What is rootDirectory for in remote storage configuration?

[lesomnus]

Assume following config:

{
	"storage": {
		"rootDirectory": "/var/lib/registry",
		"storageDriver": {
			"name": "s3",
			"region": "us-east-1",
			"bucket": "zot"
		}
	}
}

Why rootDirectory need to be specified?

• Does cache (from remote storage) are stored on that? If it is a cache storage, does that cache storage has a space limit? How much space should I prepare? Can I configure that?
• Is that for pull through cache registry? If it is a local storage for pull thorugh image, can I use features such as "commit" , "dedup" and "gc"? In the culuster configuration, the doc says "dedup" and "gc" should be disabled, is that right, or does clustering not imply a stateless? Also, can I configure the space limit of the directory?

[andaaron]

Hi @lesomnus, the rootDir needs to be specified in both cases, local storage and 3s storage.
In the local storage case it is a local folder under which images and additional files used internally by zot are stored.
In the s3 storage case it is a local folder where blobs are stored temporarily until the image is fully pushed (for multiple reasons, such as we don't stream multipart uploads directly to S3, also we make sure all blobs in the same image are pushed and valid), after which they are copied over to S3.
Note for using zot as a pull through cache, this will also be the place where blobs are temporarily kept after zot downloads them locally, before uploading them to the final storage location, be it local or S3.

For the reason above "rootDirectory" is mandatory for any zot configuration.
There is no zot mechanism to limit the space usage.

The there is a "cacheDriver" setting which is a separate mechanism from "rootDirectory", which is used to track duplicate blobs along with some other metadata. For AWS case we have a DynamoDB implementation. In case it is not specified, all data would be stored locally under the "rootDirectory" in BoltDB .
Given we don't have a 3rd alternative if DynamoDB is not available and BoltDB in local storage is not possible for multiple stateless instances, then you need to disable all feature that would need to use these database. Which means NO to "dedupe", "gc" and a bunch of other features which need the DB to function in order to make zot stateless.

[lesomnus]

Thank you for your swift response and clear explanation!