Redis Model Cache (PROJQUAY-788)

[alecmerdler]

Issue: https://issues.redhat.com/browse/PROJQUAY-788

Changelog: Add model cache implementation backed by Redis.

Docs: N/a

Testing: Add the following section to config.yaml:

DATA_MODEL_CACHE_CONFIG:
  engine: redis
  host: <redis-hostname>
  port: 6379

Details: Adds an implementation of the model cache which uses Redis. The goal is when deploying a cluster of Quay app containers, they will all use a shared Redis instance for caching requests (as opposed to a per-container memcache), which will reduce load on the database for expensive queries. Enhancements proposal.

NOTE: When using this model cache implementation, the Redis dependency changes from being low priority (effectively ephemeral) to being critical (and should be HA).
NOTE: The Quay container will still have memcached running even if not being used, unless we modify the container entrypoint to read the config.yaml.

Blocked by #745

[alecmerdler]

Rebased with master.

[kleesc]

When using this model cache implementation, the Redis dependency changes from being low priority (effectively ephemeral) to being critical (and should be HA).

It depends. Most deployment aren't large enough or don't generate enough traffic to cause a thundering herd issue, and clearing the cache wouldn't be an issue.

[kleesc]

@alecmerdler The current ticket linked is the wrong one. It should be PROJQUAY-788

[kleesc]

@alecmerdler Like you previously mentioned, I think we should make it so that two requests can't write to the same key at once (grab a Lock).

[alecmerdler]

@kleesc Ready for another review after being rebased with the GlobalLock fix.

[kleesc]

LGTM

[alecmerdler]

I'm seeing slow pushes (stuck in preparing state) when running this on my k8s cluster. Going to do a bit more testing to ensure it's unrelated to this code change.

[kleesc]

LGTM