Find majorities in Spark measurements

[bajtos]

eta: 2024-08-31

When evaluating measurements submitted by Spark checker nodes, we should check whether we have enough redundant measurements from different parties to make it safe to assume there is an honest majority.

Measurements belonging to the minority with different retrieval results from the majority should be rejected.

Fields to cross-check:

• All measurements in the majority should report the same providerId
• All measurements in the majority should report the same indexerResult
• All measurements in the majority should report the same provider_address and protocol
• Response status code
• CAR size in bytes
• CAR checksum
• Was it a timeout?
• Was the CAR too large?

As part of this work, we need to change the way how we calculate retrieval success rates.

• At the moment, we measure % of measurements that report a successful retrieval
• After the change, we should measure % of committees (tasks/deals) where the majority reports a successful retrieval

[bajtos]

I created a new roadmap item for Deal Retrievability Score (#156) so that we can prioritise it against other items in our backlog.

I created a PR to fix the RSR calculation to include minority results, i.e., to restore the algorithm we used before I implemented "committees and majorities" in CheckerNetwork/spark-evaluate#304:

• fix: include minority results in RSR calculation spark-evaluate#343

Once that PR has landed, I'd like to close this roadmap item as done.

cc @juliangruber @patrickwoodhead