Richard Brown – 67 Bricks developer blog

Things Customers Don’t Understand About Search

Note, this post is based on a dev forum put together by Chris.

Full-text search is a common feature of systems 67 Bricks build. We want to make it easy for users to find relevant information quickly often through a faceted search function. Understanding user needs and building a top notch user experience is vital. When building faceted search, we generally use either ElasticSearch (or AWS’s OpenSearch) or MarkLogic. Both databases offer very similar feature sets when it comes to search, though one is more targetted towards JSON based documents and the other, XML.

Search can seem magical at first glance and do some amazing things but this can lead to situations where customers (and UX/UI designers) assume the search mecahnism can do more than it can. We frequently find a disconnect between what is desired and what is feasable with search systems.

There are 2 main categories of problems we often see are:

Customers / Designers asking for things that could be done, but often come with nasty performance implications
Features that seem reasonable to ask for at first glance, but once dug into reveal logical problems that make developing the feature near impossible

Faceted Search

Faceted search systems are some of the most common systems we build at 67 Bricks. The user experience typicallly starts with a search box that they enter a number of terms into, hit enter and then be presented with a list of results in some kind of relevancy order. Often there is a count of results displayed alongside a pagination mechanism to iterate through the results (e.g. showing results 1-10 of 12,464). We also show facets, counts for how many results fit into different buckets. All this is handled in a single query that often takes less than 100ms which seems miraculous. Of course, this isn’t magic, full-text search systems use a variety of clever indexes to make searching and computing the facet counts quick.

Lets make a search system for a hypothetical website cottagesearch.com. Our first screen will present the user with some options to select a location, the date range they want to stay and how many guests are coming. We perform the search and show the matching results. How should we display the results and more importantly, how do we show the facets?

Let’s say we did a search for 2 bedroom cottages. We’ve seen wireframes for a number of occassions where the facet count for all bedroom numbers are displayed. So users see the number of results applicable to each bedroom count they would get if they didn’t limit the search to just 2 bedrooms (i.e. there aren’t that many 2 bed options, but look at how many 3 bed options are available). At first glance, this seems like a sensible design, but fundmanetally breaks how search systems work with faceting, they will return counts, but only for the search just done.

We could get around this by doing 2 searches, 1 limited by bedrooms and one that does not to retrieve the facet counts. This may seem like a sensible idea when we have 1 facet, but what do we do when we have more? Do we need to do multiple searches, effectively making an N+1 problem? How to we display numbers? Should the counts for the location facet include the limit of bedrooms or not? As soon as we start exploring additional situations we start to see the challenges the original design presents.

This gets harder when we consider non-exclusionary facets. Let’s say our cottage search system lets you filter by particular features, such as a wood burner, hot tub or dishwasher. Now, if we show counts of non-selected facets, what do these numbers represent? Do they include results that already include the selected facet or not? Here, the logic starts to break down and becomes ever more confusing to the end user and difficult to implement for the developer.

Other Complex Facet Situations

A common question we need to ask with non-exclusionary facets: Is it an AND or an OR search? The answer is very domain dependant, but either way we suggest steering away from facets counts in these situations.

Date ranges provide an interesting problem, some sites will purposefully search outside of the selected range so as to provide results near the selected date range. This may be a useful or annoying depending on what the user expects and is trying to achieve. Some users would want exact matches and would have no interest in results that do not meet the selected date range.

Ordering facets is also a questions that may be overlooked. Do you order lexographically or do you order by descending number of matches? What about names, year ranges or numeric values? Again, a lot of what users expect and would want comes down to the domain being dealt with and the needs of the users.

When users select a new facet, what should the UI do? Should the search immediately rerun and the results and facets update or should there be a manual refresh button the user has to select before the search is updated? An immediate refresh would be slower, but let users narrow down carefully, while a manual update would reduce the number of searches done, but then users may be able to select a number of facets in such a way that no results would be returned.

Hierarchies can also prove tricky. We often see taxonomies being used to inform facets, say subjects with sub categories. How should these be displayed? Again there are many solutions to pick from with different sets of trade-offs.

Advanced Search

Advanced search can often be a bit like a peacocks tail – something that looks impressive, but doesn’t contribute a fair share of value based on how much effort it takes to develop. A lot of designers and product owners love the idea of it but in practice, it can end up being somewhat confusing to use and many end users end up avoiding it.

Boolean builders exist in many systems where the designer of advanced search will insist on allowing users to build up some complex search with lots of boolean AND/OR options, but displaying this to users in a way they can understand is challenging. If a user builds a boolean search such as: GDP AND Argentina OR Brazil do we treat it as (GDP AND Argentina) OR Brazil or should it be interpreted as GDP AND (Argentina OR Brazil). We could include brackets in the builder, but this just further complicates the UI.

We frequently get bugs and feedback on advanced search, some of this feedback can amount to different users having contradictory opinions on how it should work. We would ask product owners to carefully consider “How many people will use it?” Google has a well build UI for advanced search that does away with the challenges of boolean logic by having separate fields for ANDs ORs and NOTs.

An advanced search facility can introduce additional complexity when combined with facets. If an advances search lets you select some facets before completing the search, does this form part of the string in the search box? We have had mixed results with enabling power users to enter facets into search fields (e.g. bedrooms:3), but this can be tricky, some users can deal with it, but others may prefer a advanced search builder while others will rely on facets post search.

Summary

In conclusion, we have 3 main takeaways

Search is much more complex than it first appears
Facets are not magic, just because you can draw a nice wireframe doesn’t make it feasable to develop
Advanced search can be tricky to get right and even then, only used by a minority of users

We’ve build many different types of search and have experimented with a number of approaches in the past and we offer some tried and tested principles:

Make searches stateless – Don’t add complexity by trying to maintain state between facets changes, simply treat each change as a fresh search. That way URLs can act as a method of persistence and bookmarking common searches.
Have facets only display counts for the current search and do not display counts for other facets once one has been selected within that category.
Only use relevancy as the default ordering mechanism – You may be tempted to allow results to be ordered in different ways, such as published date, but this can cause problems with weakly matching, but recent results appearing first.
Don’t build an advanced search unless you really need to and if you have to, use a Google style interface over a boolean query builder.
Check that search is working as expected – Have domain experts check that searches are returning sensible results and look into using analytics to see if users are having a happy journey through the application (i.e. run a search and then find the right result within the first few hits).
Beware of exhaustive search use cases – As many search mechanisms work on some score based on relevancy to the terms entered, having a search that guarantees a return of everything relevant can be tricky to define and to develop.

Dev Forum – Parsing Data

Last Friday we had a dev forum on parsing data that came up as some devs had pressing question on Regex. Dan provided us with a rather nice and detailed overview of different ways to parse data. Often we encounter situations where an input or a data file needs to be parsed so our code can make some sensible use of it.

After the presentation, we looked at some code using the parboiled library with Scala. A simple example of checking if a sequence of various types of brackets has matching open and closing ones in the correct positions was given. For example the sequence ({[<<>>]}) would be considered valid, while the sequence ((({(>>]) would be invalid.

First we define the set of classes that describes the parsed structure:

object BracketParser {

  sealed trait Brackets

  case class RoundBrackets(content: Brackets)
     extends Brackets

  case class SquareBrackets(content: Brackets)
     extends Brackets

  case class AngleBrackets(content: Brackets)
     extends Brackets

  case class CurlyBrackets(content: Brackets)
     extends Brackets

  case object Empty extends Brackets

}

Next, we define the matching rules that parboiled uses:

package com.sixtysevenbricks.examples.parboiled

import com.sixtysevenbricks.examples.parboiled.BracketParser._
import org.parboiled.scala._

class BracketParser extends Parser {

  /**
   * The input should consist of a bracketed expression
   * followed by the special "end of input" marker
   */
  def input: Rule1[Brackets] = rule {
    bracketedExpression ~ EOI
  }

  /**
   * A bracketed expression can be roundBrackets,
   * or squareBrackets, or... or the special empty 
   * expression (which occurs in the middle). Note that
   * because "empty" will always match, it must be listed
   * last
   */
  def bracketedExpression: Rule1[Brackets] = rule {
    roundBrackets | squareBrackets | 
    angleBrackets | curlyBrackets | empty
  }

  /**
   * The empty rule matches an EMPTY expression
   * (which will always succeed) and pushes the Empty
   * case object onto the stack
   */
  def empty: Rule1[Brackets] = rule {
    EMPTY ~> (_ => Empty)
  }

  /**
   * The roundBrackets rule matches a bracketed 
   * expression surrounded by parentheses. If it
   * succeeds, it pushes a RoundBrackets object 
   * onto the stack, containing the content inside
   * the brackets
   */
  def roundBrackets: Rule1[Brackets] = rule {
    "(" ~ bracketedExpression ~ ")" ~~>
         (content => RoundBrackets(content))
  }

  // Remaining matchers
  def squareBrackets: Rule1[Brackets] = rule {
    "[" ~ bracketedExpression ~ "]"  ~~>
        (content => SquareBrackets(content))
  }

  def angleBrackets: Rule1[Brackets] = rule {
    "<" ~ bracketedExpression ~ ">" ~~>
        (content => AngleBrackets(content))
  }

  def curlyBrackets: Rule1[Brackets] = rule {
    "{" ~ bracketedExpression ~ "}" ~~>
        (content => CurlyBrackets(content))
  }


  /**
   * The main entrypoint for parsing.
   * @param expression
   * @return
   */
  def parseExpression(expression: String):
    ParsingResult[Brackets] = {
    ReportingParseRunner(input).run(expression)
  }

}

While this example requires a lot more code to be written than a regex, parsers are more powerful and adaptable. Parboiled seems to be an excellent library with a rather nice syntax for defining them.

To summarize, regexes are very useful, but so are parsers. Start with a regex (or better yet, a pre-existing library that specifically parses your data structure) and if it gets too complex to deal with, consider writing a custom parser.

Organising Git Pull Requests

Recently we had an interesting dev forum on Git workflows. Git is the de facto source control tool of choice for software development. Powerful and flexible, teams have a wide range of choices to make when deciding exactly how to make use of it. This discussion originally stemmed from a Changelog podcast titled “Git your reset on” based on the blog post “Git Organized” by Annie Sexton.

To briefly summarise the above, if we imagine a situation where a merge to main leads to a production deployment and then an issue is found in production, how do we roll back? We look in the git history, find the commit that introduced the bug, revert it and then redeploy to production. But oh no! This revert changed a number of source files beyond the scope of the bug and has ended up introducing a new bug.

The proposed solution to this is to ensure a sensible, clean history of commits is used within a branch. Annie Sexton suggests using a feature branch where you commit regularly with useless messages like WIP until you are ready to submit a pull request. You then run git reset origin/head to be presented with all the changes you have done and how this differs from main, you then progressively add files and make commits with sensible messages to build up a more coherent change.

This approach has a number of advantages we discussed:

It provides a neat history of compartmentalised changes
Reviewers can then follow a structured story of commits about how the developer arrived at their solution
Irrelevant changes are excluded
Ensure that all of the included changes are relevant only to the task at hand

Disadvantages of this approach may include:

Bad approaches that were tried and abandoned are not in the history
Changes aren’t going to always compartmentalise neatly into specific changes of files meaning we would need to commit specific changes within a file, some tools do a bad job supporting this use case

It turns out that a lot of developers at 67 Bricks use a form of history rewriting when developing code, but no one used the specific approach proposed above. git commit–amend is a very useful command to tweak the previous commit. Some (including myself) use git rebase -i as a way of rebuilding the commit history, though this has a limitation in cases where you may want to split a commit up. Finally, others create new branches and build up clean commits on the new branch.

Is force pushing a force for good?

This question really split the developers, some see using git push --force as an evil that indicates you’re using git wrongly and should only be used as a last resort. Others really like the idea of force pushing, but only for branches you’ve been developing on and have yet to create a merge request for.

Squashing?

A workflow some developers have seen before involves using the --squash flag when merging into main. This can create a nice history where each commit neatly maps to a single ticket but the general view was this caused the loss of helpful information from the git history.

Should the software be valid at every commit?

Some argued that this is a sensible thing to strive for, having confidence that you can revert back to any commit and the software will work is a nice place to be at. However, others criticised this as tricky to achieve in practice. Ideally the tests would all be valid and pass too, but this goes against how some developers work; they create a failing test to show the problem or new feature, commit it and then build out code to make the test green. Others claim that tests should change in the same commit as code changes to make it obvious these changes belong together.

Tools used by Developers

At 67 bricks, we try to avoid specifying which precise tool a developer should use for their craft. Different developers prefer different approaches when it comes to using git, some are more comfortable with their IDEs plugin (e.g. IntelliJ has good git integration which can add specific lines to commits) while others may prefer using the command line. Here’s a list of tools we’ve used when dealing with git:

To conclude, specific approaches and tools used vary, but everyone agreed on the benefits of striving for a clean, structured git history.

An often asked question

I thought I’d pen a short blog post on a question I frequently find myself asking.

You’re right, but are you relevant?
Me

As software developers, it’s all too easy to want to select the new hot technologies available to us. Building a simple web app is great, but building a distributed web-app based on an auto-scaling cluster making use of CQRS and an SPA front-end written in the latest JavaScript hotness is just more interesting. Software development is often about finding simple solutions to complex problems, allowing us to minimize complexity.

You’re right, distributed microservices will allow you to deploy separate parts of your app independently, but is that relevant to a back-office system which only needs to be available 9-5?
You’re right, CQRS will allow you to better scale the database to handle a tremendous number of queries in parallel, but is that relevant when we only expect 100 users a day?
You’re right, that new Javascript SPA framework will let you create really compelling, interactive applications, but is that relevant to a basic CRUD app?

Lots of modern technology can do amazing things and there are always compelling reasons to choose technology x, but are those reasons relevant to the problem at hand? If you spend a lot of time up front designing systems with tough-to-implement technologies and approaches which aren’t needed; a lot of development effort would have been needlessly spent adding all kinds of complexity which wasn’t needed. Worse, we could have added lots of complexity that then resists change, preventing us from adapting the system in the direction our users require.

At 67 Bricks, we encourage teams to start with something simple and then iterate upon it, adding complexity only when it is justified. Software systems can be designed to embrace change. An idea that’s the subject of one of my favourite books Building Evolutionary Architectures.

So next time you find yourself architecting a big complex system with lots of cutting edge technologies that allow for all kinds of “-ilities” to be achieved. Be sure to stop and ask yourself. “I’m right, but am I relevant?”

In praise of functional programming

At 67 Bricks, we are big proponents of functional programming. We believe that projects which use it are easier to write, understand and maintain. Scala is one of the most common languages we make use of and we see more object oriented languages like C# are often written with a functional perspective.

This isn’t to say that functional languages are inherently better than any other paradigms. Like any language, it’s perfectly possible to use it poorly and produce something unmaintainable and unreadable.

Immutable First

At first glance immutable programming would be a challenging limitation. Not being able to update the value of a variable is restrictive but that very restriction is what makes functional code easier to understand. Personally I found this to be one of the hardest concepts to wrap my head around when I first started functional programming. How can I write code if I can’t change the value of variables?

In practice, reading code was made much easier as once a value was assigned, it never changed, no matter how long the function was or what was done with the value. No more trying to keep track in my head how a long method changes a specific variable and all the ways it may not thanks to various control flows.

For example, when we pass a list into a method, we know that the reference to the list won’t be able to change, but the values within the list could. We would hope that the function was named something sensible, perhaps with some documentation that makes it clear what it will do to our list but we can never be too sure. The only way to know exactly what is happening is to dive into that method and check for ourselves which then adds to the cognitive load of understanding the code. With a more functional programming language, we know that our list cannot be changed because it is an immutable data structure with an immutable reference to it.

High Level Constructs

Functional code can often be more readable than object oriented code thanks to the various higher level functions and constructs like pattern matching. Naturally readability of code is very dependent on the programmer; it’s easy to make unreadable code in any language, but in the right hands these higher level constructs make the intended logic easier to understand and change. For example, here are 2 examples of code using regular expressions in Scala and C#:

def getFriendlyTime(string time): String = {
  val timestampRegex = "([0-9]{2}):([0-9]{2}):([0-9]{2}).([0-9]{3})".r
  time match {
    case timestampRegex(hour, minutes, _, _) => s"It's $minutes minutes after $hour"
    case _ => "We don't know"
  }
}

public static String GetFriendlyTime(String time) {
  var timestampRegex = new Regex("([0-9]{2}):([0-9]{2}):([0-9]{2}).([0-9]{3})");
  var match = timestampRegex.Match(time);
  if (match == null) {
    return "We don't know";
  } else {
    var hours = match.Groups[2].Value;
    var minutes = match.Groups[1].Value;
    return $"It's {minutes} minutes after {hour}";
  }
}

I would argue that the pattern matching of Scala really helps to create clear, concise code. As time goes on features from functional languages like pattern matching keep appearing in less functional ones like C#. A more Java based example would be the streams library, inspired by approaches found in functional programming.

This isn’t always the case, there are some situations where say a simple for loop is easier to understand than a complex fold operation. Luckily, Scala is a hybrid language, providing access to object oriented and procedural styles of programming that can be used when the situation calls for it. This flexibility helps programmers pick the style that best suits the problem at hand to make understandable and maintainable codebases.

Pure Functions and Composition

Pure functions are easier to test than impure functions. If it’s simply a case of calling a method with some arguments and always getting back the same answer, tests are going to be reliable and repeatable. If a method is part of a class which maintains lots of complex internal state, then testing is going to be unavoidably more complex and fragile. Worse is when the class requires lots of dependencies to be passed into it, these may need to be mocked or stubbed which can then lead to even more fragile tests.

Scala and other functional languages encourage developers to write small pure functions and then compose them together in larger more complex functions. This helps to minimise the amount of internal state we make use of and makes automated testing that much easier. With side effect causing code pushed to the edges (such as database access, client code to call other services etc.), it’s much easier to change how they are implemented (say if we wanted to change the type of database or change how a service call is made) making evolution of the code easier.

For example, take some possible Play controller code:

def updateUsername(id: int, name: string) = action {
  val someUser = userRepo.get(id)
  someUser match {
    case Some(user) =>
      user.updateName(name) match {
        case Some(updatedUser) =>     
         userRepo.save(updatedUser)
         bus.emit(updatedUser.events)
         Ok()
        None => BadRequest()
    case None => NotFound()
}

def updateEmail(id: int, email: string) = action {
  val someUser = userRepo.get(id)
  someUser match {
    case Some(user) =>
      user.updateEmail(email) match {
        case Some(updatedUser) =>
          userRepo.save(updatedUser)
          bus.emit(updatedUser.events)
          Ok()
        case None => BadRequest()
    case None => NotFound()
}

Using composition, the common elements can be extracted and we can make a more dense method by removing duplication.

def updateUsername(id: int, name: string) = action {
  updateEntityAndSave(userRepo)(id)(_.updateName(name))
}

def updateEmail(id: int, email: string) = action {
  updateEntityAndSave(userRepo)(id)(_.updateEmail(email))
  }
}

private def updateEntityAndSave(repo: Repo[T])(id: int)(f: T => Option[T]): Result = {
  repo.get(id) match {
    case Some(entity) =>
      f(entity) match {
        case Some(updatedEntity) =>
         repo.save(updatedEntity)
         bus.emit(updatedEntity.events)
         Ok()
        None => BadRequest()
    case None => NotFound()
}

None Not Null

Dubbed the billion-dollar mistake by Tony Hoare, nulls can be a source of great nuisance for programmers. Null Pointer Exceptions are all too often encountered and confusion reigns over whether null is a valid value or not. In some situations a null value may never happen and need not be worried about while in others it’s a perfectly valid value that has to be considered. In many OOP languages, there is no way of knowing if an object we get back from a method can be null or not without either jumping into said method or relying on documentation that often drifts or does not exist.

Functional languages avoid these problems simply by not having null (often Unit is used to represent a method that doesn’t return anything). In situations where we may want it, Scala provides the helpful Option type. This makes it explicit to the reader that you call this method, you will get back Some value or None. Even C# is now introducing features like Nullable Reference Types to help reduce the possible harm of null.

Scala also goes a step further, providing an Either construct which can contain a success (Right) of failure (Left) value. These can be chained together to using a railway styles of programming. This chaining approach can lead us to have an easily readable description of what’s happening and push all our error handling to the end, rather than sprinkle it among the code.

Using Option can also improve readability when interacting with Java code. For example, a method returning null can be pumped through an Option and combined with a match statement to lead to a neater result.

Option(someNullReturningMethod()) match {
  case Some(result) =>
    // do something
  case None =>
    // method returned null. Bad method!
}

Conclusions

There are many reasons to prefer functional approaches to programming. Even in less functionally orientated languages I have found myself reaching for functional constructs to arrange my code. I think it’s something all software developers should try learning and applying in their code.

Naturally, not every developer enjoys FP, it is not a silver bullet that will overcome all our challenges. Even within 67 Bricks we have differing opinions on which parts of functional programming are useful and which are not (scala implicits can be quite divisive). It’s ultimately just another tool in our expansive toolkit that helps us to craft correct, readable, flexible and functioning software.

The First 67 Bricks Architectural Kata

On the 13th of October, 67 Bricks held its first Architectural Kata as an opportunity for developers to practice and experience architectural design. Ted Neward, the original creator of architectural katas, puts forward the argument for them quite succinctly:

  “So how are we supposed to get great architects, if
  they only get the chance to architect fewer than
  a half-dozen times in their career?”
Ted Neward

The kata exercise was quite simple to prepare for and run. Beforehand I picked a number of interesting katas from Neal Ford’s list, suitably anglicised them and prepared handouts for both online and in-person teams. Then on the day, it was a case of assigning everyone into teams and explaining the steps we’d be going through that day:

Gather into teams and read through the assigned case study
Discuss for the next 50 minutes and come up with an architecture, making sure to document any assumptions. Teams could ask me questions about the case study and I’d happily provide extra detail
Each team would then take turns presenting their architecture to the other teams and fielding questions from them
Finally a voting phase where everyone else gets the chance to show thumbs up/down/sideways to indicate how well they thought the presenting team did

Drivers

As a technical consultancy, we are often involved in architectural design of systems we create or take part in creating. Each client is individual; they seek different opportunities, are subject to different constraints and have different technical strategies. It’s critical for 67 Bricks’ success to have skilled architects able to influence and design suitable systems.

Though, how can we develop these skills? As Ted Neward identified, most only get a half-dozen tries at it over their career. While books, online courses and talks help, knowledge needs to be applied and feedback loops closed to truly improve. We’ve never tried an event like a Kata before and I was interested to see what we could learn from it.

Sharing ideas and learning from one another effectively can be a challenge for any organization and made all the harder thanks to COVID-19 prompting a rapid move to remote working. The exercise looked to provide a great opportunity for participants to meet and work with others they may not otherwise have the chance to.

Kata Afternoon

Initially I hoped to get some of the more experienced technical leads to kick off the afternoon by talking about how they architect systems. This proved difficult. Those who I spoke to either thought they didn’t know that much or didn’t feel like they could speak to an audience well about how to do architecture. As such these initial talks didn’t happen and raised a lot of questions around how to arm participants with enough knowhow to feel comfortable tackling the task. I fear this may have led to some teams struggling too much to learn and make effective progress.

Splitting teams took some thought, I wanted to be sure each team had a mix of experience and tried to aim for teams of individuals who haven’t worked together before. I’m glad I spent the time to do this, the Kata would not have had the same impact if everyone was in the same team they usually are (one of the Kata rules is to try and break up regular teams).

Running the actual Kata was reasonably straightforward, if a little awkward thanks to having to manage both online and an in-person group. Some online teams got stuck waiting for me to join their Zoom room to answer their questions. While the in-person group was easy to recognise if they needed a question answered or a nudge in the right direction.

I tried to prepare some broader scope for each case study before the event in preparation for questions being asked, I quickly found myself having to improvise. I actually found this surprisingly fun. Some teams may have made things harder for themselves by asking too many detailed questions and trying to cover every single detail in their architecture.

The presentation phase had mixed results, some teams did a strong job and were able to present their ideas well, answer questions clearly and came up with really suitable architectures. Other teams struggled a bit more, both with coming up with a suitable architecture and being able to communicate it with everyone else.

Feedback and Lessons Learned

Feedback was overwhelmingly positive, lots of participants really enjoyed the experience. They liked working with different people and felt they learned a lot from one another about how they tackle this type of task.

I think aiming for mixed team compositions was good and allowed individuals to have interactions with colleagues they wouldn’t normally work with. I’ve had feedback a number of times that with the move to remote, developers felt a lot more insular in their teams and I hoped this gave a chance to break out of those groups.

Doing both in-person and remote teams did make running the Kata a little challenging. I felt that I couldn’t effectively keep an eye on all the teams making it tough to know when to nudge a team that may be getting a little stuck or provide an answer to a pressing question.

Next time I would look at including publishing consultants to act as clients. I perhaps enjoyed being a fickle client a little too much. Having a more dedicated individual to represent the client creates opportunities for encouraging more of a discussion, simulating something closer to the real world when we work with our clients. It would also make it easier to manage the event and our publishing consultants can gain some insight to the architecture process.

Conclusions

Given the positive response, I would happily organise another Kata in the future. If you are thinking about running one for your own organisation, I would highly recommend it. It’s a great chance to meet and learn from other devs. I certainly learned a lot simply observing the various teams taking different approaches to their case studies and the variety of solutions they came up with.

Sharing Failures Dev Forum: The most “interesting” bugs you’ve caused

Ely (n.)

The first, tiniest inkling you get that something, somewhere, has gone terribly wrong.

The Deeper Meaning of Liff – Douglas Adams & John Lloyd (Thanks to Stephen for pointing this quote out)

At 67 Bricks we value open and honest communication, and aim to treat each other with respect and humility. Normalising failure helps everyone understand that mistakes happen and shouldn’t be used as a mechanism to punish but instead are an opportunity to learn and become more resilient to future failures. This week’s Dev Forum was on the topic of sharing failures, specifically talking about the most “interesting” bugs you have caused.

64 Bit vs 32 Bit Ubuntu

Our first fail came from a developer working on converting data from base 64 as part of a side project. Developing on the 64 bit version of Ubuntu, the code was stable and the tests were green and so they continued to build the application out. The application was intended to work on a number of different environments including 32 bit systems and this is where the first problems emerged.

The 64 bit OS zeroed out data after it was read and the code was designed to work with this, the 32 bit OS didn’t and instead exhibited random test failures. Sometimes expected values would match up and other times it would be wrong by a character or two, typically at the end of the strings. Finding the root cause required a trial and error approach to eliminate possibilities and eventually get a grasp on the problem and come up with a solution.

Transactions and Logging

Working on a client application this developer made two perfectly reasonable design decisions:

Use transactions with the SQL database to ensure integrity of data changes
Write logs to a table to ensure they are persisted in a convenient location

This approach was perfectly fine while the system worked, but the mystery began when the client started to complain about errors with the system. The logs didn’t show any error messages that lined up with what the client was reporting. Why? The same transaction was used to store logs and to update the data in SQL. If an exception was thrown, the transaction was rolled back preventing any data corruption problems, but also rolled back all the logs!

The application was changed to use a different transaction for logging to ensure logs were persisted. Using these logs meant the root cause of the client problem could be resolved and a lesson learnt about unintended consequences.

Overlogging

Another mystery around missing logs. A web application used a separate log shipping application to take logs from the server and send them to a remote server. However, under heavy loads the logs would become spotty and clear gaps appeared. The reason for this was due to the sheer volume of logs the shipper had to deal with eventually becoming too great and causing it to crash.

The solution was to reduce the number of logs written so the shipper would continue to function even when the main application was under heavy load. This triggered an interesting conversation at the dev forum on the ideas of how many logs should be written. Should you write everything including debug level logs to help with debugging faulty systems? Or should you write no logs whatsoever and only enable them when something starts going wrong?

Naturally we seemed to settle somewhere in the middle, but there were disagreements. Possibly a future dev forum topic?

The Wrong Emails

A client needed to receive important data daily via fax and the application did this by sending an email from a VB app to an address which would convert the body to a fax and send it on. The application needed testing and so had a separate test environment that was presumed to be safe to use for running tests.

It turned out that the test system was using the same email address as the live system, meaning the client was receiving both test and live data. Luckily the client caught on and asked why they were receiving two emails a day and the test system was quickly updated.

From then on this developer now always uses some form of override mechanism to prevent sending actual emails to end users. Others use apps like smtp4dev which will never send on emails (and just capture them so they can be inspected) or services like SES which can be configured to run in a sandbox mode.

Hidden Database Costs

AWS can be very costly, especially if no one is watching the monthly spend. In this case, one database drove up costs to astronomical new highs due to a single lambda running regularly and reading the entire database. The lambda was supposed to delete old data, but had a bug which meant it never deleted anything. Several months of operating the system meant a lot of data had piled up and so did the costs. The fix was simple enough to apply and we added additional monitoring to catch cases of runaway costs.

Similar experiences have been had before with runaway costs on clustered databases like MarkLogic. Normally a well-built index will be able to efficiently query the right data on the right node, but if the index is missing, MarkLogic will pull all the data across from the other node to evaluate it. This can drive up some eye-wateringly high data transfer prices in AWS and as such we now always monitor data transfer costs to ensure we haven’t missed an index.

Caching Fun

Our next issue is about users appearing to be logged in as other users. The system in question used CloudFront CDN to reduce server load and response times. The CDN differentiated between authenticated and unauthenticated users so different pages could be served for users who were authenticated.

The system made use of various headers set by lambdas to differentiate between authenticated and unauthenticated users. The problem arose when session handling was changed and the identifier used was accidentally stored within the CDN. This caused an issue where a user could load a page with a set-cookie header that set the identifier used for a different user’s session.

The team solved this bug by tweaking the edge lambdas to ensure only non-user specific data was cached. Caching in authenticated contexts can be challenging and need to be very carefully considered how they will be used reliably.

Deactivate / Activate

In this bug the business asked for a feature where user accounts could have a deactivated time set in the future that when reached, the user would be considered inactive and unable to access the system. This feature was implemented with a computed field in the SQL server which could be used to determine if the user is active or not.

As the system was already in use, migration scripts were developed to update the database. These needed to be applied in a certain order, however, deployment practices for this system meant that someone else applied the scripts and ended up causing an error where all users ended up deactivated preventing any users from accessing the system. To restore service, the database was rolled back and ultimately the feature was abandoned as too risky by the business.

Some viewed this as an example of why services should be responsible for their own database state and handle database migrations automatically.

Creating Bugs for Google to Index

Magic links can be a very useful feature for making authenticated access to a website easy, as long as these urls remain private to the correct user. In one case the url got cached by Google including the authentication token meaning anyone who could find the link would be able to access the authenticated content! This was fixed by asking Google to remove the URL, invalidating the token in the database and ensuring metadata was added to appropriate pages to prevent bots from indexing pages.

Another Google based bug next; after building a new system, the old one needed to be retired and part of this involved setting up permanent redirects to the new system. However, Google continues to serve up the old site’s urls as opposed to the new system’s urls, and a fix is still being worked out. A lesson learned on how important it is to carefully consider how search engines will crawl and store web sites.

As we see, failures can come from any number of sources, including ourselves. Bugs are a perfectly normal occurrence and working through them is an unavoidable part of building a robust system. By not fearing them, we can become more adept at fixing them and build better systems as a result.