Every service needs a couple of big hammers that are easy to swing.

Jennifer Mace — O’Reilly and Google

Answer: automation. Lots of automation. And automation of the automation.

Fred Lin, Harish Dattatraya Dixit, and Sriram Sankar — Facebook

Oh, how quaint! This article was written back when people traveled for the holidays.

Ashley Roof — Transposit

Surprise! Fortunately, there are some ways to fix this limitation.

Heidi Howard, Ittai Abraham — Decentralized Thoughts

A common question when a company is implementing incident management is: why do we need this process?

It turns out that the easiest way to answer this question is to look at the world of unsuccessful incident management.

Kintaba

Whether you’re new to Just Culture or an old hand, there’s a lot of great detail in this article.

Tory Thompson — Firehouse

Not sold yet on full service ownership for development teams? This interview may help.

Vivian Chan — PagerDuty

While ostensibly about Jeli.io, this article makes a great case for why incident analysis is important in general and what kind of data we should be trying to gather.

John Allspaw — Adaptive Capacity Labs

A new feature roll-out resulted in impaired service for some customers.

The adaptive universe: where adaptations to challenges feed back and cause more challenges, requiring more adaptations.

Lorin Hochstein

Our first GraphQL release was twice as slow as our old REST API. Here’s how we fixed it.

Another great example of making a duplicate request to a new API in the background to test it before deploying it.

Michael P. Geraci — OkCupid

It’s really easy to get an “uptime” SLO wrong, and a lying SLO can give you a false sense of security.

Piyush Verma — Last9

I love this quote. I feel like this is the “root cause” of every incident:

As for the underlying cause of the incident (or the “root cause” if you insist on using such language), that has to be the fact that our assumptions as teams or individuals are ultimately formed by our past experiences.

Oliver Leaver-Smith — Sky Betting & Gaming

I really love the concept of requisite complexity. This article has me thinking about a big project I’m working on in a new light.

Fred Hebert

They expected to max out an integer primary key column sometime in 2021. Then the pandemic hit and their timetable suddenly accelerated along with their traffic.

Jeff Pollard — Strava

I shouldn’t enjoy reading these so much… got any of your own to share?

Dean Wilson

The idea of borrowing expertise makes me think of Bainbridge’s Ironies of Automation.

Mandi Walls — PagerDuty

Heroku’s report explains how their service was impacted as a result of the big Amazon Kinesis outage a couple weeks back.

Heroku

This primer focuses on ensuring that your SLOs actually match up with business objectives.

Irving Popovetsky — Honeycomb

This incident report from a September Datadog outage has an interesting tidbit aboiut scaling external incident response in tandem with internal.

Alexis Lê-Quôc — Datadog

This is Google’s write-up for an interesting issue that involved repeated re-sending of invitations to edit a Google Drive document.

Google

I basically want to immediately absorb any article with this title, unless it’s just clickbait spam. This one definitely isn’t.

Ronak Nathani

Lots of juicy details in this one about the difficulty Slack has had in scaling their DB layer and how Vitess solved their problems.

Arka Ganguli, Guido Iaquinti, Maggie Zhou, and Rafael Chacón — Slack

Hitting file descriptor limits is such an annoying kind of outage. Some good tips here, clearly coming from hard-won experience.

Utsav Shah

They used two providers synced with OctoDNS.

Ryan Timken and Kiran Naidoo — Cloudflare

This is all about understanding the whole system (people and technology) and building learning, rather than finding a superficial “root cause”.

Piyush Verma — Last9

DNS-based load balancing is a nice simple solution, but unfortunately it doesn’t work well in certain circumstances. Read to find out how Algolia evolved their load balancing system in response.

Paul Berthaux — Algolia

We use percentiles all the time, so it’s really important to actually understand what they say (and what they don’t).

Piyush Verma — Last9

Thanks to An anonymous reader for this one.

The author started out as an embedded systems developer and moved into SRE. Here’s what they learned.

Eric Uriostigue — effx

Some great tips here. It’s hard to sound sincere in a public incident report, especially if you post a lot of them.

Adam Fowler

In this blog, we discuss how we built Fare Storage, Grab’s single source of truth fare data store, and how we overcame the challenges to make it more reliable and scalable to support our expanding features.

Sourabh Suman — Grab

This article covers Netflix’s gnmi-gateway, their open source tool for collecting metrics from network devices in a highly available and fault-tolerant manner.

Colin McIntosh and Michael Costello — Netflix

This year, re:Invent is online only, so you still have a chance to attend if you’re interested.

Ana M Medina — Gremlin

Cloudflare’s API service was impaired early this month. This is their incident report that describes a grey failure in a switch and downstream impact to etcd and their database system.

Tom Lianza and Chris Snook — Cloudflare

A Certificate Transparency (CT) log failed, resulting in its permanent retirement. The incident involved unintended effects from load testing being performed in a staging environment. I have a huge amount of admiration and respect for the transparency of certification authorities (CAs) when things go wrong.

Trust Asia

I like the idea that adding the ability to fail over to your system makes it much more complicated and thus more likely to fail.

Andre Newman — Gremlin

This one introduces some interesting concepts: the error kernel and property testing.

Kenneth Cross — HelloSign

[…] to be resilient, we must test everything, which consumes time that we don’t spend innovating. A good trade-off is to test in production.

Xavier Grand — Algolia

More useful tips as you develop your post-incident analysis process. I like their definition of “blameless”.

Zachary Flower — Splunk

Exactly once delivery is hard to implement and requires explicit coordination at all levels, including the client. Ably explains how their flavor works.

Paddy Byers — Ably

The most effective (if scary) way to understand how your stateless service operates under load

Utsav Shah — Software at Scale

Some good tips here — and a reminder that we may see even more traffic than normal due to social distancing.