Moved from GitHub dgraph/3988
Posted by campoy:
We should come up with a series of queries on a dataset that we can run frequently (for each PR, or at least nightly) and store the results so we can visualize the evolution of our performance over time.
This could be a great opportunity for dogfooding Dgraph, by storing all the results in Dgraph and exposing them with the GraphQL layer (idea proposed by @MichaelJCompton).
Each issue that appears in the area/performance label should have a benchmark to ensure we do not introduce regressions.