Make specifying replicas/duration optional in the experiment step

[jessesuen]

This simplifies manifests by making replicas and duration as optional fields. Also it gets progressDeadlineSeconds from the rollout spec.

Builds ontop of #239, so only look at last commit.

[codecov]

Codecov Report

Merging #241 into master will increase coverage by 0.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #241      +/-   ##
==========================================
+ Coverage   85.24%   85.25%   +0.01%     
==========================================
  Files          49       49              
  Lines        4386     4389       +3     
==========================================
+ Hits         3739     3742       +3     
  Misses        460      460              
  Partials      187      187

Impacted Files	Coverage Δ
rollout/experiment.go	83.03% <100%> (ø)	⬆️
experiments/experiment.go	93.97% <100%> (+0.06%)	⬆️

---

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 37bc1f7...59bff86. Read the comment docs.

[dthomson25]

LGTM, but quick question: If we have no duration, how does the experiment know when to stop and continue the rollout's steps? If the experiment has an analysis, then it should finish once the analysis completes. Otherwise, how will the experiment end and allow the rollout to continue? Is this a valid configuration? If not, we should add some validation to prevent the user from creating a never-ending experiment.

[jessesuen]

LGTM, but quick question: If we have no duration, how does the experiment know when to stop and continue the rollout's steps? If the experiment has an analysis, then it should finish once the analysis completes. Otherwise, how will the experiment end and allow the rollout to continue? Is this a valid configuration? If not, we should add some validation to prevent the user from creating a never-ending experiment.

Sorry I forgot to answer this question. Never ending/indefinite experiments are a desired use case. To proceed to the next step of the rollout, a manual termination of the Experiment would need to be performed.

If the experiment has an analysis, then it should finish once the analysis completes.

Actually we may need to support both use cases. Currently stopping an experiment because all analysis completed, is not supported. In this scenario, the experiment will continue to run (to it's duration, or forever).

We will need new syntax to indicate that an Experiment should stop when analysis completes. This is a decision that might need to be made for individual analyses.