jen20 / packer-ubuntu-zfs Goto Github PK

When the script it running and gets to the point where is it trying to copy /tmp/sources.list into the newly mounted disk, it fails and says the file is not there.

I verified myself while the script was running that at no point does that file exist.

The build script works great for building 18.04 after disabling ena_support but it's still using the default ubuntu packages for zfs 0.7.5 which are seriously outdates. The 0.7.5 branch is up to 0.7.13 and a few newer releases are out with significant performance upgrades on release 0.8.1 which is availible in by source as well as a few reliable PPAs.

Is there any way to update these templates and scripts to build rpool on 0.8.1?

If I try to do a build with the ena_support line still in there it fails and tells me it's not a valid parameter. However, if I remove that line from the .json file it seems to build fine. Is there a new name for it, or any way to add the enhanced network support back in?

Adding shell provisioner action to clean apt cache when building the ami.

Wonderful thing you've created here!
Something I've had to add was running 'growpart /dev/xvda 2' and 'zpool online -e rpool /dev/xvda2' to bootup, if I want it to automatically grow the root filesystem at boot.

Hey there, thanks for the scripts.

I struggled to get them to work initially, it seems that there are issues with the order in which required items are installed.

amazon-ebssurrogate: + apt-get install -y zfsutils-linux debootstrap gdisk

    amazon-ebssurrogate: Reading package lists... Done
    amazon-ebssurrogate: Building dependency tree
    amazon-ebssurrogate: Reading state information... Done
    amazon-ebssurrogate: Package debootstrap is not available, but is referred to by another package.
    amazon-ebssurrogate: This may mean that the package is missing, has been obsoleted, or
    amazon-ebssurrogate: is only available from another source
    amazon-ebssurrogate:
    amazon-ebssurrogate: E: Unable to locate package zfsutils-linux
    amazon-ebssurrogate: E: Package 'debootstrap' has no installation candidate
==> amazon-ebssurrogate: Terminating the source AWS instance...
==> amazon-ebssurrogate: Deleting temporary security group...
==> amazon-ebssurrogate: Deleting temporary keypair...
Build 'amazon-ebssurrogate' errored: Script exited with non-zero exit status: 100

==> Some builds didn't complete successfully and had errors:
--> amazon-ebssurrogate: Script exited with non-zero exit status: 100

==> Builds finished but no artifacts were created.

Changing the order to gdisk -> zfsutils-linux -> debootstrap seems to be working, though.

i got error

1594706836,,ui,message,    amazon-ebssurrogate.source: + sgdisk -Zg -n1:0:4095 -t1:EF02 -c1:GRUB -n2:0:0 -t2:BF01 -c2:ZFS /dev/xvdf
1594706836,,ui,message,    amazon-ebssurrogate.source: Problem opening /dev/xvdf for reading! Error is 2.
1594706836,,ui,message,    amazon-ebssurrogate.source: The specified file does not exist!
1594706836,,ui,message,    amazon-ebssurrogate.source: Problem opening '' for writing! Program will now terminate.

Again huge fan of this repo.

It is possible to run the ebssurrograte EC2 instance from the fifth gen nitro families c5*|m5*|r5*|i3*|z1d* by
adding block drive / partition sym-linking in surrogate-bootstrap.sh.

Similar to how the EBS mapping rules work.

To do it requires 3 steps:

1. symlink /dev/xvdf EBS volume to its NVMe device before partitioning the final EBS root volume
2. symlink /dev/xvdf2 EBS partition to its NVMe device before creating the ZFS pool
3. remove the symlinks when finished (optional)

...

# install nvme-cli

# NVMe EBS launch device mappings (symlinks): /dev/nvme*n* to /dev/xvd*
declare -A blkdev_mappings
for blkdev in $(nvme list | awk '/^\/dev/ { print $1 }'); do  # /dev/nvme*n*
    # Mapping info from disk headers
    header=$(nvme id-ctrl --raw-binary "${blkdev}" | cut -c3073-3104 | tr -s ' ' | sed 's/ $//g' | sed 's!/dev/!!')
    mapping="/dev/${header%%[0-9]}"  # normalize sda1 => sda

    # Create /dev/xvd* device symlink
    if [[ ! -z "$mapping" ]] && [[ -b "${blkdev}" ]] && [[ ! -L "${mapping}" ]]; then
        ln -s "$blkdev" "$mapping"
        
        blkdev_mappings["$blkdev"]="$mapping"
    fi
done

# Partition the new root EBS volume
sgdisk -Zg -n1:0:4095 -t1:EF02 -c1:GRUB -n2:0:0 -t2:BF01 -c2:ZFS /dev/xvdf

# NVMe EBS launch device partition mappings (symlinks): /dev/nvme*n*p* to /dev/xvd*[0-9]+
declare -A partdev_mappings
for blkdev in "${!blkdev_mappings[@]}"; do  # /dev/nvme*n*
    mapping="${blkdev_mappings[$blkdev]}"

    # Create /dev/xvd*[0-9]+ partition device symlink
    for partdev in "${blkdev}"p*; do
    	partnum=${partdev##*p}
        if [[ ! -L "${mapping}${partnum}" ]]; then
            ln -s "${blkdev}p${partnum}" "${mapping}${partnum}"

            partdev_mappings["${blkdev}p${partnum}"]="${mapping}${partnum}"
        fi
    done
done

...

# Reset device mappings
for dev_link in "${blkdev_mappings[@]}" "${partdev_mappings[@]}"; do
    if [[ -L "$dev_link" ]]; then
        rm -f "$dev_link"
    fi
done 

I can create a PR for this.

Huge fan of this repo.

It might be desirable to have "full" package parity with the default ext4 cloud-images.

The Install standard packages portion of chroot-bootstrap.sh is missing these packages from the standard ubuntu cloud-images AMIs:

ubuntu-server
acpid
ec2-hibinit-agent
ec2-instance-connect
hibagent
krb5-locales
ncurses-term
ssh-import-id
xauth

• https://cloud-images.ubuntu.com/focal/current/focal-server-cloudimg-amd64.manifest

I can create a PR to add these.

I was wondering if it was possible to create an encrypted instance with this setup, either with EC2 KMS, ZFS native encryption, or some self-decrypting LUKS setup like here. I haven't been able to achieve a working self-decrypting instance on AWS, but I've only tried LUKS so far, and unsure of how to do a KMS or ZFS-native setup here. I haven't been able to find any similar packer templates which accomplish encryption for Ubuntu.

Greeting, James! Thanks for the great article and hard work putting this together!

I was able to run the packer build for yakety just fine. However, when I specified an existing ssh key and tried to login with that with ubuntu@, no joy. (password prompt)

I even created a new key and tried that, no joy. (password prompt)

I wonder if you have any ideas whats going on here?

Let me know!

Oh, BTW.. I tested the xenial and that works just fine with ssh ubuntu@.

Thanks,
Jon.