Adding an Optional DeleteWithPrefix Method for State Stores to Allow Deletion of Actor State

[RyanLettieri]

Describe the proposal

This proposal stems from the following feature request inside dapr runimte: dapr/dapr#6992
After a discussion with various members of the project, the following solution was proposed: dapr/dapr#6992 (comment)

By adding in an optional interface called 'DeleteWithPrefix', we will allow users to delete an actor's state. The interface is optional since some databases will not be able to implement it as highlighted by ItalyPaleAle, an active contributor for dapr here: dapr/dapr#7223 (comment)

Looking at sqlite as an example, a second migration will be added in order to generate a virtual "prefix" column. This column will contain all the information from the actor state entry up to but not including the key. For example, if the key entry is as follows: app-id||actor-type||actor-id||key the new virtual prefix column will contain app-id||actor-type||actor-id

New request that would be included in state/requests.go:

type DeleteWithPrefixRequest struct {
	Prefix string `json:"prefix"`
}

New interface that would be included in state/store.go:

type DeleteWithPrefix interface {
	DeleteWithPrefix(ctx context.Context, req DeleteWithPrefixRequest) (DeleteWithPrefixResponse, error)
}

List of components that are compatible with this optional method:

• sqlite: DeleteWithPrefix support for SQLite #3265
• mysql
• Postgres
• sqlserver
• oracle

[berndverst]

Please include the actual interface method you propose adding. Without that this proposal is not meaningful.

[berndverst]

I want to know what the new request object will be called and what it will look like, and what the new method is called.

https://github.com/dapr/components-contrib/blob/main/state/store.go#L38
https://github.com/dapr/components-contrib/blob/main/state/requests.go#L43

[RyanLettieri]

@berndverst updated the proposal to include both suggestions

[berndverst]

@berndverst updated the proposal to include both suggestions

LGTM

And for say Postgres this will eventually be translated to DELETE FROM my_tbl WHERE ID LIKE '$1%' where $1 is the prefix? Is that the goal?

In other words delete every entry where the ID begins with prefix.

[ItalyPaleAle]

And for say Postgres this will eventually be translated to DELETE FROM my_tbl WHERE ID LIKE '$1%' where $1 is the prefix? Is that the goal?

The goal is that, and the query is "like" that (see what I did there? :) )

More of an implementation detail, but I've guided @RyanLettieri to avoid the use of LIKE because that normally cannot leverage indexes, so it resorts to a full-table scans. Instead, I've recommended something along the lines of creating a virtual column on the table called key_prefix (or similar) which is generated by taking the prefix from the key column. Because you can add an index to virtual columns, then that is a lot more efficient.

Also worth noting that if the key doesn't have a prefix (which is possible in Dapr, such as foo), then it should never be deleted by DeleteWithPrefix.

[github-actions]

This issue has been automatically marked as stale because it has not had activity in the last 30 days. It will be closed in the next 7 days unless it is tagged (pinned, good first issue, help wanted or triaged/resolved) or other activity occurs. Thank you for your contributions.