Netflix, Tinder and other major websites were affected for a time Sunday by glitches in Amazon Web Services’ Northern Virginia facility, offering a cautionary lesson to other companies that rely on the cloud service for mission-critical capabilities.
The problem manifested itself primarily in the form of higher-than-normal error rates. Sites affected reportedly also included IMDb and Amazon’s Instant Video and Books websites.
At the heart of the snafu were issues with AWS’s DynamoDB database, but it spread to include other services such as EC2, the mobile-focused Cognito service and the CloudWatch monitoring service, according to the AWS Service Health Dashboard.
“The root cause began with a portion of our metadata service within DynamoDB,” AWS explained in a dashboard update posted at 4:52 a.m. PDT on Sunday. “This is an internal sub-service which manages table and partition information. Our recovery efforts are now focused on restoring metadata operations. We will be throttling APIs as we work on recovery.”
After beginning at 3:00 a.m. PDT Sunday, the DynamoDB issues were fixed by 8:15 a.m. All the other services were restored by 11.05 a.m.
“This really shouldn’t be happening,” said Rob Enderle, principal analyst with Enderle Group. “A service that is sold for mission-critical systems should have massive redundancies, and there should be isolation between different customers’ implementations so a failure on one shouldn’t bring everyone down.”
If similar incidents happen in the future, AWS could start to lose customers, Enderle said.
It’s “a cautionary tale for any AWS customer,” he said. “In the end, Amazon does not have adequate failover protection, which means its customers need to make sure they do.”
Netflix, in fact, apparently experienced minimal disruption because of its own redundancy approach.
“We were able to quickly redirect traffic from the impacted AWS region to one that was fully operational,” the company said via email.
Other Amazon customers running mission-critical systems on AWS would do well to emulate Netflix’s approach, Enderle suggested.
In the meantime, the event could benefit IBM, which “has a far more robust offering in SoftLayer,” as well as firms such as BMC that incorporate AWS and have strong failover capability, Enderle said.
Of course, virtually any outage is a significant one for a cloud provider given the heavy emphasis customers place on uptime, said Stephen O’Grady, co-founder and principal analyst with RedMonk.
“Undoubtedly AWS will be having ‘less than fun’ with customers today,” he said.
That said, however, “all providers have outages,” O’Grady noted, “and thus far they have not appeared to have any lasting impact on the trajectory of businesses like Amazon’s.”
Indeed, “the fix was applied quickly, AWS owned it and recovery started almost immediately,” agreed Dave Bartoletti, a principal analyst with Forrester. “In my experience, AWS can handle one or two of these a year without significantly scaring customers.”
More than anything, he added, “it’s a wake-up call to design your cloud apps for failure.”
An earlier version of this story misstated in the fifth paragraph the times at which the affected services were restored. The DynamoDB issues were fixed by 8:15 a.m. and the other services were restored by 11.05 a.m.