Enable async backing on all testnet system chains

[mrcnski]

Built on top of #2826 which was a trial run.

Guide: https://github.com/w3f/polkadot-wiki/blob/master/docs/maintain/maintain-guides-async-backing.md

[bkchr]

• polkadot-sdk/cumulus/client/consensus/aura/src/collators/lookahead.rs

  Lines 98 to 99 in 05cdf41

  	/// The length of slots in this chain.
  	pub slot_duration: SlotDuration,

  these and similar slot_duration settings need to be dynamic to react to changes on chain.
• I actually thought that we need some migration for:

  polkadot-sdk/substrate/frame/aura/src/lib.rs

  Line 180 in 05cdf41

  pub(super) type CurrentSlot<T: Config> = StorageValue<_, Slot, ValueQuery>;

  But I realized as we go down with the slot duration, the slots will jump to a much higher slot and thus the check that the slot increases will be correct.

[rphmeier]

these and similar slot_duration settings need to be dynamic to react to changes on chain.

If configured correctly, you don't necessarily need to have 6s slots for 6s blocks.

With an unincluded segment length of 3, you can allow up to 3 blocks per 12s slot and always have something ready. 6s slots would be better.

[bkchr]

The slot duration is used to calculate the slot. If the node and runtime side do not match, the blocks will be rejected.

[rphmeier]

and has it changed in any of the chains? I just took another look and it seems to continue using the genesis slot duration of seconds, right? as I wrote above, you can still enable and benefit from asynchronous backing with a 12s slot time

[bkchr]

and has it changed in any of the chains? I just took another look and it seems to continue using the genesis slot duration of seconds, right?

For me it looks like this changed: https://github.com/paritytech/polkadot-sdk/pull/2949/files#diff-95384194e1d261e31cd6d6326580597925d02e6877fa546c6379b8f31bf6a7c3R134
See the diff, the original import of SLOT_DURATION was changed everywhere.

as I wrote above, you can still enable and benefit from asynchronous backing with a 12s slot time

Yeah I get what you said, but I also have to say that it makes stuff more "weird". But maybe it is just me that likes constant slots ;)

[bkchr]

Not your fault as the benchmark before was not correct. However, changing this will break the benchmark for anyone downstream that doesn't change the slot duration to 6 seconds. This needs to made generic.

[georgepisaltu]

Working on a fix.

[georgepisaltu]

Fix here.

[bkchr]

If someone says that they have tested this and it works, I'm fine. I just don't that we break stuff again and then need to fix it again etc.

[s0me0ne-unkn0wn]

See the diff, the original import of SLOT_DURATION was changed everywhere.

BTW, that's right, but TBH, I don't like the result. Having that SLOT_DURATION and everything in parachain_common is very error-prone; it's easy to screw things up and import SLOT_DURATION from one place and HOUR from another and get a working parachain that has the wrong perception of time. I'd say with the introduction of async backing, those constants couldn't be called "common" anymore and thus shouldn't be present in a package named whatever_common. But that can be addressed in a follow-up.

[bkchr]

I mean this parachain_common crate is only for these test chains we have there. SLOT_DURATION is also not defined anywhere else. I don't think that this is a huge problem. Consolidating this stuff between the huge number of test runtimes we have makes sense IMO.