michaelsproul
commented
2 years ago TL;DR:
- Import delay is the only delay that's truly hardware & software dependent. Everything else depends on network conditions outside our control.
Lighthouse could likely be a bit quicker to process blocks in case of re-org, so I'll open an issue for that.
This is a dashboard that I created to investigate block delays using @macladson's metrics for observation delay, import delay and set-as-head delay.
One interesting thing I noticed was that there can be really large spikes in the set-as-head delay due to re-orgs. In one case I observed a block being set as head 10 seconds after it finished being imported (because it orphaned the block from the previous slot). In another case I saw a block take 2.5 seconds to import because it re-orged across an epoch boundary (slot 2485473 on mainnet).