Category: Books

“Observability Engineering” Book Review

A great, new book “Observability Engineering” came out very recently and I had to jump on reading it. Since it is very closely related to my work, I devoured the pages and read the book in about a day (505 pages). While doing so I wrote down some thoughts that I want to share with you today. They might or might not be true, I am only speaking about the book from my own perspective. Feel free to share your own thoughts!

Overall, the book really resonated with me and it makes me very happy to see literature being written about this topic. Observability is a relatively novel concept in computing that I think will only become more popular in the future. I’d rate the book 4/5 in general but it is 5/5 between books on the same topic.

Here are my thoughts.

  • First of all, it is interesting to see tracing used in CI processes to reduce flakiness. But this probably only matters on a huge scale that most companies will not achieve. At least I haven’t worked at companies so far where it is the case. This has also reminded me of a project to put Kubernetes events as spans. Check it out if you’re interested. I hope to work on distributed tracing projects in the near future, it’s a really exciting topic.
  • Chapters by Slack engineers sometimes felt a bit like an advertisement for Honeycomb. The chapter about telemetry pipelines and their bespoke solutions felt a bit too simplistic because we have things like Vector nowadays not to mention Filebeat and so on. What’s more, Slack engineers have created their own format for storing spans. It seems like a lot of companies nowadays suffer from the “not invented here” syndrome which seems to be the case here. I would be surprised if they won’t migrate to OpenTelemetry (OTel) data format in the near future.
  • Authors spent lots of time talking about and praising OTel. Given that traces are specifically formatted logs, it’s not surprising to see the popularity of OTel. It’s a really exciting project. But we have to keep thinking about events in a system that mutates its state. Traces are only a way of expressing those changes in state.
  • The chapters about finding observability allies are enlightening. I have never thought about customer support and other people as allies that could help one instill a culture of observability in a company.
  • The observability maturity model is great and I could foresee it being used extensively.
  • Event-based service level objectives (SLOs) should be preferred to time-based ones because with distributed systems partial outages are more common than complete blackouts. Event-based SLOs is where you count the good events and the bad events in a window and divide the number of good events by the total number of events. Whereas in time-based SLOs you need to divide the time where some threshold has been exceeded by the amount of time in the window. Also, event-based SLOs reflect the reality more – instead of judging each period of time as either bad or good, with event-based SLOs it is possible to precisely tell how much error budget we’ve burned. Somehow even though I’ve worked with monitoring systems for a long time, such two different points of view escaped me. I will always try to prefer event-based monitoring now.
  • At my previous companies, I saw the same bad practices as outlined in the book. If there are barely any requests in the middle of the night then one or two failures don’t mean much and it’s not needed to alert on those conditions. I am talking about payment failures in the middle of the night if most of your clients are in one or several related timezones, for example. What’s more, I have experienced a bunch of alerts based on symptoms that don’t scale. For example, there are such alerts as “RAM/CPU is used too much”. Just like the authors, I would be in favor of removing them because they are pretty much useless and is reminiscent of the old way of using monitoring systems. I guess this is associated with the observability maturity model that is outlined in the book. My anecdotal data says that many companies are still in their infancy in terms of observability.
  • Lots of text about arbitrarily wide structured events. In an ideal world, we could deduce the internal status of service through them but I believe that it is not it all and not end it all signal. It is just one of many. If instrumentation is not perfect then it is a compression of the state space of your application. And with too much instrumentation there is a risk of high storage costs and too much noise. Sometimes it sounds like a solution to a problem that should be solved in other ways – making services with clearer boundaries and less state. Or, in other words, reduce the sprawling complexity by reducing non-essential complexity to a minimum.
  • I agree with the small section about AIOps (artificial intelligence operations). In general, I feel that it applies to anomaly-based alerting as well. How can computers tell whether some anomaly is bad or not? Instead, we should let computers sift through piles of data and humans should attach meaning to events.
  • I agree with the authors’ arguments about monitoring – again, I believe it’s a cheap signal that is easy to start with, and in my opinion, that’s why so many people rely on it / start with it. It is the same with logs. It is very simple to start emitting them. Distributed tracing takes a lot more effort because you not only have to think about your state but also how your service interacts with others. But, that’s where all of the most important observations lie in the cloud-native world.
  • The book is missing a comparison of different types of signals. The authors really drive the point of arbitrarily wide events but I feel like that isn’t the silver bullet. What about continuous profiling and other emerging signals? Probably not surprising given how much the authors talk about this topic on Twitter.
  • The example of how a columnar database works didn’t convince me and it felt out of place. It probably just needs a better explanation and/or a longer chapter. I would probably recommend you pick up a different book to understand the intricacies of different types of databases.

Of course, my notes here can’t represent all of the content of the book. I’d recommend you to read it yourself! It’s really great. Let me know what you think about it in the comments.

Related Posts

Designing API Like It Is An Everyday Thing

Recently I have read a quite popular book called “The Design of Everyday Things”. I feel that with software slowly taking over more and more parts of the world, we could say that it also became an everyday thing and that we should design it like that. This blog post will be about it.

There are certain, general design principles that were explained in that book which we should apply to the process of programming in general.

We will talk about them in terms of APIs – the programmable interfaces of applications. It is a form of an interface and one of the most prevalent ones. Having these design principles in mind should help us design better APIs.

I feel that a lot of these concepts were already expressed in books such as “Clean Code” by Robert C. Martin but nonetheless, it is interesting to look at those principles from the APIs perspective and from the general design of items – hopefully, we will learn something new.

Mode Errors

The first thing I want to start with is what Don Norman calls a group of errors called “mode errors”. Essentially, it occurs when a device can be in many different states and then the user becomes overwhelmed: they simply do not know which mode they need to use or what it even is at the moment. In the same regard, if we were to treat an API as an everyday thing, we should strive to get rid of these type of errors.

This means that the number of combinations of different values an API call can have must be reduced to the minimum. To be more precise, your API should try to compute as much stuff as possible unless it becomes an actual impediment to the performance. Thus, we should opt to calculate values which are of O(1) or, at most, O(n) complexity.

Also, the values themselves, if they are enumerations, should not have duplicate meanings, and minimal types should be used that are able to hold the needed information. This example may seem a bit superficial but, for instance, if your API that uses JSON only needs to accept numbers then it is probably much more useful to actually use a number type instead of accepting a string and only then converting it into a number.

We need to focus on our users and understand that they are constantly being interrupted by others, the attention span might not be as big, and that it is much more complex for a fresh person to understand all of the different modes that your API might be in. There needs to be a clear signal of state.

Communicate When Things Go Wrong

Feedback and feed-forward cycles are very important. This may seem a bit obvious but this still happens from time to time. In essence, we should always report errors when possible to the caller when something goes wrong (and it does inevitably). This feeds a bit into the previous tip in that the state should always be clear.

Practically, I think it means that your programming language should ideally support sum types which would force you to check for errors and report these errors accordingly to the caller of the function. For example, Rust has the std::result type for this exact purpose.

This also means that if you are making a library that is supposed to be re-used by others then it should absolutely not abruptly make the whole process exit. For example, calling os.Exit(1) in a Go program when something goes wrong is a huge no-no.

This one glog library is notorious for that. It has been written some time ago so it does not follow this recommendation. You can see that if it fails, for example, to rotate files inside of the library then it exits the whole process. But why would it do that at all? Let’s say os.Stderr is closed so the user would not know at all why their program might exit. The rotation could be nice but it should probably be left to external and well-tested programs.

Discoverability and Understanding

The next thing to keep in mind is to make easy to understand what kind of things we can even do with the API mostly after performing some actions. For instance, if you are designing a REST API then you will use verbs according to their meaning and your API will support easy-to-understand objects. This will make it easy for your users to intuitively discover what is possible.

Another common pattern is to provide links in your API responses to other things that can be done. For example, the Hypermedia API language (HAL) format uses (optionally) the _links key in the JSON responses to indicate where else the user could go to do certain actions. Or, usually APIs nowadays include pagination links in the response. The client then can go on to those URLs to do those respective actions.

In essence, it is conceptually the same as having some kind of links or buttons in real-life interfaces or response dialogues with simple verbs which would do certain actions. This is the same principle adopted for API designs.

Affordances And Signifiers

Norman formulates affordance as a property of an object – that you can do something with it, and signifiers “tell” the human what kind of operations are possible. This is in a way connected to the former point of “Discoverability”, however, not completely. Signifiers should be visible to the users from the outlook, without having performed any actions. What this means for us that we need to have some kind of way to increase the number of signifiers.

Usually, this comes up as having well-formed documentation. Nowadays it is very common to include simple interfaces like Swagger which signify to the user what kind of actions are possible.

Image result for swagger
An example Swagger UI

As you can see, all of the possible actions are presented as a neat table. This tells the user what can they afford to do via using the API. There are, of course, competing solutions but if not Swagger then we should still strive to have some kind of interface like this.

Because if there is not then it becomes hard to understand what the API lets us do without reading the actual source code. And that is like having to get into the mind of the designer/developer of the thing which is what we are trying to avoid. Ideally, this interface should be generated from the source code itself. For more information, refer to Chapter 29 “It’s Just A View” from the book “Pragmatic Programmer“.

Constraints

Constraints allow the user to just intuitively know how different parts should fit together. We can think of this in terms of a standard library of functions/classes that some framework, your API provides.

In my opinion, a good example of this is the Python mantra:

PEP 20

There should be one – and preferably only one – obvious way to do it.

Some people might argue that with all of the new additions (and the old relics in the Python’s standard library) this is not so true anymore but still, we should strive to achieve this.

As always, the goal is to reduce the likelihood of an error and accidental complexity. If we were to have more than one way of doing things then we would start having questions like:

  • which functions or class should we use in what case?
  • which way has the bigger efficacy?
  • and so on.

On the other hand, the antithesis of this, in my belief, is the C++ programming language. Over the years it had accumulated a lot of historical cruft due to always trying to be backward compatible and other reasons. Modern C++ language style guides even recommend you to only use a subset of the language itself. That is how bad it became. For instance, it is forbidden to use exceptions even though they are in the language itself.

Mappings

Last but not least, Don has introduced a concept of mappings. The actions that are available to the user should map logically to the items that are provided to them. He gave an example of gas controls on a stove – the controls should clearly be connected to the different outputs.

In our case, you probably would not want to include random methods in your API specification which are completely unrelated to it. Also, the methods should do operations only on items that you have passed to it. Otherwise, you might be running into the risk of creating unclear relationships between different parts of your API.

Conclusion

I liked this book a lot – since the start, I was completely hooked by it and it was a page-turner. The paradoxical book cover caught my eye and I just had to read it.

It has brought me some perspective over software design from the point of view of the general design of things around us that we use every day. With software becoming more and more prevalent, I think that the concepts introduced here will be more widely adopted and respected.

Related Posts