This is temporary anyway, but it will allow us to finish the backfill without having to worry about newer tags being displayed with just their V1IDs
Fixes https://jira.coreos.com/browse/QUAY-1353
Some customers are hitting this endpoint rapidly for repositories with many, many tags. This change drops the unnecessary joins, which should reduce database load somewhat.
We were occasionally trying to compute schema 2 version 1 signatures on the *unicode* representation, which was failing the signature check. This PR adds a new wrapper type called `Bytes`, which all manifests must take in, and which handles the unicodes vs encoded utf-8 stuff in a central location. This PR also adds a test for the manifest that was breaking in production.
* endpoint/api/repository: limit the number of tags returned
- Limit the number of tags returned by /api/v1/repository/<ns:repo> to 500.
- Uses the tag history endpoint instead, with an active tag filte.
- Update UI to use tag history endpoint instead.
After discussion, we decided the best solution for the missing content checksum problem was to lookup the proper blobs in the repository and, if not present, mark the manifest as broken, as this would reflect the actual issue the user faces if they pull the repository tag today via V2
The tag manifest creation operation is now quite a bit heavier (because it is populating a bunch of new-model tables as well), and it is not necessary for the transaction to include this operation, because tags are valid without manifests currently.
The ALTER TABLE operations previously used were causing the DB to die when run on the production TagManifest table which has 7 million rows. We instead now use new mapping tables, which is less nice, but these are temporary anyway, so hopefully we only have to deal with their ugliness for a short duration.
This change also starts passing in the manifest interface, rather than the raw data, to the model for writing.
Note that this change does *not* backfill the existing rows in to the new tables; that will occur in a followup PR. The new columns in `tagmanifest` and `tagmanifestlabel` will be used to track the backfill, as it will occur in a worker.
Instead of 41 queries now for the simple manifest, we are down to 14.
The biggest changes:
- Only synthesize the V1 image rows if we haven't already found them in the database
- Thread the repository object through to the other model method calls, and use it instead of loading again and again
Will let the users know they can recover the tag via time machine
Note: This was tested with the Docker protocol, but the new error code is *technically* out of spec; we should make sure its okay.
