Fast Confirmation Rule #011
Transcript
- Jun Song
gm
- Jun Song
Newbie question: do we have any discord channel for discussing FCR?
- Mikhail Kalinin
- Boma’s iPhone
For Grandine no much update waiting for review although devs are just coming back from vacation so probably get a review soon then merge hopefully
- Mikhail Kalinin
Reacted to "For Grandine no much update waiting for review although devs are just coming back from vacation so probably get a review soon then merge hopefully" with 👍
- Dmitrii Shmatko
The question regarding metrics: slot: fallback to finalized slot+1: still on finalized Do we count 2 fallbacks in this case?
- Jun Song
Prysm side: still testing and reviewing, I hope we can merge this soon (as I’ve done reviewing). I forgot the metrics, will add metrics on that branch as well. https://github.com/OffchainLabs/prysm/pull/17122
- Lion dapplion
We disable FCR is clock - head > X where X is 4 epochs I believe
- Mikhail Kalinin
Reacted to "We disable FCR is clock - head > X where X is 4 epochs I believe" with 👍
- Boma’s iPhone
Reacted to "We disable FCR is clock - head > X where X is 4 epochs I believe" with 👍
- Dmitrii Shmatko
We run FCR only for the current slot with timeout of 1/2 slot
- Boma’s iPhone
I guess the can run from your branch
- Lion dapplion
@Dmitrii Shmatko so what do you send in the FCU after block import if block is imported before 1/2 slot?
- Dmitrii Shmatko
Reacted to "@Dmitrii Shmatko so what do you send in the FCU after block import if block is imported before 1/2 slot?" with 👀
- Mikhail Kalinin
- Lion dapplion
Post gloas confirmation time doubles to two slots right? Kind of unfortunate, will you need to update fastconfirm.it?
- Jun Song
It’s nice to see fast confirmation badge in Dora: https://dora.glamsterdam-devnet-7.ethpandaops.io/
- Dmitrii Shmatko
Reacted to "It’s nice to see fast confirmation badge in Dora: https://dora.glamsterdam-devnet-7.ethpandaops.io/" with 🔥
- Mikhail Kalinin
Reacted to "It’s nice to see fast confirmation badge in Dora: https://dora.glamsterdam-devnet-7.ethpandaops.io/" with 🔥
- Mikhail Kalinin
- Jun Song
Reacted to "https://github.com/fradamt/verified-fast-confirmation/" with 👍
- Dmitrii Shmatko
Replying to "@Dmitrii Shmatko so ..." 1/2 slot is only limit for FCR task, to drop it if it somehow is stalled, we don’t need outdated result. Regarding FcU, we send it before FCR, so it lags for 1 slot
- Lion dapplion
Reacted to "https://github.com/f..." with ❤️
- Dmitrii Shmatko
Reacted to "https://github.com/fradamt/verified-fast-confirmation/" with 👍
- Boma’s iPhone
Am interested
Call summary
Decisions
- •Lodestar will disable FCR during range sync (keyed off gossip topic subscription state) - 00:30:58
- •Reversion metric counts the initial fallback to finalized as one event, not once per subsequent slot remaining on finalized - 00:26:57
Highlights
- Open Issues:
- ·Lighthouse node restart resets FCR to finalized, causing apparent safe-hash regression in EL; persisting confirmed root to disk deemed unsafe under current synchrony assumptions - 00:11:34
- ·Safe block hash defaults to justified when FCR is off, causing finalized→justified flicker; proposal to default to finalized instead - 00:47:52
- ·No one currently working on fuzzing FCR; differential fuzzing between implementation pairs proposed as practical approach - 00:44:57
- Client Updates:
- ·Lighthouse: FCR merged, not yet released; runs in ~30ms/slot; no reversions observed outside node restarts - 00:06:06
- ·Teku: FCR implemented on branch, merging to master; ~150-200ms/slot (unoptimized); passes spec tests; mainnet tested - 00:23:28
- ·Lodestar: performance improvement landed; FCR will be disabled during range sync to suppress spurious reversion metrics - 00:29:46
- ·Prysm: FCR PR under review (Offchain Labs branch); performance improvements identified; not yet on mainnet - 00:34:19
- ·Grandine: FCR PR awaiting review; devs returning from vacation - 00:07:21
- Organizational:
- ·Next FCR call (in ~2 weeks) tentatively the last; continuation to be decided then - 00:50:53
- Spec And Research:
- ·New spec PR #5449: adds edge-case test for total-active-balance source and adds safeExecutionBlockHash property to FCR checks - 00:38:36
- ·Post-Gloas: safeExecutionBlockHash will reference parent's execution payload (not confirmed beacon block); FCR confirmation time doubles to two slots - 00:39:28
- ·Updated FCR paper published: algorithm now matches implementation closely; formal proofs moved to appendix - 00:42:29
- ·LLM-generated Lean formalization shared by Francesco; caution: link formalization must be verified before claiming formal verification - 00:43:29
Action Items
- •Mikhail Kalinin - Share beacon-metrics standardized metrics repo link in FCR Telegram group - 00:09:40
- •Mikhail Kalinin / Jun Song - Coordinate with EthPandaOps to integrate Prysm into FCR simulator for mainnet-data backtesting - 00:36:05
Key decisions
Lodestar will disable FCR during range sync (keyed off gossip topic subscription state)
performance improvement to suppress spurious reversion metricsReversion metric counts the initial fallback to finalized as one event, not once per subsequent slot remaining on finalized
clarification on reversion metric behavior