Disallow usage of from_ref_time for weights after the relevant polkadot version

[Chralt98]

Hey valuable Polkadot community,

we were able to construct a runtime with from_ref_time here, although weights-v2 are active by our used polkadot version. This resulted in transactions not being able to be processed by the nodes. If a transaction doesn't use from_ref_time we were able to execute those transactions.

Could you please deprecate the usage of from_ref_time so that everyone sees that this function shouldn't be called anymore by the specific polkadot version?

[ggwpez]

This resulted in transactions not being able to be processed by the nodes. If a transaction doesn't use from_ref_time we were able to execute those transactions.

I dont follow this logic. from_ref_time(x) is exactly equivalent to from_parts(x, 0), it was just a convenience wrapper.

Anyhow, it was deprecated in paritytech/substrate#13475 and removed in paritytech/substrate#14397 already.

[Chralt98]

We are currently unable to process transactions on battery station here. A transfer call just doesn't get executed. I looked in the code and found out that each transaction with from_ref_time inside the weight function wasn't able to get processed. However, the transaction set_heap_pages() was able to get processed by the local node setup (it just was the first example that I could find without from_ref_time).

Thanks for your quick response! And cool that you already removed it.

EDIT: this commit resulted txs not being processed with from_ref_time inside the weight function (assume a local dev node setup).

[ggwpez]

I just queried System::blockWeight and it reports about this:

  normal: {
    refTime: 0
    proofSize: 0
  }
  operational: {
    refTime: 0
    proofSize: 0
  }
  mandatory: {
    refTime: 7,348,715,184
    proofSize: 69,273,040
  }
}

Which means that some on_initialize or on_finalize hooks are consuming about 70 MB of pov size, while your limit is (correctly) set to 5 MiB.
So that could be why only transactions that have from_ref_time in them are being processed.

It looks like there is a sudo key set, so you can try Sudo(Utility::WithWeight(SetCode(...), (0, 0)) to force a transaction in.
I think there is currently no PoV base weight per transaction, so could be that this works. But better test with a Remark first. There should be a debug log in finalize, in case that you have debug logging enabled somewhere.

[ggwpez]

Yes pretty much confirmed. Prediction market on_init is using about 70 MB proof per block as introduced here: zeitgeistpm/zeitgeist@bb0e671#r127466669

We are adding some sanity checker for parachains here #1493, but currently the mandatory weight needs to be checked manually or by using something like subweight: https://github.com/ggwpez/substrate-weight-compare