The Amazon S3 Outage - When the Internet's Hard Drive Fails …
March 06, 2017

Denis Goodwin
SmartBear

Share this

Last week, SmartBear observed a sudden and protracted 5X increase in web page timeout errors associated with the failure of Amazon's S3 cloud-based storage service. Looking a bit more closely at our data, we dug up a few more interesting angles on the impact of the failure.

Event Timeline

The issue hit suddenly – we saw an immediate spike in errors at 12:35 p.m. EST, and by 12:45 p.m. EST, error rates were 5X normal. For some specific types of timeout errors, the spike was more than 10X normal. At 3:30 p.m. EST, the error rate began dropping and by 3:50 p.m. EST, rates had returned to normal.


Web vs. API

The issue hit web pages hard, while API monitors were not noticeably impacted by this outage. Web pages and web apps often utilize content storage hosted by cloud services such as Amazon S3.

Common failure scenarios on Tuesday included page elements failing to load, which could cause either the whole web page to time out or specific content on a page might not render. Depending on the design of a given page, this partial content failure could be relatively minor or it could render a critical web journey non-functional. File uploads and downloads that rely on S3 storage endpoints were particularly hard hit.

In order to get a complete picture of application health, it's necessary to monitor your real user's journey through the application. The monitored user journeys that depended heavily on content hosted in S3 failed. Those that didn't have that dependency continued functioning. I personally experienced this with Slack – I was able to use the app, however files could not be uploaded presumably because these files are stored by Slack using S3 as the storage mechanism.

While far less pronounced than the spike in errors, some response time degradation was observed in API monitors that continued running successfully. Given that the issue affected Amazon's storage services rather than their hosting services for applications, this makes sense.

Geographic Impact

The issue was more acutely felt in the United States, but we observed impacts all over the globe. The spike in page errors was seen on websites dependent on Amazon S3, many of which are U.S.-hosted websites that are likely monitored from U.S. locations. Unsurprisingly, error counts spiked by as much as 25X in some U.S. monitoring locations. While not as significant as the U.S. locations, timeout and page error increases were also observed from Canada, Europe and Asia.

Takeaways

Much of the web is built on the backs of cloud providers. Most of the time, these cloud services provide a great user experience. Amazon will learn from the root cause of this issue and likely emerge from this outage more resilient than ever. It's impossible to control all aspects of these shared services – but here are three steps to take that are in your control.

1. Identify your business critical applications

2. Proactively monitor user journeys on these applications

3. Don't rely on your third party provider to tell you when it is down

It is key to utilize independent monitoring services to ensure your applications are up, functioning correctly and fast. Furthermore, missing content can be catastrophic or merely inconvenient to a critical user journey – it's important that your monitoring tool can be configured to know the difference.

Denis Goodwin is Director of Product Management, AlertSite, SmartBear Software.

Share this

The Latest

June 15, 2018

When it comes to their own companies, 50% of IT stakeholders think they are leaders and will disrupt, while 50% feel they are behind and will be disrupted by the competition in 2018, according to a new survey of IT stakeholders from Alfresco Software and Dimensional Research. The report, Digital Disruption: Disrupt or Be Disrupted, is a wake-up call for the C-suite ...

June 14, 2018

If you are like most IT professionals, which I am sure you are, you are dealing with a lot issues. Typical issues include ...

June 13, 2018

The importance of artificial intelligence and machine learning for customer insight, product support, operational efficiency, and capacity planning are well-established, however, the benefits of monitoring data in those use cases is still evolving. Three main factors obscuring the benefits of data monitoring are the infinite volume of data, its diversity, and inconsistency ...

June 11, 2018

Imagine this: after a fantastic night's sleep, you walk into the office ready to attack the day. You sit down at your desk ready to go, and your computer starts acting up. You call the help desk, but all IT can do is create a ticket for you and transfer it to another team to help you as soon as possible ...

June 08, 2018

As many IT workers develop greater technology skills and apply them to advance their careers, many digital workers in non-IT departments believe their CIO is out of touch with their technology needs. A Gartner, Inc. survey found that less than 50 percent of workers (both IT and non-IT) believe their CIOs are aware of digital technology problems that affect them ...

June 07, 2018

CIOs of 73% of organizations say the need for speed in digital innovation is putting customer experience at risk, according to an independent global survey of 800 CIOs commissioned by Dynatrace ...

June 05, 2018

Digital Transformation requires more than just the latest technology, it's a mindset that iterative change is on the way and should be embraced. This also requires us to factor in the people and process parts of the equation and find ways to measure the end-user-experience (EUE). One way to do this is to sponsor an Application Performance Monitoring (APM) initiative that can provide visibility to the business, help communicate the progress, and highlight the impacts to the organization. ...

June 01, 2018

While DevSecOps, much like DevOps itself, is more about changing IT culture than employing certain types of technology, some tools can be an important support. To find out what the right tools are, DEVOPSdigest asked experts from across the IT industry for their opinions on the top tools to support DevSecOps ...

May 31, 2018

Availability has cemented itself as one of the most essential elements of any successful data center. However today, what many organizations are beginning to realize — sometimes the hard way — is that traditional methodologies and technologies for high availability (HA) have limits. What's needed instead is a new approach that enables the dynamic transfer of workloads in IT environments based on optimizing the particular job at hand ...

May 30, 2018

Self-driving cars, integrated toys, smart home appliances, and even critical infrastructure have all become part of the ecosystem of Internet of Things (IoT) devices, which begs the concerning question, "How will network administrators process all the data generated?" ...