Nightly protocol runtime config includes wrong regular_op_cost

[matklad]

This line is wrong:

nearcore/nearcore/res/runtime_configs/123.json

Line 154 in 5cbb3a3

"regular_op_cost": 3856371,

It should say this instead:

nearcore/nearcore/res/runtime_configs/48.json

Line 154 in 5cbb3a3

"regular_op_cost": 2207874,

This'll probably get caught once #4961 is implemented. But I wonder if we should improve, mid-term, how we represent config changes?

Configuration bugs are notorious for causing outages, this is the evidence that "let's just dump config to JSON" is not fool-proof enough.

I think it makes sense to refactor the code to allow storing just diffs, which explicitly say "we change from this value to this value".

I am also 0.7 sure that for source of truth for costs, we should switch from JSON to plain-text readable format from estimator:

https://github.com/near/nearcore/blob/master/runtime/runtime-params-estimator/costs.txt

but that is blocked on #4633

[Longarithm]

Great catch. I see why this happened:

• first, PR with 123.json was introduced, which was based on older config without regular_op_cost updated
• then another PR introducing lower regular_op_cost was introduced and merged
• and then 123.json was merged, but it still was based on old op cost, so it was increased back

This seems to be a good reason to switch to diffs, because the speed of merging different costs indeed may be different, and consequences may be worse if smth like this will leak to mainnet.

For now I will hotfix the cost, because it affects only betanet which is hardforked every day.

[matklad]

@Longarithm Closed?

[Longarithm]

Yeah, the cost is fixed. I will come back to storing diffs later

[Longarithm]

Regarding diffs: today I found that max_transaction_size field is taken as default value. If I want to put it into config, I have to put it into each config instead of only the first one.