Instance fails to boot after changing instance type (Xen ↔ Nitro) when hibernation is enabled

[831kirimi]

Issue Description

When hibernation is enabled on an EC2 instance, amazon-ec2-hibinit-agent sets kernel command line parameters resume and resume_offset. However, the resume parameter is set using device names (e.g.,
resume=/dev/nvme0n1p1, resume=/dev/xvda1). When changing instance types between different hypervisor types (Xen ↔ Nitro), the device naming convention changes, causing the instance to fail boot
with instance status check failures.

Root Cause

The issue occurs in the patch_grub_config function where the swap device is identified by device name:

https://github.com/aws/amazon-ec2-hibinit-agent/blob/master/agent/hibinit-agent#L131

grub_update_kernel = "grubby --update-kernel=ALL --args='no_console_suspend=1 " + \
                         "resume_offset={offset} resume={swap_device}'"
grub_update_kernel = grub_update_kernel.format(offset=offset, swap_device=swap_device)
check_call(grub_update_kernel, shell=True)

https://github.com/aws/amazon-ec2-hibinit-agent/blob/master/agent/hibinit-agent#L189

if config.grub_update:
    dev_str = find_device_for_file(SWAP_FILE)
    patch_grub_config(dev_str, offset)

https://github.com/aws/amazon-ec2-hibinit-agent/blob/master/agent/hibinit-agent#L210

def find_device_for_file(filename):
    # Find the mount point for the swap file ('df -P /swap')
    df_out = check_output(['df', '-P', filename]).decode(sys.getfilesystemencoding())
    dev_str = df_out.split("\n")[1].split()[0]
    return dev_str

Reproduction Steps

1. Launch hibernation-enabled instance with Xen-based instance type:

aws ec2 run-instances \
    --image-id ami-07faa35bbd2230d90 \
    --instance-type t2.micro \
    --key-name test \
    --security-group-ids sg-1234567890abcdef0 \
    --hibernation-options Configured=true \
    --block-device-mappings '[
        {
            "DeviceName": "/dev/xvda",
            "Ebs": {
                "VolumeSize": 30,
                "VolumeType": "gp3",
                "Encrypted": true
            }
        }
    ]'

2. Stop the instance:

aws ec2 stop-instances --instance-ids i-1234567890abcdef0

3. Change to Nitro-based instance type:

aws ec2 modify-instance-attribute --instance-id i-1234567890abcdef0 --instance-type Value=t3.micro

4. Start the instance:

aws ec2 start-instances --instance-ids i-1234567890abcdef0

5. Check instance status:

aws ec2 describe-instance-status --instance-ids i-1234567890abcdef0

Expected Result

Instance should boot successfully after instance type change.

Actual Result

Instance status check fails with "reachability: failed" status:

{
    "Details": [
        {
            "ImpairedSince": "2025-09-03T02:17:00+00:00",
            "Name": "reachability",
            "Status": "failed"
        }
    ],
    "Status": "impaired"
}

Proposed Solution

Use UUID or filesystem labels instead of device names for the resume parameter to ensure compatibility across different hypervisor types.