I used to be a die-hard advocate of containers, back when they were Solaris Zones. They could destroy the performance of Xen VMs (and I wrote posts explaining the technical reasons why.) But a lot of things changed with HW hypervisors, with the leaders nowadays being Amazon Nitro and Firecracker.
It's something the article doesn't address. Containers used to be a slam dunk, so I'm not sure I'd say that we were wrong _at the time_. But HW hypervisors evolved, so now I can see HW hypervisors like Nitro and Firecracker being compelling for some* workloads (e.g., to give the end user a dedicated kernel where they can do anything, including run any BPF program, run PGO kernels, etc.)
It's something the article doesn't address. Containers used to be a slam dunk, so I'm not sure I'd say that we were wrong _at the time_. But HW hypervisors evolved, so now I can see HW hypervisors like Nitro and Firecracker being compelling for some* workloads (e.g., to give the end user a dedicated kernel where they can do anything, including run any BPF program, run PGO kernels, etc.)
* I'm not sure yet whether it's "some" or "most".