AllCoreDevs - Testing #098
Transcript
- Mario Vega
gm!
- Ladislaus
Reacted to "gm!" with 👋
- Justin Traglia
Reacted to "gm!" with 👋
- Justin Traglia
- Toni Wahrstätter
- Justin Traglia
Reacted to "https://github.com/ethereum/pm/issues/2230" with 😄
- Barnabas
would any client team NOT be ready by EOD tomorrow?
- Nico Flaig
lodestar has a release as well
- emma (reth)
Reth released this AM
- Daniel Lehrner
Besu has a release as well
- Dustin
Nimbus will be releasing tomorrow
- james
Ours is in the works should be out soon hopefully tomorrow will check with preston
- Barnabas
Replying to "lodestar has a relea..." 10 min ago lol
- saulius
Grandine should be out tomorrow.
- Barnabas
we are rolling out releases on devnet 8
- Nico Flaig
Replying to "lodestar has a relea..." we had rc on friday which is the same commit
- Barnabas
Replying to "Besu has a release a..." hyperledger/besu:26.9.0 ? release comments don’t mention sepolia glamsterdam support.
- Justin Florentine
did we resimulate without the tracing on?
- Miguel Tenorio
Reacted to "we are rolling out releases on devnet 8" with 👍
- Łukasz Rozmej
yes we should
- Łukasz Rozmej
probably yes, unless there is something off with our assumptions, always re-validate
- Justin Florentine
and eventually mainnet shadowfork
- Dragan Rakita
There is probably diff between building and validating block
- Dustin
justification never made sense
- Barnabas
any reason the tag is not just fcr ?
- Julian Ma
- Barnabas
seems like fast could be used elsewhere too
- Julian Ma
fcr tag also works for me but would want to check with @Mikhail Kalinin and @Francesco D'Amato
- Potuz
It can't go backwards
- Justin Traglia
Replying to "any reason the tag i..." I typically suggest not using acronyms; for that reason I slightly prefer fast
- Potuz
@Julian Ma I dont understand why a new endpoint instead of just using the same endpoint with different CL behavior, are RPC providers claiming that they need to serve both?
- Jihoon
Replying to "any reason the tag i..." Not sure if it has to be 4-letters, but I'd vote for confirmed . fast block makes me wonder what's slow block then 🤷♂️
- Francesco D'Amato
Replying to "@Julian Ma I dont un..." Yes, right now at least some of them are planning to serve both, if they want to adopt FCR
- Francesco D'Amato
Replying to "@Julian Ma I dont un..." (Or just not adopt FCR)
- Potuz
Replying to "@Julian Ma I dont un..." ah ok, that explains it
- Justin Traglia
Replying to "any reason the tag i..." Ah yeah something like that would be great too
- Jihoon
Reacted to "Ah yeah something like that would be great too" with 👍
- Julian Ma
Replying to "@Julian Ma I dont un..." Yes that is what they are seeing
- Mikhail Kalinin
Replying to " fcr tag also works..." We can have “confirmed” tag If we want to. Don’t have a strong preferences, “fcr” kind of attached to a specific algo now and we will have fast confirmation in-protocol coming with DC upgrade. So, I am slightly against “fcr”
- Justin Florentine
confirmed
- Enrico Del Fante
Not a fan of “fastBlockHash” . So it will be optional or default to safe?
- cayman
Reacted to "Not sure if it has t..." with 👍
- Justin Florentine
Reacted to "confirmed" with ➕
- cayman
Reacted to "confirmed" with 👍
- Enrico Del Fante
Reacted to "confirmed" with ➕
- Justin Traglia
Reacted to "confirmed" with ➕
- Kevaundray Wedderburn
Would we eventually deprecate safe or will safe be served in perpetuity?
- Potuz
Reacted to "Yes that is what the..." with 👍
- Dustin
gloas is very, very late
- Barnabas
this is all optional can be rolled when clients ready no?
- Kevaundray Wedderburn
Does this require a hardfork?
- Dustin
~half the client shave releases out
- cayman
Reacted to "this is all optional..." with 👍
- Potuz
I am confused, what requires coordination?
- Marius van der Wijden
Requires an update to FCU
- Jihoon
Which clients already support FCR? Nimbus, Lighthouse and Nethermind?
- Dustin
requiring an update to FCU is major at this stage
- Julian Ma
Replying to "Which clients alread..." Teku and Prysm as well
- Julian Ma
Replying to "Which clients alread..." Lodestar
- Justin Traglia
Replying to "requiring an update ..." Yep I agree.
- Julian Ma
Replying to "Which clients alread..." From my understanding
- Jihoon
Replying to "Which clients alread..." A half of clients!
- Barnabas
can’t we serve this on beacon api?
- Toni Wahrstätter
Which json rpc would respond with the confirmed block?
- Toni Wahrstätter
My understanding is, we need it in the engine api for ELs to expose it to rpc users?
- Mikhail Kalinin
Replying to "Which json rpc would..." Any which accept “safe”, “late”, “finalized” block tags. e.g. getBlockByNumber
- Justin Florentine
Reacted to "My understanding is, we need it in the engine api for ELs to expose it to rpc users?" with 👍
- Mario Vega
Reacted to "My understanding is, we need it in the engine api for ELs to expose it to rpc users?" with 👍
- Mikhail Kalinin
Replying to "can’t we serve this ..." You mean add a new API with the fast confirmed block hash?
- Toni Wahrstätter
Or put it in as default but fill it with justification slot if fcr not active
- Francesco D'Amato
For RPC providers I think this should work just fine
- Nico Flaig
Reacted to "confirmed" with ➕
- Justin Traglia
CL breakout: https://ethereumfoundation.zoom.us/j/89441658268?pwd=yjqf4TheEb1R7ZSuesCdOAy5tdyVk7.1
- Potuz
Replying to "Or put it in as defa..." that changes every node
- Toni Wahrstätter
Replying to "For RPC providers I ..." If they run a specific client combo that passes it over the engine api
- Kevaundray Wedderburn
Replying to "Or put it in as defa..." I think there would need to be a flag to say that fcr was not active, so one doesn’t confuse fcr with justification
- Barnabas
“unmuting is responsibility”
- Toni Wahrstätter
Replying to "Or put it in as defa..." But makes running any client pajr with fcr work
- spencer
Reacted to "“unmuting is resp..." with 😂
- Mario Vega
Reacted to "“unmuting is responsibility”" with 😂
- Parithosh Jayanthi
Reacted to "“unmuting is responsibility”" with 😂
- spencer
Replying to "“unmuting is resp..." never unmute
- danceratopz
Technical issues :)
- Barnabas
- Marius van der Wijden
Can we also quickly make an announcement about the bootnodes?
- Barnabas
- Parithosh Jayanthi
https://dora.glamsterdam-devnet-8.ethpandaops.io/slot/331715#arrival
- Parithosh Jayanthi
https://discord.com/channels/595666850260713488/892088344438255616/1554135728919806066
- Dragan Rakita
We can hear you 🙂
- Mario Vega
I could hear both
- Parithosh Jayanthi
Ah lol
- Parithosh Jayanthi
okie
- Marius van der Wijden
lool
- Barnabas
micdao
- Marius van der Wijden
micdao
- Justin Florentine
speakerdao
- Barnabas
I’m gonna jump to the cl call
- danceratopz
Please halp. Sound down.
- Marius van der Wijden
yep
- Parithosh Jayanthi
Reacted to "yep" with 👍
- Marius van der Wijden
much faster
- Łukasz Rozmej
we should have control group without debug tracing
- spencer
Reacted to "much faster" with 💙
- Mario Vega
Reacted to "much faster" with 💙
- Marius van der Wijden
Reacted to "we should have con..." with 👍
- Mario Vega
Reacted to "we should have control group without debug tracing" with 👍
- Ben Adams
Reacted to "we should have control group without debug tracing" with 👍
- Parithosh Jayanthi
Reacted to "we should have control group without debug tracing" with 👍
- Marius van der Wijden
The tracing for Pawels attack block should be somewhere in the gigabytes of data, so slowdowns through tracing are expected
- Mario Vega
I think the qs is whether the combination of loading the contract + jumpdest is the problem in particular
- Mario Vega
why not loading the contract + other compute operation?
- Łukasz Rozmej
you can regenerate the bals and then do it again, but that needs custom orchestration
- Daniel Lehrner
Reacted to "you can regenerate the bals and then do it again, but that needs custom orchestration" with 👍
- Karim Taam
One node doing full sync without pruning and this one will be a peer for the other nodes
- Łukasz Rozmej
Reacted to "One node doing full sync without pruning and this one will be a peer for the other nodes" with 👍
- Justin Florentine
sepolia forks in a week
- Justin Florentine
this sounds to me like we have a pretty scary attack but its isolated in the lab, and we haven't been able to exercise it in an APPLIED form
- Mario Vega
Reacted to "this sounds to me like we have a pretty scary attack but its isolated in the lab, and we haven't been able to exercise it in an APPLIED form" with 👍🏼
- Łukasz Rozmej
Replying to "this sounds to me like we have a pretty scary attack but its isolated in the lab, and we haven't been able to exercise it in an APPLIED form" hard to say, should be investigated more
- Maria Silva
Can we also run Jochem’s attack?
- Stefan Starflinger
sure
- Łukasz Rozmej
master branch, we are not main hippies ;)
- Ben Adams
Replying to " master branch, we are not main hippies ;)" main master branch
Call summary
Targets
- •EOD tomorrow — all client releases due for Sepolia Shadow Fork preparation - 00:13:59
- •~1.5–2 weeks — Sepolia mainnet fork - 00:23:40
Decisions
- •200M gas limit remains safe pending devnet retest without debug tracing; no repricing decisions until clean measurement obtained - 00:19:34
- •FCR new block tag to be opt-in only; no change to default FCU behavior; CL+EL pairs that support it coordinate independently - 00:33:25
- •New block tag name preferred: 'confirmed' (over 'fast' or 'fcr') - 00:29:57
Highlights
- Testing Progress:
- ·Devnet-8 stable over weekend under three attack variants; most clients processing blocks within ~6 seconds - 00:12:54
- ·Security team found memory-explosion bug in ethrex on devnet-8; already fixed - 00:14:12
- ·200M gas limit assessed safe: Pavel's attack peaks at 2.5s execution; Jochem's slow blocks were compounded by debug tracing running simultaneously - 00:17:41
- ·Significant gap between Benchmarker (~2.5s worst case) and devnet (~5s) for JUMPDEST analysis on 100M gas block; cloud-attached disks and active debug tracing are suspected causes - 00:46:00
- ·Plan: disable debug tracing on devnet-8, then rerun Pavel's and Jochem's attacks; Besu requests trunk branch be used instead of release for this test - 00:57:08
- Critical Infrastructure:
- ·EF bootnodes migrated from EthPandaOps to a new EF team; old nodes remain active until at least mainnet Glamsterdam, then deprecated - 00:41:17
- ·ELOC keys missing 0x prefix in JSON genesis file; Nethermind fails without it — PR open in eth-clients/mainnet for client approval - 00:39:46
- Fork Status And Schedule:
- ·Client release deadline is EOD tomorrow to prepare for Sepolia Shadow Fork - 00:13:59
- ·Releases confirmed: Teku (out), Lodestar (out), Reth (out), Besu (out), Nimbus (tomorrow), Prysm (tomorrow), Grandine (tomorrow) - 00:15:28
- ·After Sepolia fork, devnet-8 images will revert to latest trunk branches - 00:16:23
- Fast Confirmation Rule Fcr:
- ·Proposal: add a new 'confirmed' block tag (not repurpose 'safe') to enable FCR adoption; backwards-compat concerns from RPCs are blocking uptake - 00:24:11
- ·Chainlink went live with FCR today; other bridges and L2s want it; RPC providers blocked by needing to serve both safe and FCR simultaneously - 00:25:42
- ·Does not require a hard fork; consensus that FCR should be opt-in via compatible CL+EL pairs without changing default FCU behavior for all nodes - 00:33:25
Action Items
- •EthPandaOps (Stefan) + Pavel - Disable debug tracing on devnet-8 and rerun JUMPDEST attacks (Pavel's and Jochem's) to isolate tracing overhead; use Besu trunk branch - 00:57:18
- •Julian Ma - Open async thread (EF R&D Discord or EF Magicians) on 'confirmed' block tag PR (execution-apis #908) - 00:31:23
Key decisions
200M gas limit remains safe pending devnet retest without debug tracing; no repricing decisions until clean measurement obtained
Safety of 200M gas limit confirmed pending further testing without debug tracing.FCR new block tag to be opt-in only; no change to default FCU behavior; CL+EL pairs that support it coordinate independently
Fast Confirmation Rule will be optional via a new block tag without altering default forkchoice update behavior.New block tag name preferred: 'confirmed' (over 'fast' or 'fcr')
Preferred naming for the new block tag is 'confirmed'.
AI Disclaimer: Some content or metadata on EIPsInsight may be AI-inferred or automatically compiled. If you find any discrepancy, please contact us at dev@avarch.org.