robert

https://github.com/paritytech/polkadot/pull/4640/files#r805223564 Looks suspicious. I pushed a branch reverting this fix to https://github.com/paritytech/polkadot/tree/rh-revert-connectivity-fix if you'd like to try it out.

@crystalin Specifically the collators need it. This should only affect collator logic. That might be an argument against this being the underlying cause, because AFAIK it began even before the...

https://github.com/paritytech/substrate/compare/polkadot-v0.9.15...polkadot-v0.9.16 These are all Substrate changes from that period https://github.com/paritytech/polkadot/compare/v0.9.15...v0.9.16 It doesn't look like lip2p changed at all; both versions `Cargo.lock` point to v0.40.0 These are all Polkadot changes. It...

> is it safe to revert validators to v0.9.15 on Alphanet with runtime 9160 ? (I know there is a new host, but I'm not sure if/when it is actually...

Can we get a graph on Grafana indicating the amount of reorgs?

@crystalin The idea is that the block is propagated immediately after authoring, which also means we can use more time for transferring larger PoVs. With 1 block delay we have...

@crystalin This depends on the type of parachain. If there's a plausibly unique block author, the ideal honest node would broadcast to validators at the same time as gossiping to...

I added a bunch more info to this issue, and will file some more concrete issues to build a roadmap to completion.

@bkchr one interesting byproduct of this model is that we don't necessarily require the relay-parent to increase from candidate to candidate. We could accept 2 candidates in a row with...

Yes, that is what I'm saying. If the relay parent doesn't advance, the same author may be selected multiple times in a row on a parachain, so we may need...