Re: [RFC PATCH 1/2] sdk: support creation of container image

[Jan Kiszka <jan.kiszka@siemens.com>]

On 12.01.21 18:54, [ext] Silvano Cirujano Cuesta wrote:
> 
> On 12/01/2021 18:36, Henning Schild wrote:
>> Am Tue, 12 Jan 2021 11:33:37 +0100
>> schrieb "[ext] Silvano Cirujano Cuesta"
>> <silvano.cirujano-cuesta@siemens.com>:
>>
>>> Extend task "populate_sdk" to support the creation of a container
>>> image containing the SDK.
>>>
>>> Signed-off-by: Silvano Cirujano Cuesta
>>> <silvano.cirujano-cuesta@siemens.com> ---
>>>  meta/classes/image-sdk-extension.bbclass | 99
>>> ++++++++++++++++++++++-- 1 file changed, 92 insertions(+), 7
>>> deletions(-)
>>>
>>> diff --git a/meta/classes/image-sdk-extension.bbclass
>>> b/meta/classes/image-sdk-extension.bbclass index a8c708a..9317256
>>> 100644 --- a/meta/classes/image-sdk-extension.bbclass
>>> +++ b/meta/classes/image-sdk-extension.bbclass
>>> @@ -6,10 +6,77 @@
>>>  # This class extends the image.bbclass to supply the creation of a
>>> sdk 
>>>  SDK_INCLUDE_ISAR_APT ?= "0"
>>> +SDK_GENERATE_FORMATS = "${@d.getVar("SDK_FORMATS", "tar")}"
>>> +
>>> +sdk_tar() {
>>> +    # Copy mount_chroot.sh for convenience
>>> +    sudo cp ${SCRIPTSDIR}/mount_chroot.sh ${SDKCHROOT_DIR}
>>> +
>>> +    # Create SDK archive
>>> +    cd -P ${SDKCHROOT_DIR}/..
>>> +    sudo tar --transform="s|^rootfs|sdk-${DISTRO}-${DISTRO_ARCH}|" \
>>> +        -c rootfs | xz -T0 >
>>> ${DEPLOY_DIR_IMAGE}/sdk-${DISTRO}-${DISTRO_ARCH}.tar.xz +}
>>> +
>>> +sdk_container_images() {
>>> +    local cmd="/bin/dash"
>>> +    local empty_tag="empty"
>>> +    local full_tag="latest"
>>> +    local oci_img_dir="${WORKDIR}/oci-image"
>>> +    local sdk_container_formats="$1"
>>> +
>>> +    # prepare OCI container image skeleton
>>> +    sudo umoci init --layout "${oci_img_dir}"
>>> +    sudo umoci new --image "${oci_img_dir}:${empty_tag}"
>>> +    sudo umoci config --image "${oci_img_dir}:${empty_tag}" \
>>> +        --config.cmd="${cmd}"
>>> +    sudo umoci unpack --image "${oci_img_dir}:${empty_tag}" \
>>> +        "${oci_img_dir}_unpacked"
>>> +
>>> +    # add SDK root filesystem as the flesh of the skeleton
>>> +    sudo cp -a "${SDKCHROOT_DIR}"/* "${oci_img_dir}_unpacked/rootfs/"
>>> +
>>> +    # pack container image
>>> +    sudo umoci repack --image "${oci_img_dir}:${full_tag}" \
>>> +        "${oci_img_dir}_unpacked"
>>> +    sudo umoci remove --image "${oci_img_dir}:${empty_tag}"
>>> +    sudo rm -rf "${oci_img_dir}_unpacked"
>>> +
>>> +    # no root needed anymore
>>> +    sudo chown --recursive $(id -u):$(id -g) "${oci_img_dir}"
>>> +
>>> +    # convert the OCI container image to the desired format
>>> +    sdk_id="sdk-${DISTRO}-${DISTRO_ARCH}"
>>> +    image_name="isar-${sdk_id}"
>>> +    image_archive="${DEPLOY_DIR_IMAGE}/${sdk_id}-${sdk_format}.tar"
>>> +    for sdk_format in ${sdk_container_formats} ; do
>>> +        case "${sdk_format}" in
>>> +            "docker-archive" | "oci-archive")
>>> +                if [ "${sdk_format}" = "oci-archive" ] ; then
>>> +                    target="${sdk_format}:${image_archive}:latest"
>>> +                else
>>> +
>>> target="${sdk_format}:${image_archive}:${image_name}:latest"
>>> +                fi
>>> +                skopeo --insecure-policy copy \
>>> +                    "oci:${oci_img_dir}:${full_tag}" "${target}"
>>> +                xz -T0 "${image_archive}"
>>> +                ;;
>>> +            "oci")
>>> +                tar --create --xz --directory "${oci_img_dir}" \
>>> +                    --file "${image_archive}.xz" .
>>> +                ;;
>>> +            "docker-daemon" | "containers-storage")
>>> +                skopeo --insecure-policy copy \
>>> +                    "oci:${oci_img_dir}:${full_tag}" \
>>> +                    "${sdk_format}:${image_name}:latest"
>>> +                ;;
>> Jan proposed to make the SDK class wider than x86(64). How is this
>> going to affect docker? (i intentionally use "docker" as synonym for all
>> sorts of ...)
> I intentionally use container instead of Docker where it's not docker-only.
>>
>> The proposed changes should be tested for riscv and arm64 as well.
> 
> Do you mean hosts or targets? Assuming you're talking about RISC-V and ARM64 targets...
> 
> This functionality is merely packaging the cross-buildchain that up-to-now could be generated with "populate_sdk", whatever was supported should be supported by this patch.
> 
> Of course, the changes should be tested for different combinations. Feel free to test it for those architectures, as long as we are in the discussion round I'm fine with the "whatever was supported should be supported by this patch" that I've written above. I'd really test if only before contributing the final patches for integration.
> 

Agreed. I don't see any arch-specific blocker here. Both skopeo and
umoci are available for our "new" host arch, arm64. RISC-V is still no
official Debian target, thus may break for any reason on any day.

Jan

-- 
Siemens AG, T RDA IOT
Corporate Competence Center Embedded Linux