Unable to use any tools under /sbin in cloud-init script

Question

I have a cloud init script which attaches a volume, creates a partition on said volume and then mounts that partition. For that I use parted and mkfs as usual. The image is based on Amazon Linux 2023 and it's AMI id is ami-005ad02a0f1729234 (questdb version 8.2.3).

The script is like this (stripped to essential part):

```
#!/bin/bash
TOKEN=`curl -X PUT "http://169.254.169.254/latest/api/token" -H "X-aws-ec2-metadata-token-ttl-seconds: 21600"`
INSTANCE_ID=$(curl -H"X-aws-ec2-metadata-token: $TOKEN" http://169.254.169.254/latest/meta-data/instance-id)
aws ec2 attach-volume --device /dev/xvdc --instance-id $INSTANCE_ID --volume-id $VOLUME_ID
echo "volume attached, waiting for device to become available"
while ! lsblk | grep xvdc; do
    sleep 0.1
done
parted -s -a opt /dev/xvdc mklabel gpt
parted -s -a opt /dev/xvdc mkpart primary ext4 0% 100%
mkfs -t ext4 /dev/xvdc1
mount /dev/xvdc1 /mnt
```

* The above will result in `parted: command not found`.
* If I change to `/sbin/parted`, the mkfs is not found
* If I change mkfs to `/sbin/mkfs` then I get `.../sbin/mkfs.ext4: no such file or directory`
* If I edit path at the top of the scrip `export PATH="/sbin/:/usr/sbin/:$PATH"` it has no effect
* If I call `which parted` I get `no parted in ((null))`
* If I run with `#!/usr/bin/env bash` it has no effect

The only workaround is to switch paths to `/usr/sbin/` and not use `mkfs` but use the filesystem-specific program directly. I.e. this works.

```
#!/bin/bash
TOKEN=`curl -X PUT "http://169.254.169.254/latest/api/token" -H "X-aws-ec2-metadata-token-ttl-seconds: 21600"`
INSTANCE_ID=$(curl -H"X-aws-ec2-metadata-token: $TOKEN" http://169.254.169.254/latest/meta-data/instance-id)
aws ec2 attach-volume --device /dev/xvdc --instance-id $INSTANCE_ID --volume-id $VOLUME_ID
echo "volume attached, waiting for device to become available"
while ! lsblk | grep xvdc; do
    sleep 0.1
done
/usr/sbin/parted -s -a opt /dev/xvdc mklabel gpt
/usr/sbin/parted -s -a opt /dev/xvdc mkpart primary ext4 0% 100%
/usr/sbin/mkfs.ext4 /dev/xvdc1
mount /dev/xvdc1 /mnt
```

I am creating the instance with the following CDK code:

```
    const instance = new ec2.Instance(this, "Instance", {
      instanceType: instanceType,
      machineImage: ec2.MachineImage.lookup({
        name: `questdb-${cfg.questdbVersion}*`,
      }),
      vpc: vpc,
      availabilityZone: vpc.availabilityZones[0],
      vpcSubnets: {
        subnetType: ec2.SubnetType.PRIVATE_WITH_EGRESS,
      },
      propagateTagsToVolumeOnCreation: true,
      securityGroup,
    });
    const volume = new ec2.Volume(
      this,
      `QuestdbVolume${hashFromString(`${cfg.questdbVersion}${cfg.snapshotId ?? ""}`, 4)}`,
      {
        encrypted: true,
        availabilityZone: vpc.availabilityZones[0],
        snapshotId: cfg.snapshotId,
        size: cdk.Size.gibibytes(cfg.dbSize),
        iops: 16000,
        throughput: 1000,
        volumeType: ec2.EbsDeviceVolumeType.GENERAL_PURPOSE_SSD_GP3,
      },
    );
    instance.applyCloudFormationInit(
      ec2.CloudFormationInit.fromElements(
        ec2.InitFile.fromFileInline("/opt/init/init.sh", path.join(__dirname, "init", "init.sh"), {
          mode: "000755",
        }),
        ec2.InitCommand.shellCommand("/opt/init/init.sh", {
          env: {
            VOLUME_ID: volume.volumeId
          },
        }),
      ),
    );
```

I found absolutely nothing online about this. Is this something specific to this QuestDB image? Why is documentation not mentioning anything about that?

Varunprakash Shanmugam · Accepted Answer

On Amazon Linux 2023 and specifically in your QuestDB-based AMI, utilities like `parted`, `mkfs.ext4`, etc., are located in `/usr/sbin` and `/sbin`, but those directories may not be in `PATH` during execution unless explicitly set.

However, simply setting `export PATH=...` in the script often has no effect when using `InitCommand`, because each command runs in a fresh shell.

**Solution: Use full paths for commands**

To avoid issues, use the full absolute paths for the tools. Here’s a cleaned-up version of your script that works reliably:

```bash
#!/bin/bash

TOKEN=$(curl -X PUT "http://169.254.169.254/latest/api/token" \
  -H "X-aws-ec2-metadata-token-ttl-seconds: 21600")

INSTANCE_ID=$(curl -H "X-aws-ec2-metadata-token: $TOKEN" \
  http://169.254.169.254/latest/meta-data/instance-id)

aws ec2 attach-volume --device /dev/xvdc --instance-id "$INSTANCE_ID" --volume-id "$VOLUME_ID"

echo "volume attached, waiting for device to become available"

while ! lsblk | grep xvdc; do
    sleep 0.1
done

/usr/sbin/parted -s -a opt /dev/xvdc mklabel gpt
/usr/sbin/parted -s -a opt /dev/xvdc mkpart primary ext4 0% 100%
/usr/sbin/mkfs.ext4 /dev/xvdc1

mkdir -p /mnt
mount /dev/xvdc1 /mnt
```

**Alternative: Use a wrapper script**

If you want to avoid using full paths everywhere and ensure PATH is available, wrap your logic in a script and invoke it using `bash -l` (login shell), which sources `/etc/profile` and sets up the environment.

Modify `InitCommand` like this:

```ts
ec2.InitCommand.shellCommand("bash -l -c '/opt/init/init.sh'", {
  env: {
    VOLUME_ID: volume.volumeId,
  },
}),
```

This ensures the system paths from login shells are respected (`/sbin`, `/usr/sbin`, etc.).

Unable to use any tools under /sbin in cloud-init script

Relevant content