Update the updater.rs code.

[MathieuDutSik]

Motivation

The updater.rs is currently using the blob_last_used_by with the error case returning the missing entries which is a bd design:

• It might get you a wrong index if the error is not not_found.
• It does not group the data.

Proposal

The following was done:

• Introduce the missing_blob_states that provides what we are after.
• Remove the blobs_last_used_by that was previously introduced, but keep it in the db_storage.rs.
• The replacement of stream and the attention to only getting whether a blob_state is present or not allows for grouping the operation efficiently.

Possible criticism of the design:

• We could have kept the blobs_last_used_by but there was no use case and in particular, we could not use it for getting the missed blob ids.
• The API is left in the db_storage.rs. Possibly we could remove it as well as other unused APIs.
• The missing_blob_ids does a single operation. We could split it into blocks and then use a stream in a future PR. However, the size is quite small, only BlobIds.

There is no remaining read_certificate(s) that is done for extracting only small data.

Test Plan

The CI

Release Plan

This changes the binaries but does not change the storage. So, a switch

Links

• reviewer checklist

[afck]

Good catch! We actually are only interested in the missing blobs in this case.

[ndr-ds]

[Re: lines +385 to +401]

Like @afck mentioned, `missing_blob_states` should be semantically equivalent to this one (we'll only have a state in storage if we have a blob also). So we can probably just use this existing one instead and delete `missing_blob_states`

See this comment inline on Graphite.

[MathieuDutSik]

Ok, thank you.