Linus Tech Tips Discussion

d there was one critical component that was inadvertently dependent on a single region, but there are also plenty of others who didn't think about that (Amazon generally only encourages spreading across availability zones, which are part of a region, but that would not have helped here). I imagine there will also be a few companies that decide that the cloud isn't worth it after this, and go back to self-hosting, but I also suspect there will be a lot fewer of those companies than some people would like to think - the cloud does have downsides, but it also does bring advantages too, and those advantages are particularly valuable to big companies (changing capex into opex, and requiring less in-house staff).   On the root cause of this issue, Amazon have already listed in the article some of the changes that they will be making to prevent this issue, and others like it, from happening again. However, in an organization as large as AWS, I think these types of outages are going to be inevitable - it's very difficult to eliminate every bug, and once there is a bug it's very difficult to make sure that every service is resilient to extended failures of its dependencies. I do think there are some useful lessons for both Amazon and other companies to learn from this, about avoiding cascading failures, but because these issues are rare, it's too easy to forget it somewhere, and you only need one part of the chain to be impacted for the outage to be widespread and severe.   Sources https://aws.amazon.com/message/101925/