Should you keep things anonymous (“an engineer”), or should you say exactly who did what? Here’s a solid argument for the latter.

Lorin Hochstein

This article explores the downsides to a design composed of independent parts such as with microservices.

Ephraim Baron

Uber designed a tool they call Blackbox to perform simulated user requests and measure availability. I was struck by the candid discussion of complexity — no one person can understand how all of Uber’s microservices go together.

Carissa Blossom — Uber

They’ve made a YAML specification and validator for expressing SLOs in a machine-readable format.

Mike Vizard — Devops.com

A new spin: this one makes the distinction between “experimental tools” that affect the state of the system, and “observability tools” that are read-only.

Brendan Gregg

“Contributing factors: moose and squirrel.”

JJ Tang — Rootly

Every once in awhile, I need to pull out gdb. In times like those, it’s useful to have this kind of thing floating around in the back of my mind.

Brendon Scheinman — okcupid

This is an in-progress document about the kinds of patterns we see or use when designing systems. The author warned me that it’s a work in progress and maybe not ready for prime-time, but I think this is exactly the time when I should get it in front of your eyes.

I’d love your help growing this list. If you know of a name that is missing from the list please send me a tweet with the name and a short description of it and I’ll include it in the list with a link to your tweet

Mads Hartmann

Whoa, a podcast dedicated to picking apart public incident postings! I love this, because there’s a lot that’s left to shorthand, and a live conversation is a great way to flesh it out.

Tom Kleinpeter and Jamie Turner

There’s a really interesting undercurrent in this story about resilience. Nurses can catch these kinds of errors, but this just one layered protection among many. If the system is reduced to relying on that second-layer defense, the overall resilience is diminished.

Daniel Keane — ABC News

Of course, before reaching this stage, all of the pieces are tested in isolation. But until they’re all put together, it’s almost impossible to predict the behavior of the finished product during an accident.

Mikolaj Pawlikowski

The attributes discussed are:

• Problem solving
• Awareness building
• Collaboration
• Empathy

Jayne Groll

Wait, more attributes? Oh, and by the same author, too:

• “Great SREs have a passion for high-quality automation.”
• “A great SRE ensures SLOs (Service Level Objectives) are set at correct boundaries of service; […]”
• Prize Communication.
• Look for longer-term support experience.
• Look for a person that demonstrates empathy.

Jayne Groll

This one explore the application of SRE principles to mobile app design.

Abhijith Krishnappa

This two-part series uses a narrative case study format to show how SLOs can be misleading. You might have great numbers, but what are the numbers actually measuring?

Adam Hammond — Squadcast

We built Edgar to ease this burden, by empowering our users to troubleshoot distributed systems efficiently with the help of a summarized presentation of request tracing, logs, analysis, and metadata.

Kevin Lew, Maulik Pandey, Narayanan Arunachalam, Dustin Haffner, Andrei Ushakov, Seth Katz, Greg Burrell, Ram Vaithilingam, Mike Smith and Elizabeth Carretto — Netflix

The PDF covers 5 main areas:

1. Availability
2. Performance
3. Monitoring
4. Incident Response
5. Preparation

No account required or form to fill out to download the PDF.

Splunk/VictorOps

This one’s especially interesting for the section about what MTTx metrics aren’t good for, and the following section on how to improve them.

Emily Arnott — Blameless

If you’re interested in deploying Kafka in a multi-region configuration, eBay has put quite a bit of thought into this and has a lot to share.

Engin Yoeyen — eBay

Straight from someone who was there from the start. The “what chaos engineering is not” section is especially enlightening.

Casey Rosenthal — Verica

The last paragraph regarding “unknown unknowns” is noteworthy.

Heroku

There are some great questions in here on blamelessness and full service ownership.

James Thigpen — Gremlin

The SRE book has a chapter covering on-call, but it’s best suited for huge-scale companies. What should the rest of us do?

Utsav Shah

If you’re feeling hesitant about chaos engineering, or you’re trying to convince someone who is, this might be useful. The myths are:

Myth #1: Chaos engineering is testing in production
Myth #2: Chaos engineering is about randomly breaking things
Myth #3: Chaos engineering is only for large, modern distributed systems
Myth #4: We don’t need more chaos – we already have plenty!
Myth #5: Chaos engineering is only for very mature teams/products

Mikolaj Pawlikowski

Drawing parallels to the high modernism movement during the cold war, this article raises interesting questions about the direction SRE is going, and system administration in general.

Laura Nolan — USENIX

Riffing off of a tweet by Charity Majors, this article explores the idea that moving faster can actually be safer, despite an urge one may feel to slow down.

Bruce Johnston

An extreme oversimplification of this incident would be: multiple engine failure on a plane subsequent to a maintenance error on all engines. This accident is cited as a reason to have separate mechanics work on each engine, in hopes of avoiding duplicated errors.

US National Transportation Safety Board (multiple authors)

[…] in order to ship new features and improvements faster while lowering the risk in our deployments, we have a simple but powerful tool: feature flags.

Alberto Gimeno — GitHub

This one blew my mind. By recording instruction execution traces in a ring buffer, they’re able to reconstruct enough information to step through the execution leading up to a crash — even though they weren’t running the application under a debugger!

Walter Erquinigo, David Carrillo-Cisneros, Alston Tang — Facebook

Automation is supposed to take some of the load off of the human operator, right? But in reality, humans need to build a mental model of what the automation is doing in order to use it safely and effectively.

Shem Malmquist — WIRED

Yet more proof that DNS behavior varies way more than is obvious at first glance. Who the heck thought longest common prefix matching was a good idea?

Charles Li — eBay

The application may log multiple lines during the lifecycle of a request. Stripe has found it invaluable to also log one final line with a fully summary of the request.

Brandur Leach — Stripe

This is a followup with more detail on the G-Suite outage I reported here last week. A database issue caused two separate outages.

Google

Really great advice about 3 common pitfalls in implementing SL*s.

Cortex

This research paper explores the marginal boundary, a set of conditions beyond which a system enters a different operating mode and an accident is much more likely. It discusses the concept of coupling between seemingly unrelated parts of the system and shows how economic incentives can push a system toward this boundary.

Dr. Richard Cook and Jens Rasmussen (Original paper)

Thai Wood — Resilience Roundup (summary)

This is an analysis of a recent BGP leak with a discussion about how the impact from such events can be mitigated through emerging best practices.

Alessandro Improta and Luca Sani — Catchpoint

How do you hand over ownership of a system, transferring enough knowledge that the new owners can maintain its availability and reliability successfully?

Aleksandra Gavrilovska — SoundCloud

Shopify works toward Black Friday / Cyber Monday all year long, through a combination of load testing, failure mode analysis, game days, and incident analysis.

Ryan McIlmoyl — Shopify