benchmarking-pallets

Benchmarking

Contents

• Benchmarking Overview
• Pallet Benchmarking
• Running Benchmarks
• Generated Weight File
• Precompile Benchmarking
• Best Practices
• Troubleshooting

Benchmarking Overview

Why Benchmarking Matters

• Block Weight Limits: Ensures blocks don't exceed computational limits
• Fee Accuracy: Users pay proportional to actual resource usage
• DoS Prevention: Prevents underpriced operations from being exploited
• Network Stability: Predictable block production times

Weight Components

Weight {
    ref_time: u64,      // Computational time
    proof_size: u64,    // Storage proof size
}

Pallet Benchmarking

1. Add Benchmark Feature

# pallets/my-pallet/Cargo.toml
[dependencies]
frame-benchmarking = { workspace = true, optional = true }

[features]
runtime-benchmarks = [
    "frame-benchmarking/runtime-benchmarks",
    "frame-support/runtime-benchmarks",
    "frame-system/runtime-benchmarks",
]

2. Create Benchmark Module

// pallets/my-pallet/src/benchmarking.rs
#![cfg(feature = "runtime-benchmarks")]

use super::*;
use frame_benchmarking::v2::*;
use frame_system::RawOrigin;

#[benchmarks]
mod benchmarks {
    use super::*;

    /// Benchmark for `do_something` extrinsic
    #[benchmark]
    fn do_something() {
        // Setup: Create any required state
        let caller: T::AccountId = whitelisted_caller();
        let value = 42u32;

        // Optional: Fund the account if needed
        // T::Currency::make_free_balance_be(&caller, 1_000_000_000_000u128.into());

        #[extrinsic_call]
        do_something(RawOrigin::Signed(caller.clone()), value);

        // Verify: Assert the expected state change
        assert_eq!(Something::<T>::get(), Some(value));
    }

    /// Benchmark with linear complexity
    #[benchmark]
    fn do_something_complex(n: Linear<1, 100>) {
        let caller: T::AccountId = whitelisted_caller();

        // Setup scales with n
        for i in 0..n {
            SomeStorage::<T>::insert(i, i);
        }

        #[extrinsic_call]
        do_something_complex(RawOrigin::Signed(caller), n);

        assert!(SomeStorage::<T>::iter().count() as u32 >= n);
    }

    /// Benchmark for storage-heavy operation
    #[benchmark]
    fn clear_storage(n: Linear<1, 1000>) {
        let caller: T::AccountId = whitelisted_caller();

        // Populate storage
        for i in 0..n {
            SomeStorage::<T>::insert(i, vec![0u8; 100]);
        }

        #[extrinsic_call]
        clear_storage(RawOrigin::Signed(caller), n);

        assert_eq!(SomeStorage::<T>::iter().count(), 0);
    }

    impl_benchmark_test_suite!(
        Pallet,
        crate::mock::new_test_ext(),
        crate::mock::Test
    );
}

3. Define Weight Trait

// pallets/my-pallet/src/weights.rs
use frame_support::weights::Weight;

pub trait WeightInfo {
    fn do_something() -> Weight;
    fn do_something_complex(n: u32) -> Weight;
    fn clear_storage(n: u32) -> Weight;
}

/// Placeholder weights for development
impl WeightInfo for () {
    fn do_something() -> Weight {
        Weight::from_parts(10_000, 0)
    }

    fn do_something_complex(n: u32) -> Weight {
        Weight::from_parts(10_000 + n as u64 * 1_000, 0)
    }

    fn clear_storage(n: u32) -> Weight {
        Weight::from_parts(10_000 + n as u64 * 5_000, 0)
    }
}

4. Include in Pallet

// pallets/my-pallet/src/lib.rs
#[cfg(feature = "runtime-benchmarks")]
mod benchmarking;
pub mod weights;
pub use weights::WeightInfo;

#[pallet::call]
impl<T: Config> Pallet<T> {
    #[pallet::call_index(0)]
    #[pallet::weight(T::WeightInfo::do_something())]
    pub fn do_something(origin: OriginFor<T>, value: u32) -> DispatchResult {
        // ...
    }

    #[pallet::call_index(1)]
    #[pallet::weight(T::WeightInfo::do_something_complex(*n))]
    pub fn do_something_complex(origin: OriginFor<T>, n: u32) -> DispatchResult {
        // ...
    }
}

5. Add to Runtime Benchmarks

// runtime/moonbase/lib.rs
#[cfg(feature = "runtime-benchmarks")]
mod benches {
    frame_benchmarking::define_benchmarks!(
        // Existing benchmarks...
        [pallet_my_pallet, MyPallet]
    );
}

Running Benchmarks

Using the Script

# Run all benchmarks for a runtime
./scripts/run-benches-for-runtime.sh moonbase release

# Run specific pallet benchmarks
./scripts/run-benches-for-runtime.sh moonbase release pallet_my_pallet

Manual Benchmark Execution

# Build with benchmarks
cargo build --release --features runtime-benchmarks

# Run frame-omni-bencher
frame-omni-bencher v1 benchmark pallet \
    --runtime target/release/wbuild/moonbase-runtime/moonbase_runtime.wasm \
    --pallet pallet_my_pallet \
    --extrinsic "*" \
    --steps 50 \
    --repeat 20 \
    --wasm-execution compiled \
    --output runtime/moonbase/src/weights/pallet_my_pallet.rs \
    --template benchmarking/frame-weight-template.hbs

Benchmark Parameters

Parameter	Purpose	Typical Value
--steps	Number of sample points	50
--repeat	Repetitions per step	20
--heap-pages	WASM heap size	4096
--wasm-execution	Execution mode	compiled

Generated Weight File

The benchmark generates a weights file like:

// runtime/moonbase/src/weights/pallet_my_pallet.rs
#![allow(unused_parens)]
#![allow(unused_imports)]

use frame_support::{traits::Get, weights::Weight};
use sp_std::marker::PhantomData;

pub struct SubstrateWeight<T>(PhantomData<T>);
impl<T: frame_system::Config> pallet_my_pallet::WeightInfo for SubstrateWeight<T> {
    fn do_something() -> Weight {
        Weight::from_parts(12_345_000, 3456)
            .saturating_add(T::DbWeight::get().reads(1))
            .saturating_add(T::DbWeight::get().writes(1))
    }

    fn do_something_complex(n: u32) -> Weight {
        Weight::from_parts(10_000_000, 1000)
            .saturating_add(Weight::from_parts(500_000, 100).saturating_mul(n as u64))
            .saturating_add(T::DbWeight::get().reads(1 + n as u64))
            .saturating_add(T::DbWeight::get().writes(n as u64))
    }
}

Precompile Benchmarking

1. Create Benchmark Module

// precompiles/my-precompile/src/benchmarking.rs
#![cfg(feature = "runtime-benchmarks")]

use frame_benchmarking::v2::*;

#[benchmarks]
mod benchmarks {
    use super::*;

    #[benchmark]
    fn do_something() {
        let caller = H160::from_low_u64_be(1);
        let value = U256::from(42);

        #[block]
        {
            MyPrecompile::<T>::do_something(caller, value).unwrap();
        }
    }
}

2. Precompile Gas Costs

// In the precompile implementation
const BASE_GAS_COST: u64 = 200;
const PER_ITEM_GAS_COST: u64 = 100;

fn gas_cost(items: u64) -> u64 {
    BASE_GAS_COST.saturating_add(PER_ITEM_GAS_COST.saturating_mul(items))
}

Best Practices

1. Worst Case Scenarios

Always benchmark the worst case:

#[benchmark]
fn bounded_operation(n: Linear<1, MAX_ITEMS>) {
    // Setup with maximum complexity
    for i in 0..n {
        HeavyStorage::<T>::insert(i, vec![0u8; MAX_SIZE]);
    }

    #[extrinsic_call]
    bounded_operation(RawOrigin::Root, n);
}

2. Include All DB Operations

// Weight should reflect actual DB access
Weight::from_parts(compute_time, proof_size)
    .saturating_add(T::DbWeight::get().reads(read_count))
    .saturating_add(T::DbWeight::get().writes(write_count))

3. Test Benchmarks

# Run benchmark tests to verify they work
cargo test -p pallet-my-pallet --features runtime-benchmarks

4. Validate Results

After generating weights:

• Compare with expected complexity
• Check for outliers
• Verify linear relationships match code

Checklist

• Add runtime-benchmarks feature to Cargo.toml
• Create benchmarking.rs with all extrinsics
• Define WeightInfo trait in weights.rs
• Use weight trait in pallet calls
• Add pallet to runtime benchmark list
• Run benchmarks and generate weight file
• Copy generated weights to runtime
• Verify benchmark tests pass
• Review generated weights for accuracy

Troubleshooting

Benchmark Fails

# Check benchmark compiles
cargo check --release -p moonbase-runtime --features runtime-benchmarks

# Run with more verbose output
RUST_LOG=runtime::benchmark=trace frame-omni-bencher ...

Unrealistic Weights

• Check benchmark setup creates realistic state
• Verify worst-case is actually benchmarked
• Look for missing storage operations

Memory Issues

# Increase heap pages
--heap-pages 8192