Problem Background

The data disk failed to mount, causing log files to be written to the root disk, leading to multiple disk alarms.

The kernel had not yet refreshed the partition table to generate the /dev/nvme1n1p1 device file when the mkfs command was executed, resulting in a failed formatting.

Solution

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12

...
yum install parted -y
mkdir -p /data
parted -s /dev/nvme1n1 -- mklabel gpt
parted -s /dev/nvme1n1 -- mkpart primary ext4 0% 100%
partprobe /dev/nvme1n1
udevadm settle
while [ ! -e /dev/nvme1n1p1 ]; do sleep 1; done
mkfs -t ext4 /dev/nvme1n1p1
mount /dev/nvme1n1p1 /data
echo '/dev/nvme1n1p1 /data ext4 defaults 0 0' >> /etc/fstab
...

Difference between 0 and 0% in parted

• 0: Indicates a mandatory requirement for the partition to start from the 0th byte/sector of the disk.

• 0%: Aligns automatically, starting the partition at the best alignment point. Parted automatically skips the reserved space at the beginning and typically sets the starting position at sector 2048.

partprobe /dev/nvme1n1

Notifies the kernel (Kernel) to immediately reread the partition table.

udevadm settle

Blocks (pauses) script execution until udev has processed all device events. The script can only continue after all device files have been created.

User Data Script

Location of user script: /var/lib/cloud/instance/user-data.txt

Location of logs: /var/log/cloud-init-output.log