Frame-omni-bencher weights are worse than old bench

[mordamax]

a6ce732
This diff shows

• red (frame-omni-bencher)
• green (old bench)

the environment is the same (Github runner)

[mordamax]

@kianenigma @ggwpez ^^ pinging here
because of this issue we are still defaulted to old benchmarks

[iulianbarbu]

I'll look into this

[bkchr]

@iulianbarbu try to change the following line:

polkadot-sdk/.github/workflows/cmd.yml

Line 367 in 89b0228

cargo install --path substrate/utils/frame/omni-bencher --locked

to:

cargo install --path substrate/utils/frame/omni-bencher --locked  --profile production

[mordamax]

@iulianbarbu note that to test this workflow change can be done through paritytech-stg, as this event trigger takes always workflow from master, so it's easier to test from there

[ggwpez]

Was this closed by accident or do we know that this was the issue?

[mordamax]

@ggwpez yes, by accident as PR was merged, it auto-closed issue, thanks

[iulianbarbu]

Latest test here, which used @bkchr 's suggested fix, ran against corresponding 9edaef0 on paritytech/polkadot-sdk/master, resulted in a smaller difference than what was shown initially in the issue description.

My previous test which resulted in similar numbers as @mordamax showed in the issue description was done against: 350a6c4, which is not much different from 9edaef0 (as in nothing happened in between based on my understanding, that would impact pallet_balances weights on westend).

I would conclude that @bkchr 's fix resolves the issue. Will follow up with the same change on master.