Segment discovered the hard way that their move to a microservice architecture had brought far more problems than benefits. Here’s why they transitioned back and how they pulled it off. Awesome article!

Alexandra Noonan — Segment

Drawing on the work of Dr. David Woods and the rest of the SNAFU Catchers, this article discusses the concepts behind resiliency and how to measure and achieve it.

Beth Long — New Relic

Serverless is not the magical gateway to the land of NoOps. You still have to operate your system even if you’re not directly running the servers. This article does a great job of explaining why.

Bhanu Singh — Network World

New to me: Wireshark’s statistics view and how it can be useful.

Julia Evans

How do you define whether your system is available and healthy? This article uses an anology to medical health.

Claiming that our system is doing well means nothing if users can perceive an outage.

José Carlos Chávez — Typeform

These folks are experiencing mysterious latency with HTTP/2 traffic to their ALB and published this report on their investigation. There’s no happy ending here — ultimately they disabled HTTP/2 support. I hope they update if they do discover the culprit.

Peter Forsberg — ShopGun

I had some fun this week unearthing the cause for the chronic lockups in Rsyslog that we’ve experienced at work. I found the cause (overeager retries of socket writes) and put together a bug report and a pull request.

Full disclosure: Fastly, my employer, is mentioned.

I love science! Grab wrote a nifty tool to help them select cohorts of users and perform experiments on them.

Abeesh Thomas and Roman Atachiants — Grab

Titus is the container orchestration system that Netflix created and open sourced. Rather than building a new auto-scaling feature for Titus, they instead got Amazon to productize EC2’s auto-scaling engine as a generalized auto-scaling tool, which Netflix then integrated with Titus. Neat!

See Amazon’s Application Auto Scaling announcement, published this past week.

Andrew Leung, Amit Joshi, and the rest of the Titus team — Netflix

What do you do when your hosts have kernel crashes at random every day? It turns out that you don’t need to be a seasoned kernel programmer to find a solution.

Pavlos Parissis — Booking.com

This is my first introduction tcpconnect (part of BCC). Pretty nifty!

fREW Schmidt

At Facebook, […] It is simply too difficult to rewrite caching/admission/eviction policies and other manually tuned heuristics by hand. We have to fundamentally change how we think about software maintenance.

Vladimir Bychkovsky, Jim Cipar, Alvin Wen, Lili Hu, and Saurav Mohapatra — Facebook

A couple weeks back, I linked to a postmortem template. Here’s a gameday report template from the same author.

Michael Kehoe

I had a really hard time choosing whether to include this one. On the one hand, it’s a really interesting article about service discovery in franchises that has to work right every time. On the other hand, Chick-fil-A has a terrible track record on GLBT rights, and I can’t overlook that.

Ultimately, I’m choosing to link to this article for its educational content, but I urge you to join me as I continue to boycott Chick-fil-A.

Brian Chambers, Caleb Hurd, and Alex Crane — Chick-fil-A

At 9 years old, this may be the oldest article I’ve linked to, but it’s worth it. The analogy to a home mortage is spot on.

Eric Lee

Click through to read about an interesting monitoring challenge and an account of how they solved it. I appreciate the emphasis on the importance of educating engineers to spread the knowledge of how the new system works among more people.

Joy Zheng and Jeeyoung Kim — Plaid

Another chaos engineering introduction. Why should you read it? If nothing else, the architecture diagram with the skull and cobwebs on it is pretty great. It’s also well worth reading if you’re looking to create a chaos engineering game plan.

Benjamin Wilms — Codecentric

Sometimes, a reliability risk can come in the form of a bunch of angry customers.

Ben Kuchera — Ars Technica

Humor for SREs! This is the most hilarious thing I’ve read all week.

James Mickens — USENIX ;login:logout

This focuses on various ways that Linux systems can fail to boot.

Chris Siebenmann

A (raw) transcript of a chat about Bloomberg’s adoption of SRE practices. It might be worth dropping it in a text editor and removing all occurrences of the phrase “sort of”. The real meat is in the discussion of what Bloomberg has learned (text search: “lessons learned”) and how to sell SRE as necessary in a company (text search: “ROI”).

Alan Shimel — devops.com

Channels employs three time-honored techniques to deliver these messages at low latency: fan-out, sharding, and load balancing. Let’s look inside the box!

Jim Fisher — Pusher

An in-depth explanation of how consistent hashing works. Love the hand-drawn diagrams!

Srushtika Neelakantam — Ably

Have you ever needed to generate a random number in code? whether it’s for rolling a dice, or shuffling a set, this tweet thread is here for you! There’s no reason that it should be easy or obvious, very experienced programmers repeat common mistakes. I did, before I learned …

Not strictly SRE-related, but then again it’s by Colm MacCárthaigh, who is SRE-related.

Colm MacCárthaigh

What should you do if you blow your error budget? Depends on whether you leaked it like a dripping faucet or splurged it all on big outages. Either way, you’ll need to investigate and make a plan.

Adrian Hilton, Alec Warner and Alex Bramley — Google

I love the two-method approach: a simple migration path for users that aren’t active all the time, and a more careful (and more complex) path for very busy users.

Xiang Li and Thomas Georgiou — Facebook

If you haven’t implemented alerts on support page views yet, do it now!! and thank me later. Here’s a view of how our dashboard looked as of a few minutes ago – a clear demonstration of user impact that supplements existing monitors and alerts.…

Click through for the graph. Monitor status and support page views… do we actually need any other monitoring? Only half-kidding.

Sri Harsha Kalavala

Visa wrote a letter to the Chair of the Treasury Committee of the UK House of Commons, explaining their outage from a few weeks ago and answering the questions they posed. The good bits are in the first few pages, and the question answers mostly reiterate them. The last question about steps to prevent recurrence has some additional detail.

[…] a component within a switch in our primary data centre suffered a very rare partial failure which prevented the backup switch from activating.

Visa

This is really nifty!

The website has two sections: Country Statistics and Traffic Shifts.

Such an awesome idea:

@eanakashima: Alerting on spikes in status page views: so wrong, or so right?

Emily Nakashima

How (and why) should an SRE team communicate with Dev and the rest of the organization? I especially enjoy the section on how communicating outwardly helps SRE.

HostedGraphite

o11ycon has posted a Call for Failures:

Send us a slide or two, including a graph or other visual artifact of observability that represents the worst day of your (professional) life. Or a graph that drives home some important, deeply unexpected, or just plain interesting point about your systems.

o11ycon

There’s a great description of their current setup, but what really makes this article awesome is the explanation of what was wrong with their old system and why they replaced it.

Shlomi Noach — GitHub

Hilights of this article:

• description of the pros and cons of two techniques for automating database migrations
• a surprising number of instances of the word “tentacle”

Hen Peretz — BlazeMeter

Rather than firing the driver that caused a rear-end collision, this company looked deeper and found an underlying flaw in their procedures.

The organization had unknowingly created a system that was risk-promoting, rather than risk-averse.

Larry Boxman and Paul LeSage — Journal of Emergency Medical Services

Our friends in the GrabFood team now save up to 70% development time on creating a new service. We have also recorded improvements in stability and availability of our services.

Karen Kue and Michael Cartmell — Grab

Some tips on surviving peak traffic as we head into World Cup season. I like the discussion in #10 (load testing): accurately testing your CDN is all but impossible.

Hadar Weiss — Peer 5 (CDN)

This is a video recording of a talk by Charity Majors at Monkigras 2018. She has a lot of awesome stuff to say about making on-call enjoyable and owning your code, including this gem:

Babies, by the way, are engineered by evolution to be too cute for you to want to kill them. Your code is not.

Charity Majors — Honeycomb

A power disruption occurred at our service provider resulting in a number of instances going offline. Heroku databases running on these instances were impacted.

Presumably this was the us-east-1 power issue I reported on in Issue 124.

The first article in this new series is about the evolution of the Network Engineer into a Network Reliability Engineer. It’s part of the broader breakdown of silos with the goal of understanding holistic reliabilty.

Michael Kehoe

I hadn’t realized that GDPR has provisions related to site/service reliability.

Theresa Abbamondi — Netscout

To shamelessly steal a line from this recorded talk, it’s very rarely the right thing for your observability system’s scale to match that of the system it’s observing. To avoid that, you need to throw away some event data rather than storing and indexing everything. How do you do that while still achieving functioning observability?

Ben Hartshorne — Honeycomb

I’m looking forward to seeing where this article series goes. Database changes can be a huge reliability risk, and getting them right is critical.

Bob Walker — Octopus Deploy