Feat request: Check runtime weights in CI for excess

[wischli]

Description

A recent incident has shown that we need to ensure our generated weights don't lead to excess of the maximum block weight or PoV size. Therefore, we need a CI job which checks that for at least Altair and Centrifuge runtime for every release. It can be coupled with the WASM build trigger as both have the same frequency.

The script needs to through all files inside the weights directory of a runtime, calculate the call weights based on the runtimes parameters (e.g. MaxKeys) and compare it with the maximum block weight and maximum pov size.

Research/based on

• function update_collection weight having 175% PoV size for MaxKeys = 1000 and MaxFeeders = 10.

How will this affect the code base

• Improve security

What are foreseen obstacles or hurdles to overcome?

• Mapping the right parameters from the runtime lib.rs to the generic n, m etc. parameters inside the weight files. The correct parameters can be determined by looking into the corresponding pallet's lib.rs, so we might need manual cases for parametrized calls which then requires maintenance for breaking changes (i.e. renaming of parameters)

// pallets/oracle-collection/src/lib.rs
#[pallet::weight(T::WeightInfo::update_collection(
	T::MaxFeedersPerKey::get(),
	T::MaxCollectionSize::get(),
))]
#[pallet::call_index(2)]
pub fn update_collection(...)

// runtime/centrifuge/src/weights/pallet_oracle_collection.rs
fn update_collection(n: u32, m: u32, ) -> Weight

// runtime/centrifuge/src/lib.rs
parameter_types! {
        // snip
	pub const MaxActiveLoansPerPool: u32 = 1000;
	pub const MaxFeedersPerKey: u32 = 10;
}
impl pallet_oracle_collection::Config for Runtime {
	// snip
	type MaxCollectionSize = MaxActiveLoansPerPool;
	type MaxFeedersPerKey = MaxFeedersPerKey;
}

Whoever tackles this issue, needs to check whether such a tool already exists in the Polkadot Ecosystem as it is agnostic to Centrifuge and needed by every Substrate chain.

[lemunozm]

The correct parameters can be determined by looking into the corresponding pallet's lib.rs, so we might need manual cases for parametrized calls which then requires maintenance for breaking changes

I think we do not need it. We can get the weight from the DispatchInfo of the extrinsic: extrinsic.get_dispatch_info().weight.

What I think we need is to call all extrinsics manually. But maybe we can create something similar to what currently works for benchmarking. Some kind of macro in the runtime that calls all extrinsics under the hood if they are registered for benchmarks