common: abort load on ENOSPC and output clear error

[zvenigorodsky]

Fix: containers/podman#26197

When loading images, exit archive checks when receiving ENOSPC error, and display a clearer exit message for the user.

tests for podman at: containers/podman#27374

[podmanbot]

✅ A new PR has been created in buildah to vendor these changes: containers/buildah#6452

[zvenigorodsky]

Hey, I saw the issue containers/podman#26197 is stale, and decided to look into it.

Looked at zhengkezhou1's PR and researched about fmt.Errorf.

It does implement the unwrap method on the error (https://pkg.go.dev/fmt#Errorf) , and errors.Is does call the unwrap on fmt.Errorf (https://pkg.go.dev/errors#Is)

[zvenigorodsky]

Tested this by creating a mounted filesystem with minimal storage and pointed store.graphroot to said filesystem,
works as intended

[Luap99]

Thanks for the PR.

As I noted on containers/common#2446 (review) I don't think this can work because the chroot extraction and thus ENOSPC can happen in a subprocess so the error is just a string not the actual ENOSPC object so this cannot match I believe.

Can you provide the full reproducer you used to test this patch?

[mtrmac]

can happen in a subprocess

(Note that this is platform-specific; testing on macOS or Windows would not show this.)

[zvenigorodsky]

I forgot about other platforms, I tested this on a fedora 41 by creating a filesystem with minimal storage and pointing graphroot to the filesystem with a config file.

Here's a bash script that basically reproduces the setting:

#!/bin/bash

USER=$(whoami)

dd if=/dev/zero of=temp_storage bs=1M count=128
mkfs -t ext4 temp_storage
mkdir temp_storage_fs

sudo mount temp_storage temp_storage_fs
sudo chown $USER:$USER temp_storage_fs

cat > custom.conf <<EOF
[storage]
graphroot="$PWD/temp_storage_fs"
EOF

go mod edit -replace go.podman.io/common=github.com/zvenigorodsky/container-libs/common@main

go mod tidy

go mod vendor

make bin/podman 

# You can replace with any image that will be larger than 128Mb
HEAVY_IMAGE=quay.io/openshift-release-dev/ocp-release:4.20.2-ppc64le

./bin/podman pull $HEAVY_IMAGE 

./bin/podman save $HEAVY_IMAGE -o image.tar

export CONTAINERS_STORAGE_CONF="$PWD/custom.conf" 

./bin/podman load -i image.tar

[mtrmac]

That’s just guessing, and discarding the original cause.

[mtrmac]

This is still discarding the original error value [maybe we would want to modify the text, but it is still valuable to retain the regional error so that errors.As works higher in the stack], and not handling chrootarchive; both from the very first #419 (comment) .

[Luap99]

Note that I couldn't get go mod edit -replace go.podman.io/containers/common=github.com/zvenigorodsky/container-libs/common to work, it looked like container-libs/common is still not a drop-in replacement to containers/common (correct me if I'm wrong).

It works fine, after all we already use the monorepo in podman. It is your command which is wrong. The module name is go.podman.io/common (no containers)

https://github.com/containers/container-libs/blob/main/CONTRIBUTING_GO.md#testing-changes-in-a-dependent-repository

[zvenigorodsky]

... and not handling chrootarchive; ...

Could you guide me in how can I test this for other platforms?

[baude]

any chance you would want to push this through or would you like someone to take this over?

[zvenigorodsky]

any chance you would want to push this through or would you like someone to take this over?

I would like to know if there is a way for me to test this on other platforms without replicating the environment from scratch on virtual machines.

If that's the only way to test this for other platforms, then yes I would like for someone else to take over

[mtrmac]

I suppose, for testing purposes, adding an

if err != nil {
   err = errors.New(err.Error())
}

to the stub chrootarchive implementations would type-erase the error, matching the Linux behavior, and allow testing things. That would, at least, demonstrate why the proposed PR is insufficient.

But, given that I expect that a real fix would involve detecting this error class in the chrootarchive subprocess, and passing that information over the process boundary, I think running/developing this on Linux or a Linux VM is necessary.

[zvenigorodsky]

I am running a Linux machine, I'll be happy to try again if I may

[packit-as-a-service]

Packit jobs failed. @containers/packit-build please check.

[packit-as-a-service]

Packit jobs failed. @containers/packit-build please check.