[Bug] FS Metadata Duplication

[turan18]

Description

When converting TOC to metadata, we create a root FS bucket for every TOC, where each FS bucket is identified by a random ID.

soci-snapshotter/metadata/reader.go

Lines 120 to 139 in 8752222

	func (r *reader) init(toc ztoc.TOC, rOpts Options) (retErr error) {
	// Initialize root node
	var ok bool
	for i := 0; i < 100; i++ {
	fsID := xid.New().String()
	if err := r.initRootNode(fsID); err != nil {
	if errors.Is(err, bolt.ErrBucketExists) {
	continue // try with another id
	}
	return fmt.Errorf("failed to initialize root node %q: %w", fsID, err)
	}
	ok = true
	break
	}
	if !ok {
	return fmt.Errorf("failed to get a unique id for metadata reader")
	}
	return r.initNodes(toc)
	}

This means that we create a new FS tree every time we create a reader for a TOC, regardless of whether it exists in the DB. This causes a-lot of duplication and eats up a-lot of disk space on the host machine (eg: when restoring the snapshotter).

Steps to reproduce the bug

1. Pull an image

nerdctl pull --snapshotter soci public.ecr.aws/soci-workshop-examples/rabbitmq:latest

2. Check the size of metadata DB

ls -l --block-size=MB metadata.db

-rw------- 1 root root 17MB Dec 11 19:28 metadata.db

3. Kill and restart the snapshotter

systemctl kill soci-snapshotter
systemctl restart soci-snapshotter

4. Re-pull the image

nerdctl pull --snapshotter soci public.ecr.aws/soci-workshop-examples/rabbitmq:latest

5. Check the size of metadata DB

ls -l --block-size=MB metadata.db

-rw------- 1 root root 23MB Dec 11 19:33 metadata.db

Note: I was originally expecting the size to double, there may be some fancy bbolt storage optimization going on here.

Describe the results you expected

Ideally, we shouldn't write a TOC to the DB if it already exists. One possible solution is to have the FS bucket's identified by the hash of the TOC/zTOC instead of using a random id.

Host information

1. OS:
2. Snapshotter Version:
3. Containerd Version:

Any additional context or information about the bug

No response