Open joshwlewis opened 1 year ago
@joshwlewis I took a look at the indexer logs in production, and whilst we're no longer hitting any rate limits after #115, the whole indexing process takes ~5-6mins from start to finish with the current "pull everything" approach. This will worsen over time, as the number of buildpack/version combinations naturally grows.
As such, I think it would still be worth merging this PR, to both lower that 5-6 mins (which affects the latency of publishing a new buildpack release) and also protect against future rate limit issues.
As an added bonus, after this PR merges the updated_at
field returned by the API will actually match the last time an entry's metadata changed, and not the last time the metadata was needlessly refreshed :-)
Another effort to reduce the number of pulls: check the database to see if we have the exact entry already. We don't need to refetch the upstream image when the image address (which includes the sha) hasn't changed.