[frontend] List named blobs SQL

[snalli]

Fix list-named-blobs SQL to use smaller joins. This improves query performance by 1000x. Yes you read that right.

[SophieGuo410]

We should be careful about what we changed here.
I suggest we could have a new query and switch by config. so we can rollback easily if something is wrong.
Also we need to add integration test to make sure it cover every case.
For example,
if we have the delete_ts is null for older version and deleted_ts is not null for a new version, we will return the old version blobName.

[snalli]

We should be careful about what we changed here. I suggest we could have a new query and switch by config. so we can rollback easily if something is wrong. Also we need to add integration test to make sure it cover every case. For example, if we have the delete_ts is null for older version and deleted_ts is not null for a new version, we will return the old version blobName.

will add a test, the coverage seems limited, because the tests passed even though my prev query was not what is required.

[snalli]

We should be careful about what we changed here. I suggest we could have a new query and switch by config. so we can rollback easily if something is wrong. Also we need to add integration test to make sure it cover every case. For example, if we have the delete_ts is null for older version and deleted_ts is not null for a new version, we will return the old version blobName.

converted the query itself to a config to avoid a v2/v3 switch, as we far too many of those.

[SophieGuo410]

I feel like having the actual query as a config is not good. We want to make the config as simple as possible, besides, having a sql cmd in config instead of in the MySqlNamedBlobDb makes the code hard to track. if you think we have too many boolean configs, you can easily remove them after the query has been fully verified by actual traffic.

[SophieGuo410]

Can you please add some test like like

1. add one permanent version.
2. upsert the named blob and delete it.
   In this case, we should list nothing cause the latest version already been deleted.

[SophieGuo410]

I'm a little bit confused here.
You didn't change the LIST_ALL_QUERY_V2, why you can update the statement settings?

[SophieGuo410]

I think GROUP BY logic operates independently of the WHERE clause, this GROUP BY account_id, container_id, blob_name ensures that the aggregation (e.g., MAX(version)) respects the boundaries of each account and container. Again, we are not the expert of mysql queries, and you might want to add some test case like adding named blob for different account and container and test the list. Even though you use where to select the corresponding account and container, it might not give you the right max version based on my understanding.

[SophieGuo410]

I feel like this query is still pretty complex. I suggest we can think of some simpler queries which makes the code easy to read. Sometimes having two queries is also ok cause we are trying to avoid the complex sql queris which might missing edge cases. cc @justinlin-linkedin

[SophieGuo410]

Can we add some time like this:
blobName1 blobId1 version1
blobName1 blobId2 version2
so listing we should return blobName1 blobId1 version2 right?
but I think in some case we might return the blobName1 blobId2 version.
This is ok but we should not return the NamedBlobRecord with blobId info. Instead, we can only return the blobName, cause blobId1 and blobId2 all map to blobName1 and we don't need blobId for listing.