For every 9 you add to SLO, you’re making the system 10x more reliable but also 10x more expensive.

  Alex Ewerlöf

In this incident story, the feature flags were served by the main application server. When a new feature caused the server to crash, there was no way to flip the flag back off to stop the crashes.

  rachelbythebay

The author of a classification system for human error reflects 20 years later on the harm that such systems can cause by using deficit-based language.

  Dr. Steven Shorrock

Here’s Fred Hebert’s analysis of Cloudflare’s write-up of their incident on November 2.

I’m hoping they’re going to do a more in-depth review.

  Fred Hebert — VOID

In this post, we introduce a hybrid approach that seamlessly combines the precision of manual instrumentation with the comfort, efficiency, and performance of automatic instrumentation.

  Ron Federman — Odigos

Change is not the problem. It’s unaddressed risk

  Bruce Johnston — High Scalability

A shell script with a loop running a DB client can fill up your ephemeral ports in a hurry.

  Oren Eini — RavenDB

When you get right down to it, it’s all human communication, even assembly code. It’s human factors all the way down.

  Michael Hart

A great overview of SLIs, covering event-based vs time-based SLIs, commonly used SLIs, and examples of things that don’t make good SLIs.

  Alex Ewerlöf

When it’s time to declare an incident, I want to spend ten seconds or less getting things kicked off.

  Matilda Hultgren — incident.io

This short article covers three important aspects of error budgets:

1. Understanding Your Error Budget
2. Make Informed Decisions
3. Proactively communicate

  Code Reliant

SRE’s Golden Signals are four key metrics used to monitor the health of your service and underlying systems. We will explain what they are, and how they can help you improve service performance.

  Blameless   Full disclosure: Honeycomb, my employer, is mentioned.

I hadn’t really appreciated some of the subtler details of CPU requests in k8s until I read this.

  Ara Pulido — Datadog

Reading this, I can see hints of the contributing factors in many incidents I’ve been involved in.

To these folks, it feels like giving a damn is a huge career liability in your organization. Because it is.

  David Caudill

They went to impressive lengths to make the upgrade process reversible.

Amusingly, this post was directly relevant to me 30 minutes ago when I discovered mojibake all over sreweekly.com due to upgrading MySQL from 5.7 to 8.0+ last week. Oops.

  Jiaqi Liu, Daniel Rogart, and Xin Wu — GitHub

In order to learn from incidents, we need to know that they happened. That means someone needs to report them, but a lot can get in the way of reporting incidents.

  Dr. Steven Shorrock — Humanistic Systems

Wow, this interactive tool for choosing SLOs is fun to play with! Dragging the sliders really gives you a feel for the math involved, and then you get a formula that you can actually use.

  Alex Ewerlöf

A riveting story of a service that was the victim of its own success, a potential solution, and then further challenges to overcome.

  Tanat Lokejaroenlarb — Adevinta

Here’s a classic example of “work as imagined” vs “work as done”, as health care workers struggle against difficult security constraints while trying to care for patients.

  Fred Hebert — summary
  Ross Koppel, Sean Smith, Jim Blythe, and Vijay Kothari — original paper

This article covers a lot of ground, touching on a lot of components of a successful SRE program, and even includes a code example for SLO calculation.

  Vishal Padghan — Squadcast

More on the weird EBS performance regression I linked to last week. Still no full explanation of what changed, but at least they have a solution (gp3 volumes).

  Dustin Brown — dolthub

After a massive 73-hour outage, Roblox set out to redesign their infrastructure to make that kind of incident much less likely. They’ve charted a path through several intermediate architectures, with the ultimate goal of active-active datacenters.

  Daniel Sturman, Max Ross, and Michael Wolf — Roblox

Now here’s one that really makes me think. I can’t really summarize it in a sentence, so just go read it.

  Lorin Hochstein

Maybe you’re thinking of skipping over “yet another article about blamelessness”? Don’t. This one has some great examples and stories and is well worth a read.

  Michael Hart

I’m definitely guilty of a couple of these.

  Code Reliant

New podcast relevant to our interests!

In this series, you’ll hear insightful conversations with engineers, product managers, co-founders and more, all about the debatable topic of incident management.

  Luis Gonzalez — incident.io

A puzzling performance regression in EBS volumes, seemingly reproducible across instances. Anyone else seeing anything like this?

  Dustin Brown — dolthub

This article presents a framework for scaling SRE teams by defining SRE processes, automating, and iterating.

   Stelios Manioudakis — DZone

Some tips on what makes a good alert and how to design your alerts to be actually useful, rather than just noise.

  Leon Adato — Kentik

Why would you want multiple different targets for the same SLO? Read this one to find out.

  Alex Ewerlöf

Conflict-free Replicated Data Types are powerful, but they have downsides explained in this article, so it’d be great if we could avoid them when possible.

  Zak Knill

The network is not reliable. What are the implications and what can we do about it?

  Anadi Misra

Beyond a run-of-the-mill severity levels article, this one goes into a couple of common pitfalls.

  Jonathan Word

Some good tips in here, esp. the one about brevity.

  Ashley Sawatsky — Rootly

Subtitle:

Or, Eleven things we have learned as Site Reliability Engineers at Google

   Adrienne Walcer, Kavita Guliani, Mikel Ward, Sunny Hsiao, and Vrai Stacey — Google

Good lessons to learn here that apply more broadly than just EKS.

  Christian Alexánder Polanco Valdez — Adevinta

This article is about project management, but a lot of the skills discussed apply to aspects of SRE at Staff+ levels.

  Sannie Lee — Thoughtworks (via martinfowler.com)

Now this is more like it: there’s a healthy does of skepticism woven through this article, including things genAI probably won’t be good for, and potential pitfalls.

  Jesse Robbins — Heavybit

There are two different ways of alerting on SLOs, for two very different audiences, as explained in this article. Ostensibly this is a product feature announcement, but you don’t need to be using the product to get a lot out of this.

  Fred Hebert — Honeycomb
  Full disclosure: Honeycomb is my employer.