Refactoring Scheduler for Pluggable Backends

[Mierdin]

The existing Antidote API is fairly abstract - primitives like lessons, collections, livelessons, livesessions, etc. were conceived intentionally to obscure underlying infrastructure details, so that lesson authors didn't have to care about those things.

However, within the Antidote scheduler, translation of those abstract primitives to the backend infrastructure was done fairly statically; since Kubernetes was the only backend implementation that was planned, these primitives always translated to some kind of Kubernetes API call. So, while the lesson author never had to know anything about Kubernetes, everything still had to run on top of Kubernetes in order to function.

This PR begins the effort changing this requirement by formalizing the boundary between the core scheduler, and the backend implementations, and ensuring that all backend-specific details are well-contained within these implementations. This PR does not include an alternative backend implementation, such as for OpenStack, vSphere, or a cloud provider, but rather refactors the existing scheduler to make this possible. It also converts the existing Kubernetes implementation into this new modular architecture. Future efforts could implement alternative backends by following the example and documentation created here.

Summary of Changes

• Refactors scheduler so the backend is pluggable. This involved formalizing backend functionality using an AntidoteBackend interface, but also documenting plenty of other conventions - see included scheduler/backends/README.md for more details.
• Added a field to livelesson state to track last active time, and refactored garbage collection to use this instead of labels on infrastructure. Removed "BOOP" functionality, as this is no longer needed as a result.
• The antidote-core configuration now includes a section for backend-specific configuration options

[codecov]

Codecov Report

Merging #212 (87f1dc7) into master (f03049e) will increase coverage by 0.53%.
The diff coverage is 19.80%.

❗ Current head 87f1dc7 differs from pull request most recent head d13e8b5. Consider uploading reports for the commit d13e8b5 to get more accurate results

@@            Coverage Diff             @@
##           master     #212      +/-   ##
==========================================
+ Coverage   28.00%   28.54%   +0.53%     
==========================================
  Files          24       24              
  Lines        2246     2032     -214     
==========================================
- Hits          629      580      -49     
+ Misses       1553     1402     -151     
+ Partials       64       50      -14     

Impacted Files	Coverage Δ
api/exp/livelessons.go	0.00% <0.00%> (ø)
db/db.go	0.00% <ø> (ø)
scheduler/backends/kubernetes/ingresses.go	0.00% <0.00%> (ø)
scheduler/backends/kubernetes/namespaces.go	0.00% <0.00%> (ø)
scheduler/backends/kubernetes/services.go	0.00% <0.00%> (ø)
services/utils.go	0.00% <0.00%> (ø)
scheduler/backends/kubernetes/kubernetes.go	16.34% <16.34%> (ø)
scheduler/backends/kubernetes/jobs.go	24.09% <25.00%> (ø)
db/driver_inmem.go	62.06% <42.85%> (-2.06%)	⬇️
scheduler/backends/kubernetes/pods.go	46.55% <46.66%> (ø)
... and 8 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f03049e...d13e8b5. Read the comment docs.