Regression in pallet-scheduler: integrity_test fails after upgrading to Polkadot SDK stable2412

[SunTiebing]

After upgrading our parachain project to Polkadot SDK version stable2412 (git rev: cff69157b985ed76), our build succeeds, but running runtime tests fails.

We are using the following command to run runtime tests:

SKIP_WASM_BUILD=1 cargo test --features "with-all-runtime" --lib

The following test fails:

---- __construct_runtime_integrity_test::runtime_integrity_tests stdout ----
thread '__construct_runtime_integrity_test::runtime_integrity_tests' panicked at .../substrate/frame/scheduler/src/lib.rs:397:4:
Must be possible to submit a small lookup

From tracing the code in pallet-scheduler, the issue comes from the integrity_test function:

let small_lookup = lookup_weight::(128);
assert!(small_lookup.all_lte(limit), "Must be possible to submit a small lookup");

However, our runtime does not override any specific weight values or logic, and we did not see such errors prior to this SDK upgrade. This seems to be a regression caused by new integrity constraints introduced in the scheduler.

Steps to Reproduce:
1. Upgrade to latest polkadot-sdk (rev cff69157b985ed76)
2. Run cargo test for the runtime crate
3. Observe failure in __construct_runtime_integrity_test

Please advise whether:
• This is an intentional change and requires us to increase MaximumWeight, or
• This test is too strict and should be relaxed, or
• There is a missing migration or documentation for handling this

Thanks!

[bkchr]

CC @ggwpez

[ggwpez]

What weights do you have configured for pallet scheduler, just ()?
It seems that there is a possible issue with your config and this test fails to indicate that.
Are you an open-source project where I could check the code?

[SunTiebing]

@ggwpez
Thanks for the response!

In our actual runtime, we configure WeightInfo as:
WeightInfo = pallet_scheduler::weights::SubstrateWeight<Runtime>;

But in our mock runtime for testing, WeightInfo is set to (), following the pattern seen in other pallets, such as: polkadot-sdk

Is this now expected behavior, requiring real weight implementations even in mocks? Or is this a regression in how the test handles default/mock weights?

Our project is open source. You can find the code here: bifrost

Thanks again for your support!

[ggwpez]

It only fails in your mock, not production runtime? So the test works for production bifrost runtime? I will give it a look, thanks for the link.

[ggwpez]

The issue is that on stable2412 we have incorrect default weights. I checked all other stable releases and this seems to be the only one affected.
It does have the benchmarking fix in it though - but no weight update 🫠

Proper solution would be to run benchmarks for this pallet and use your own generated weights until we publish a fix. Alternatively, you could also copy&paste the Scheduler weights of the Asset Hub Runtime, but you should still sanity check them manually and it is a bit hacky.