Closed eskimor closed 4 months ago
It sounds fine. Do we have information about the actual variance in practice? It's maybe worse than 0.5 s.
Status on this one?
In the works, hope to publish something next week.
After looking at the code again, it seems that no code changes are required.
no_show_duration
is computed based on the config param no_show_slots
, which is currently set to 3 on Polkadot and Kusama
This translates to 6 * 3 = 18s no-show timeout.
Backing and approval voting timeouts are now configured via runtime (ExecutorParams). I'll prepare a motion for Kusama and then Polkadot to set both in one go.
Here are the calls for Kusama and Polkadot:
With asynchronous backing it is expected for PoVs to take up to 2 seconds in execution. The backing timeout, which is a hard timeout is currently set to exactly those 2s. Even if in memory validation is faster than block production, this seems too tight to be practical. To ensure good performance with 2s blocks, we should be more lenient with the backing timeout.
Rollout
We should bump the approval and no-show timeout first, then once enough validators have upgraded, backing timeout comes next.