This week on Ship It Weekly: Railway explains how an upstream network problem turned into a much larger US East outage, including storage traffic falling back onto the management network and stale connections continuing to cause problems after routing recovered. Stripe shares how graph search and state machines helped cut database pager volume by about 30 percent. Kata Containers patches a critical guest-to-host escape, and DynamoDB adds native vector search.

The bigger theme this week is what happens after the obvious failure. Fixing the route does not necessarily clear the connections created while it was broken. Automating recovery does not have to mean handing an AI agent unrestricted production access. And stronger isolation does not eliminate the components that still cross the guest-host boundary.

In the lightn ...