there, you need a job ad which will attract them. Candidates are about to invest a significant amount of their own, personal, outside-work time in applying for and interviewing at your company, and then you’re going to ask them to resign from their current position, where they have friends and valued colleagues and know the system, to join your team. Give them good reasons.

Please don’t give them a sales pitch. Don’t spend most of the ad spruiking the company. Talking about how great you are is for your marketing material, or your annual shareholders’ report.

Just talk honestly about your culture, values, and benefits.

Anatomy of a Good job ad

Here it is. This is what your job ad should look like.

  • Talk about the role.
    Tell prospective candidates a little bit about the company and the role. This should be short. What does your company do? Who are you looking for? How will this position further the company’s mission?
  • Say who you’re looking for.
    Really cut it down. Don’t have 10 bullet points. Keep it to 3 or 4. Try to focus on higher-order skillsets, rather than specific technologies, unless that technology really is core to the role.
    If “Good Communication Skills” is making it to your short-list of 3 or 4 key skillsets, I hope you’re hiring a radio operator or air traffic controller. Otherwise, please, just leave it off.
  • Tell them what’s in it for them.
    Describe your benefits. Tell them about the great company culture. Talk about training and conference budgets and career opportunities. Once again, keep it short and to the point.
    If you can’t think of anything to put here, you might have bigger problems.
  • Tell them how to apply.

That’s it. Really. No nice-to-haves – those just give suitable candidates a reason not to apply. Don’t do that. If you get two suitable candidates, and one of them happens to have one of your nice-to-haves, you might still offer the role to the other based on the bigger picture. If it’s not necessarily a differentiator even at the end of the recruitment process, it definitely has no place at the beginning.

But I want to list a bunch of stuff!

Resist the temptation. Are you hiring a team lead? Just say you’re hiring a team lead. Don’t list out all the stuff that team leads need to do.

  • Lead a team of software experts to deliver innovative products.
  • Help to work with product owners to deliver real business value!
  • Mentor senior engineers, and help them to mentor juniors.
  • Engage with the business about technical challenges.
  • Foster a strong team culture.

Don’t do this! Team leads already know the day-to-day detail of the role, and they won’t be going back to the job ad to work out how to spend their time. Just say that you’re looking for an experienced team lead or a senior engineer looking to step up to a team lead role – and get on to covering the more important points! What kind of team is it? What is the team mission? Are you looking for someone to drive a major change, to keep an already-high-performing team pointed in the right direction, or to build a whole new team? That stuff is much more useful than saying things like “Ensuring the team is aligned with business priorities”.

Where do I put things like “Strong work ethic” and “Ability to work in a collaborative team environment”?

In the same place you put “Good Communication Skills”. People who don’t have a strong work ethic or the ability to work in teams are going to apply anyway, so all you’re doing is making the job ad longer and more boring.

This sounds like you’re a really cool person and I’d like to work with you.

If you have great software experience and a DevOps mindset, get in touch with me! Elabor8 really is a cool place to work, and we have an awesome mission. We also have a bunch of other roles we’re hiring for: check out our careers page.

Wait, was this secretly a job ad?

No. This doesn’t match my “Anatomy of a Good job ad” at all! But recruitment is bigger than just job ads, and this was, secretly, a bit of guerrilla recruitment. That’s another idea I’m hoping you’ll take away from this post: recruitment is about a lot more than just writing a good job ad. It’s about being a place people will want to work, and making sure the right people know it.

I’m Joining Elabor8!

I’ve thought about joining a consultancy before, but I’ve always  resisted. Software consultancies are often all about delivery: get in and ship some software! And that’s something I love doing, but it’s not the thing I love doing most.

I love enabling teams.

Enabling teams takes many shapes, and that variety is part of why I love to take on that role. I’ve joined more than one team which wasn’t even using version control, and so my role was to take the team on a basic CI journey: build automation, branch strategy, and automated testing. Of course, part of that process is training: coach, coach, coach. Teach them how to do CI, and of course, teach them how to learn. Don’t give a man a fish, teach him how to fish: there’s no point setting up a build server if there’s nobody to keep it going after you move on!

There’s nothing wrong with needing help with the basics: everyone has to start somewhere. However, I’ve been privileged to work with some amazing teams over the years. Teams which were well past needing help with the basics. Even at those, though, I’ve never been content to just write code. Whether it’s been working with executive on group strategy, or finding better ways to attract and retain technical talent, I’ve always been keen to find ways to leverage my time more effectively. And as always: coach, coach, coach! Teach engineers how to recruit. Teach sales groups how to understand technology. Teach executives how to engage with their technical managers.

I want to stay on the tools, but keep my eyes up on the broader organisational goals. Never be content with the status quo: always be looking for ways to deliver more business value, faster, with higher quality. Don’t worry about how to deliver more value as an individual: find ways to leverage every single team member better. There’s only so much an individual can achieve. If you focus on improving your team, you can deliver so much more – and continue delivering value even after you leave.

The big secret, of course, is that this is exactly what technology consulting is all about. It’s never just about delivery. In all my work with software consultants though – mostly as a client – the focus has been on the delivery. “We have a project, and we need help. Send us people.” There are consultancies which have a great reputation for making sure capability uplift and coaching happens – alongside projects. I think I would have felt right at home in any of these organisations.

Elabor8 stood out for me because coaching and uplift is right at the core of their offering. We don’t offer coaching as an unwritten benefit of bringing people in on a project. Instead, we see project work as a mechanism for delivering on our core mission: capability uplift and coaching. Elabor8 was the first consultancy which came to me and offered me the chance to do what I love – enabling teams – as a core focus.

We call our primary mission Engineering Effectiveness. It’s about building the right software, the right way – and we’re not doing it for you, but instead helping your teams to do it. I get to become the strategic partner, the coach, the trainer – not just for one team, or one organisation, but for any company which needs our help.

That is why I’ve joined Elabor8. If you’re on a team which needs help – at any level – get in touch! I’d love to hear from you. And if you’ve read all of that and it resonates with your own passions, perhaps you should think about joining us.

(If there are no positions there which suit you, but you really think you’d be a great fit anyway, reach out to me and let’s see what we can make happen.)

Measure What Matters

What’s your average API response time? Do you know? Is it important to your business? What about the 90th percentile? Do response times suffer during peak demand?

Do you think about those questions? How about these ones:

How long does it take to get a software change reviewed? Do you know? Is it important to your business? Is it a bottleneck? Do reviews get skipped during busy periods?

If you care about code reviews, you should measure them. Put them on your system dashboard. They’re as much an indicator of the health of your software environment as your API response times. Minimising Work In Progress and Mean-Time-To-Release are important parts of your QA process, and making sure your pull requests are reviewed and merged in a timely fashion is a great way to improve those numbers.

What existing products are there out there to do this? Depending on the tools you use, you can probably pull out a few relevant reports. Jira is popular, and I’ve seen PMs produce some great graphs to include in their monthly management update. The problem is, the numbers you get out of these tools don’t give you direct, real-time feedback. Their very nature as longer-term averages mean they can’t represent a call to action.

Enter TeamLab

As a software shop, if the tools I’m using don’t do what I want, I have an option: build something. This is a dangerous option to have, and countless business hours have been wasted solving the wrong problems, but I really needed a nice visual prompt of how we’re doing at our code reviews in-the-moment. I also wanted a side-project for the team to tinker with new ideas for writing web applications – so even if the project didn’t turn out to be useful, the experiment would teach us something.

I had a specific technology I wanted to try out: React Storybook. This is a really nice way to visualise your React components in various different states, and I wanted something relevant to use as a demo for the team. It was very quick and easy to get up and running with a create-react-app project including Storybook, and I hacked together a quick picture of what my PR display should look like:
On the right, you can see my quick mock-up of a board displaying ten pull requests, and the left is the Storybook control panel.

I decided it would be useful to colour-code the pull requests, and display any reviewers and approvers on the PR cards. A new PR is yellow, and an approved one is green. A PR with reviewers turns blue, and most importantly, any PR which is older than 48 hours turns red.

This was a nice little mock-up, but there was no real data behind it at this stage. Fortunately, the Git server we use has a fairly straightforward API, and so it didn’t take long to get some real data behind this component.

It’s really easy to see when we have PRs which are starting to get stale, and need attention. Quick – at a glance, how many PRs here have been hanging around too long and need attention?

This has become the go-to way of seeing our outstanding PRs at a glance, and has since gone up on a big screen on the wall in our dev team office. I soon got requests for a few other widgets to go on the same dashboard, and our little side project has become a key part of our DevOps toolkit.

Has It Worked?

Having those cards up where we can see them during the day has been good – but the biggest signal is during stand-up each morning. A quick glance at the TeamLab PR board has become part of the ritual, and if those cards start to build up – especially if they start to turn red – the team has a really strong signal that we’re getting behind on our code reviews.

I don’t currently have a report which tells me the Mean-Time-To-Merge for our PRs – but I don’t think I need it. Mean-Time-To-Merge isn’t as strong or immediate a signal as a pile of glaring red PR cards looming over our morning stand-up, nor does it provide the immediate sense of relief when we clear the board.

What Next?

I’m not sure what will go on the dashboard next, but I have some idea what kinds of things I’m looking for.

I need things I can measure – things I can pull straight out of an API. Things which can directly influence numbers like Mean-Time-To-Release – but I don’t want to display averages like that. I’m going to give people a dial they can turn directly. I’ll pick an angry colour like red for things which are outside targets, and nice friendly colours like blue and green for things which are on track. Once something is off the list, I’ll make it go away.

In short, I want to find things which I can measure, which team members can directly influence, and which will improve our overall quality – and I want to put them up where everyone can see them.

Announcing NSchemer 1

If you already know all about NSchemer, you can jump straight to the Version 1 release notes.

What is NSchemer?

Database schema management has been an interest of mine for a very long time. I’ve seen all sorts of approaches tried: folders full of .sql files, schema version tracking in Excel, and of course the tried-and-true1 manual approach using schema diffing tools. I’m a keen proponent of automated schema management. Automated deployment is all the rage these days, and if you can’t automate your schema updates, you can’t automate your releases.

I prefer to go one step further: I like to aim towards single code path schema management. Any database, whether a brand-new one to support a new installation, or an ancient database restored from a backup for a returning client, should get to the current version using the same code path – or at least, as close as possible.

When I started pushing this idea – that developers should write their own SQL migrations as they went, rather than leaving it to the designated DBA to do in the lead-up to a release – I got some push-back. Some of my team didn’t want to write SQL. Thus, NSchemer was born2. The framework languished in alpha status for many years, despite being actively used in a number of production systems. Recently, I finally decided to tidy up the API, add a few new features I’d been meaning to for a while, and bump it up to version 1.

Why Automated Schema Management?

The number one reason for automating your schema management is testing and reliability. Assuming, for a moment, that you have test environments, automated schema management means your test environments should go through the same migrations as your production environments will – automatically, with no opportunity for a manual step to get skipped or done incorrectly. This gives you a lot of confidence that when you hit the Big Red Button to go live with a new version, your schema migrations will work: the same automated set of steps which have run against all of your other environments will run against production.

You get a lot of other nice bonuses, as well. When you merge master into your own branch, not only do you get all of the new code; you also get the migrations that update your database schema to match. No more pulling in another branch, only to have to manually update your local database schema to match.

Digging up a backup from a couple of years ago? No worries, NSchemer will bring it up to current without any hassle at all.


While you can install NSchemer into an existing assembly, I typically create a new assembly just for managing schema transitions (I use a console app, so I can run the transitions from a script during deployment). Once you’ve created YourProjectName.Schema, just

install-package NSchemer

and you’re ready to go.

Show me the code!

NSchemer uses a single class which inherits from SqlClientDatabase to represent a versioned schema. Just create one, implement the Versions collection, and start writing transitions (beginning from 1 – NSchemer uses version 0 internally). If you’re starting with an existing schema, just use your favourite SQL tool to generate a full CREATE script, and drop it in as version 1 (use the embedded resource transition mentioned below).

public class TestSchema : SqlClientDatabase
    public TestSchema(string connectionString) : base(connectionString) {}
    public override List<ITransition> Versions
            return new List<ITransition>
                new CodeTransition(1, "Initial Schema", BuildTheWorld),
                new CodeTransition(2, "Add Widget Table", "This script adds a very important table", AddWidgets)
    private bool BuildTheWorld()
            new Column("ThingId", DataType.BIGINT).AsIdentity(1, 1).AsPrimaryKey(),
            new Column("ThingName", DataType.STRING, 50)                
            new Column("AnnotationId", DataType.BIGINT).AsIdentity(1, 1).AsPrimaryKey(),
            new Column("Text", DataType.STRING, 50),
            new Column("ThingId", DataType.BIGINT, false).AsForeignKey("Thing", "ThingId")
        return true;
    private bool AddWidgets()
        RunSql(@"CREATE TABLE DBO.Widget (WidgetId [int],WidgetName [nvarchar](50)) ON [PRIMARY]");
        return true;

The core of your schema class is the Versions collection, which contains a numbered list of transitions to be run in order. NSchemer will automatically create a table to track which versions have and haven’t been run, and whenever you call Update() on your class, it will work out which transitions haven’t run yet, and apply them. Assuming your transitions all ran without exceptions and returned true, your schema should now be up-to-date, and the version history table updated. If any of your transitions either returned false, or threw an exception, Update() will throw an exception.

Why not just SQL?

You’ll notice, in the sample above, that as well as being able to write SQL transitions, there are helper methods like CreateTable. These are here for four reasons:

  1. Some developers refuse to write SQL.
  2. Some developers write terrible SQL.
  3. They’re actually pretty convenient.
  4. If/when NSchemer officially supports non-MSSQL databases, your transitions should be cross-platform.

If you don’t want to use these convenience methods at all and you’re happy just using SQL, you could also look at some of the SQL-only frameworks which do the same thing as NSchemer, such as DbUp (which has explicit support for a range of other databases as well as Microsoft SQL Server).

If you have larger blocks of SQL to run, don’t put them all into a string like in the example above: NSchemer also supports resource files. You can create a transition like this:

new SqlScriptTransition(3, "Add another table", "NSchemer.SystemTests.EmbeddedFile.sql")

It will look in the same assembly for an embedded resource file with that name. There is also an overload which allows you to specify a different assembly for the resource file.

NSchemer uses a similar format to SQL Server Management Studio: it uses GO on a line by itself as a command separator, allowing you to submit multiple chunks of the file as separate commands.


NSchemer has a couple of options you can use to control its behaviour. I’m afraid the API is inconsistent and the options are limited: I plan to address this when I do the overhaul in 2.0 (see below).

  • VersionTable
    Override this property to change the table NSchemer uses. Default: NSCHEMER_VERSION
  • SchemaName
    Set this property to use a schema other than dbo (use with caution: this has limited test coverage).

Can I rely on NSchemer?

You should be testing all of your migrations in testing and staging environments before they make it to production. These tests will also ensure NSchemer is behaving itself in your environment. Should you start using NSchemer in production without ensuring it goes through a testing pipeline? No, but you shouldn’t be running your own code that way either.

If you find any bugs or problems in NSchemer, please report them on the NSchemer GitHub repository. I use NSchemer myself, and I’m keen to fix any bugs you find as soon as possible. I’m also open to pull requests.

Implementation Advice

I like to run NSchemer from two different places: one in development environments, and another in deployed environments (testing, staging, UAT, production, whatever you prefer to call them).

To run in development environments, I throw some guards in (to make really sure it never runs in a deployed environment), and put it somewhere it will run on application start-up. It might look like this:

if (Debugger.IsAttached && _connectionString.Contains(".\sqlexpress")) {
    new MySchemaClass(_connectionString).Update();

For deployed environments, I make sure the assembly containing my transitions is a console app, run the transitions from there, and return or output a success/failure message so my deployment tool knows whether to continue or alert me that something went wrong.

Upgrading from 0.x

The only changes you should need are to sprinkle a few using NSchemer.Sql statements at the top of your files.

You should notice some API improvements:

  • When you create columns, there are new options you can provide using a fluent syntax:
    • .AsPrimaryKey()
      Only allowed when creating a new table. Specifies that this column is part of the primary key. Supports composite keys.
    • .AsForeignKey(…)
      Allowed when either creating a table or adding columns to an existing table. Allows you to indicate the referenced table, column, and (optionally) cascade options.
    • .Identity(…)
      Allows you to create the column as an IDENTITY (auto-increment) column. You can specify the initial seed, and increment value.
  • Description is no longer required on transitions.
  • CreateTable(…) now has a params syntax, so you don’t need to create a List every time you use it.
  • You can now specify the nullable status of a column for data types which don’t require a length.

The Future: NSchemer 2.0

As I’ve used NSchemer, I’ve discovered a few short-comings of the existing API. I made some minor breaking changes when I bumped NSchemer to 1.0, mostly just relating to namespaces, but 2.0 is likely to be a more significant overhaul of the API.

I also want to split Microsoft SQL Server support out into a separate package, and provide official support for other database servers. If you have strong knowledge of MySQL, PostgreSQL, Oracle, or another database platform and would like to help maintain support for that platform, please get in touch.


I’ve chosen to release NSchemer under the LGPL because I want it to be generally usable, but I want to make sure any improvements are available to everyone who uses it. If you want to use NSchemer but the LGPL is a problem for you, please let me know. I’m sure we can work something out.

Healthcare Software: Past, Present, Future

A guest post by myself from a different blog.


I did this for a slide deck on the past, present, and future of software at Ramsay Health Care. Disclaimer: opinions are my own, and future may or may not represent actual future. Also I think our wards are usually nicer than this!

Helping Your Best Developers Leave

It’s my job to help my team members find better jobs. I know that sounds a little counter-intuitive, but stick with me for this one – I hope I can convince you.

There are two types of software developers: wanderers and lifers. Wanderers drift from job to job. They may stick around at a job for a year, or five, or even ten, but the one constant throughout their career is that they don’t intend to stay in their current job forever. The reasons change from person to person, and from job to job: sometimes they just hate their job; sometimes they’re in over their head; some people just love variety. Whatever the reason, they know that their next job won’t be the one they stay in any more than the last one was.

Then, there are the lifers. They may not stay in any one job for good either – people get made redundant, companies go broke, and changing circumstances force lifers to change jobs, but their goal is to find a team they like (or can tolerate) and stay in it as long as they can. The motivations for lifers vary, too, and it’s not just being scared of change: they may enjoy developing a deep knowledge of the business domain, the company culture, or the niche industry they’re in, and prefer leveraging that knowledge to starting fresh somewhere new.

Lifers can become a problem, though: if you’re a good environment for lifers, you will tend to collect them. Even if you’re a bad environment for lifers, you will tend to collect them. Every time you replace someone, they probably left because they were a wanderer – and some of the time, you’ll end up replacing them with a lifer. The reverse almost never happens: your lifers aren’t leaving (if they are, you might have bigger problems), so you can’t replace them with wanderers.

Before we get too far into this, I want to clarify something: I don’t buy into the stereotype that lifers are bad developers, and wanderers are good ones. There’s a bit of an attitude that anyone who stays in one place for too long “got stuck there”, while people who move from job to job are “in demand”. I’ve had plenty of recruiters talk to me in these terms. Too-frequent job hopping is bad (“they can’t hold down a job”), but staying in one place for too long is considered bad too. To speak frankly: this stereotype is rubbish. There are plenty of smart, productive developers who find themselves great jobs – which let them do interesting, intellectually-challenging computer science or engineering – and stay in them. Conversely, there are plenty of developers who drift from job to job, never really contributing much, but neither being quite bad enough to be worth firing.

But that’s an aside: I’m not writing this to tell you how to hire and retain good developers (I’ve written plenty of other articles about that). This time, I’m telling you how to get rid of your good developers! But first, more about why.

Remember how I told you that teams naturally accumulate lifers? Well, if you’re too complacent, you’ll also accumulate bad developers who have discovered your team is a safe place to hide. Wanderers who are just not much good (and don’t want to improve) will latch onto a team which tolerates them, and will milk it for all it’s worth.

There’s another big problem: market rates for developers have consistently risen faster than inflation, while salary increases almost never keep up. That means that the intermediate developer you hire today is probably getting paid more than the junior you hired five years ago – even though the ex-junior may well be worth more by now. Worse: if the junior hangs around to become a senior, they may well be one of your most valuable team members – and one of your lowest-paid ones. Why is this a problem? Well, aside from it being really unfair, it also sets them up to be poached. Even the staunchest lifer will eventually be tempted away by market rates, and if they’re even a little good at math, when they do the calculus, they’ll resent you for all the missed income over the years.

It gets worse: the longer they hang around before being poached, the more reliant on them you become. You may well be incurring some really significant business risks by hanging onto your developers for too long.

You could solve all of these problems by aggressively monitoring productivity and performance, firing the under-performers, and rewarding the achievers – and if you try this approach, you won’t be alone: software giants (and many other major employers) have dubbed this the “up or out” system (it was originally termed the “Cravath System”), and it kind of works OK – for them. Unless you’re Cisco or Google, I bet you can’t make it work at all: measuring developer effectiveness is famously difficult and error-prone. You might just find you end up promoting the networkers, self-promoters, and empire-builders, and firing your good engineers. In fact, even the software giants probably do this more than they’d like to admit.

What is a CTO, team lead, or development manager to do? Easy. Help prepare your developers to get better jobs elsewhere. One of my favourite software quotes came in response to a question about funding developer training: “What if we pay for all this training, and they leave?” The response: “What if we don’t, and they stay?” (I’m not certain of the origin of the quote, but Martin Risgaard tweeted something similar back in 2012). I think every good software team needs to double down on this idea. Don’t just pay for PluralSight accounts. Don’t just send your developers to a token conference every year. Really invest in turning your team members into people who are just too good to stay. If someone hangs around for six or eight years, you’re failing – or perhaps they will just never become the sort of developer teams want to hire. Perhaps you will eventually need to force them out – but now you will know it’s not just because of some silly “up-or-out” rule, but because you’ve done everything you can to help them thrive, and it hasn’t worked. This is good for you, and it’s good for them: you’re not dooming them to a career as one of the “got stuck there” brigade, and you’re also not letting them hang around in a job which just isn’t succeeding at building their career. Let’s face it: it’s entirely possible that their lack of success is as much your fault as it is theirs.

So, is there room in this philosophy for the genuine, high-achieving lifer? That rare individual who develops a deep understanding of your industry, continually improves themselves, boosts their team performance, and has a track record of innovation, year in and year out, for five, ten, or more years? Yes, absolutely. The best-laid plans rarely survive first contact, and you will absolutely run into people throughout your career who buck this trend, and should definitely be allowed to hang around for decades.

If this sounds like I’m back-tracking on everything I’ve said, you’re right – sometimes, there’s just no alternative but to have experienced and knowledgeable team leads, managers, and company officers who know when to exercise discretion and ignore all the rules. My central message here is not that you should fire anyone who makes it to their 10-year anniversary: rather, you should focus on doing your very best to turn your developers into the sorts of professionals who are in-demand and will definitely be hired away. On the way through, you’ll build a more effective team. You’ll create an environment which will encourage past employees to return – and they’ll be developers you’ll want back. Your team reputation will spread, as past employees go out into the general software industry and talk about everything they learned, and everything they accomplished. It will cost you more per developer, but you will reap the rewards many times over – and please, never underestimate the enormous benefit of having a team which can easily attract high-quality developers when you need them.

Most software teams have a long way to go – so you need some first steps. Here they are:
1. Invest in your employees. Don’t just allocate budget to buy them a hotel and a conference ticket every year – find real ways to help them learn and grow.
2. Support your employees in finding the next step in their career once you’ve finished learning from each other.
3. Expect great things of genuinely outstanding long-term employees – and find ways to reward them commensurately.

Above all, don’t make the mistake so many employers do – the mistake of encouraging employees to stay too long.

One final note: if you haven’t been embracing these ideas, don’t try to implement things too quickly. If you’ve spent the past ten years not investing in your employees, trying to move senior talent out could well be disastrous: it takes time to build the right sort of turnover, and to decide to hang onto the rare lifer who you really want to keep around. If you’re not sure, it’s safer to err on the side of spending more time investing in your existing employees, and giving them more time to find their next career move (or proving that they’re genuinely worth keeping around, at above-market rates).

On FizzBuzz and interviewing developers

Lots of people have heard of the FizzBuzz interview test (if you haven’t, Google it!), and Jeff Atwood once famously asked: “Why can’t programmers.. program?” But is it a useful test?

I’ve interviewed lots of developers, and hired quite a few of them. I’ve only regretted a handful of hires, and I’ve spent a lot of time trying to work out how to improve. I’ve made a career out of building teams, and hiring good people is a key part of that. Posing a simple programming challenge – often referred to as a FizzBuzz problem – is a common strategy, and it’s one that interviewers and job-seekers should both understand.

FizzBuzz is Pass/Fail.

One of the mistakes I think people make is in judging the code people write. I always put candidates through a FizzBuzz-type test, but I don’t really care about how good their implementation is. I have one very specific thing I want to know: can they write code, or can they not?

The pass/fail nature of FizzBuzz isn’t the sort of pass/fail you write a unit test for. I have no interest in whether the string they output has correct spacing, or whether they even get the calculation correct. I want an answer to this question:

Has this candidate spent even a little bit of their recent career writing code?

If I’m asking someone to solve FizzBuzz, I’m not hiring a program manager, a technical writer, or a business analyst. I’m hiring someone to write code. I’m hoping to hire someone who can write good code, which solves the correct problem, and produces a good user experience, and doesn’t introduce performance problems, but the core skill I’m looking for is the ability to write code. If they can’t write code at all, the quality or correctness of the code they write isn’t a concern.

FizzBuzz is trivial.

I’ve heard people lump FizzBuzz in with algorithmic problems, like asking a candidate to solve the traveling salesman problem. I’ll admit: if I was asking someone to solve FizzBuzz and send me their answer, it’s an algorithm problem. A very simple one, which I’d expect a high-school student doing a programming course to cope with, but an algorithm problem nonetheless. I don’t ask people to submit a solution, though: I ask them to do it in front of me, and what I’m really interested in is the first step.

Loops are one of the simplest programming concepts.

Fundamentally, programming is about loops and conditions. There are higher-level concepts that are really important, but you really don’t get any simpler than loops and conditions. FizzBuzz has a really simple beginning: “go through the numbers between one and twenty, and …”

The rest doesn’t really matter. I’ll pass someone who isn’t sure about how to work out if a number is a multiple of 3, or 4, or both. I want to know if the candidate can take a really simple problem statement, with an extremely obvious first step, and make a start writing a really simple solution.

People Fail FizzBuzz.

Do 199 candidates out of 200 fail FizzBuzz? No way. If that many fail, you are interviewing people you shouldn’t. Most people I interview have no trouble at all passing FizzBuzz, because I don’t interview people unless I think I might want to hire them. I simply don’t have the time to interview 200 people to find one who can pass FizzBuzz. Nobody has that kind of time to waste.

FizzBuzz is pass/pass.

You shouldn’t be interviewing people who can’t pass FizzBuzz. FizzBuzz is trivial. It’s the sort of simple problem that professional developers can’t have trouble with. Asking a professional developer to write a solution to FizzBuzz is like asking a professional mathematician to solve 5+4.

If FizzBuzz is so simple, why even ask it?

I get candidates to solve FizzBuzz because I’m going to actually test their technical skills later in the interview, and I want them to be comfortable. Interviews are stressful, and the best way to take someone from stressed to comfortable is to let them succeed – and not just succeed, but easily succeed. FizzBuzz lets someone with even the most basic programming ability succeed, and that lets them relax – and that makes it easier for them to show me why they’re worth hiring.

“Interviews are stressful” is no excuse.

I’ve seen plenty of people complain that asking developers to write code during an interview is unfair, because interviews are stressful, and that makes it hard for candidates to perform.

Yes. That’s the point. Let me tell you a story.

My team was releasing a new feature to a business-critical site. I do similar things all the time – that’s my job – but this time, something went wrong. The moment someone hit the site, the server went to 100% processor utilization and stopped responding. Ten minutes later, we managed to kill the process and roll the update back. We postponed the update until tomorrow, and started trying to diagnose the problem.

We couldn’t.

Several person-days worth of testing and analysis later, we hadn’t been able to replicate the problem in any of our test environments, so we decided to deploy the new version again (with a few minor tweaks). Once again, the server went to 100% CPU usage, and after about 10 minutes we were able to roll back the update. We were behind schedule, and senior management started to get involved.

Evenings and weekends were cancelled, experts were consulted, and we put a number of measures in place to ensure the new features went out successfully. We rolled back a number of non-critical changes. We put additional testing in place. We put some data collection in place to collect memory dumps, and we deployed – and our production system came to a screaming halt. Everything froze, and we rolled back. Senior management were upset, and my team’s credibility was at stake. Consultants were being brought in. We collected dump files, fired up debuggers – and diagnosed a faulty third-party library which was misbehaving on some edge-case which only happened in production.

Excising the third-party library and getting a working version tested and released wasn’t an easy task, but it had to happen fast. With the problem identified, we wrote a pile of code at very short notice, got it tested, and pushed it into production – and everything worked. The whole situation lasted only a few days, but the pressure to identify and fix the problem was tremendous, and my team was suddenly under a spotlight.

I need to hire people who can write code under stress.

As a team lead, it’s my job to make sure our team doesn’t end up in high-stress, tight-deadline situations. As a manager, it’s my boss’s job to ensure that stress doesn’t get passed on to my team. But sometimes it goes that way, and when it comes right down to it, I want a team filled with people who can write good code in stressful situations.

Professional developers write code.

When you get right down to it, the job of a developer is to write working code. However you boil it down, someone with any kind of experience – even experience as a student – should have spent plenty of time writing software. Trivial problems should be trivial, even under stress (the kind of stress that happens in real life, whether it’s in exams, during assignment periods, or at work) – and in fact, even moderate or difficult problems should be manageable under stress.

Someone who can’t solve FizzBuzz under stress isn’t someone I want on my team.

This is what it gets down to. FizzBuzz is trivial. It’s not the problem: as I discussed earlier, FizzBuzz is the simple introduction, designed to help people relax. I’ve seen it work, over and over: stressed people, nervous in a job-interview situation, are distracted by their interest in writing code. Someone who came in to an interview nervous has an easy win, and goes on to tackle some of the harder technical problems I have for them with confidence.

At the end of it all, if you can’t solve FizzBuzz under interview-stress conditions, I can’t trust you to be on my team.

I’ve probably turned down one or two developers I shouldn’t have, over the years, because they froze and couldn’t solve FizzBuzz in the moment. I have successfully built teams full of successful people, though. It hasn’t been by making people solve a FizzBuzz-like problem before hiring them – but watching candidates try to solve such simple problems has been a key part of deciding whether to hire them or not.

In Summary…

FizzBuzz, on its own, is a terrible way to judge whether to hire someone or not – but it is a tremendously useful tool for a team lead who is trying to decide whether someone will be a great team member or not.

%d bloggers like this: