io_uring regression in 6.1 kernel

[arnaldo2792]

What happened?

The latest Linux 6.1 kernel introduced a regression in applications using the io_uring syscalls, which could result in processes hanging and not being killed. In particular, nodejs applications seem susceptible to this regression (see amazonlinux/amazon-linux-2023#840 for details).

Affected versions

Bottlerocket 1.27.1 variants using the 6.1 kernel:

• aws-k8s-1.28*
• aws-k8s-1.29*
• aws-k8s-1.30*
• aws-k8s-1.31*
• vmware-k8s-1.28*
• vmware-k8s-1.29*
• vmware-k8s-1.30*
• vmware-k8s-1.31*

[arnaldo2792]

Workaround for nodejs applications

Set the UV_USE_IO_URING=0 environment variable in your affected workloads:

apiVersion: apps/v1
kind: DaemonSet
metadata:
  name: <>
spec:
  selector:
    matchLabels:
      name: <>
  template:
    metadata:
      labels:
        name: <>
    spec:
      containers:
      - name: <>
        image: <>:<>
        #...
        env:
          - name: UV_USE_IO_URING
            value: '0'

[larhauga]

Any updates on when to expect bottlerocket 1.27.2 or 1.28.0 with this fix?