AllCoreDevs - Testing #075
Transcript
- Potuz
GM
- nixo
rare potuz sighting
- jochem-brouwer
gm
- Potuz
why rare?
- FLCL
gm
- Justin Traglia
Gm
- Potuz
Oh Justin is here! we can have a breakout then, I don't care :)
- Parithosh Jayanthi
Potuz flexing some math in the background
- David (dionysuz)
gm ☀️
- Barnabas
Reacted to "Oh Justin is here! w..." with 😂
- Parithosh Jayanthi
Reacted to "Oh Justin is here! w..." with 😂
- David (dionysuz)
Reacted to "Potuz flexing some m..." with 😂
- Barnabas
Replying to "Potuz flexing some m..." you think he understands?
- Barnabas
Reacted to "Potuz flexing some m..." with 😂
- Parithosh Jayanthi
Replying to "Potuz flexing some m..." Sometimes I use the research room whiteboard as a background to seem smarter
- Barnabas
Reacted to "Sometimes I use the ..." with 😂
- nixo
Replying to "why rare?" ¯\_(ツ)_/¯ it’s the differential geometry variant
- nixo
Reacted to "Oh Justin is here! w..." with 😂
- Potuz
Replying to "why rare?" ah, yeah, gave manifolds last semester, that shows how often I use my office's board :)
- nixo
Reacted to "Sometimes I use the ..." with 😂
- Potuz
Replying to "why rare?" surprised you can recognize the topic :)
- Marius Van Der Wijden (M)
49? I have 3 locally
- Łukasz Rozmej
a lot of our failures are error message missmatches - we will fix them soon, then we will see what exactly is the problem
- Mario Vega
Reacted to "a lot of our failure..." with 👍🏼
- Marius Van Der Wijden (M)
I will look into it
- Barnabas
I’d propose to not have a bal-devnet-4 but instead try to scope glam-devnet-0
- spencer
Reacted to "I’d propose to no..." with ➕
- Ishika
Reacted to "I’d propose to no..." with ➕
- felipe
Reacted to "I’d propose to no..." with ➕
- Soubhik
Reacted to "I’d propose to not h..." with ➕
- Justin Traglia
This is the newest version, right? https://github.com/ethereum/consensus-specs/pull/5020
- Stefan Bratanov
Reacted to "This is the newest v..." with 👍
- nflaig
but that doesn't run anything in process slots, it's in process epoch
- Potuz
yeah I didn't understand Stefan's comment
- Potuz
lol
- Potuz
5020 is not in process slots
- Enrico Del Fante (tbenr)
That’s a convincing argument to me actually
- Marius Van Der Wijden (M)
How big would it be in state?
- Dustin
256-384kB is tiny and this is an O(1) thing
- Dustin
state is huge
- nflaig
it same as 96 new validators from what I last checked if we have 3 epochs in state
- Potuz
Nico?
- Marius Van Der Wijden (M)
And it churns per epoch?
- Potuz
someone from Lodestar?
- Barnabas
everyone is happy with every option then nobody is happy about any options lol Can we please agree to do one today on the call please?
- Eitan Seri-Levi
definitely prefer having the full epoch worth (or three epochs worth) of committees instead of just the last PTC of the epoch if we’re going to do this
- Potuz
Replying to "everyone is happy wi..." yeah, we'll decide
- Potuz
Replying to "everyone is happy wi..." don't worry
- Potuz
if we go with the epochs one then I think we can agree on making it 3 epochs and I'll modify the PR
- Eitan Seri-Levi
Reacted to "if we go with the ep…" with 👍
- nflaig
Reacted to "if we go with the ep..." with 👍
- Marius Van Der Wijden (M)
Reacted to if we go with the ep... with "👍"
- Justin Traglia
Reacted to "if we go with the ep..." with 👍
- Potuz
I'll have it ready in 5 minutes
- Justin Traglia
Replying to "I'll have it ready i..." Be sure to update tests & specs for other upgrades (heze) too 😄
- Barnabas
which pr is the final one then?
- Potuz
Replying to "I'll have it ready i..." nah, I'll have the core changes, and will work tests after
- Barnabas
can we please close the other 2 open PRs ?
- NC
I think it’s the one that was closed before?
- Potuz
- Enrico Del Fante (tbenr)
New one or change 4979 ?
- Potuz
I'll update this one
- Enrico Del Fante (tbenr)
ok
- Barnabas
Reacted to "https://github.com/e..." with 👍
- Justin Traglia
Reacted to "https://github.com/e..." with 👍
- Barnabas
and do we do variable ptc deadline?
- Justin Traglia
Replying to "and do we do variabl..." In general, it doesn’t sound like there’s sufficient support for it.
- Marius Van Der Wijden (M)
Months??? In months we should be finishing H*
- Justin Traglia
Toni’s doc: https://hackmd.io/@Nerolation/B1GKtAUcZg
- Parithosh Jayanthi
Replying to "Months??? In months ..." Yeah… that was my understanding as well
- terence
Replying to "Months??? In months ..." Its fine to not have it at the time of the fork though
- Parithosh Jayanthi
Reacted to "Its fine to not have..." with 👍🏽
- terence
Replying to "Months??? In months ..." We can always roll it out later
- Toni Wahrstätter
I agree, since it's a simple (based on clients), non-consensus critical change, it can be further postponed
- Justin Traglia
Reacted to "I agree, since it's ..." with 👍
- Jihoon
Reacted to "I agree, since it's ..." with 👍
- Jihoon
Replying to "https://github.com/e..." FYI, I've closed #4992 and #5020 based on this decision.
- jochem-brouwer
Can you hear me?
- Mario Vega
Replying to "Can you hear me?" no
- Justin Traglia
Reacted to "FYI, I've closed #49..." with ❤️
- Ameziane Hamlat
Replying to "Can you hear me?" no
- Barnabas
Reacted to "FYI, I've closed #49..." with ❤️
- jochem-brouwer
lemme rejoin
- Mario Vega
Reacted to "lemme rejoin" with 👍🏼
- felipe
Reacted to "lemme rejoin" with 👍🏼
- nixo
Reacted to "lemme rejoin" with 🎤
- spencer
Reacted to "lemme rejoin" with 🎤
- lightclient
sry had to step away for another call
- terence
Is epbs done? can we leave and go back to coding? lol
- lightclient
Reacted to "Is epbs done? can we..." with 🔥
- Mario Vega
Reacted to "Is epbs done? can we..." with 🔥
- jvn
Reacted to "Is epbs done? can ..." with 🔥
- jochem-brouwer
- Mario Vega
Replying to "Is epbs done? can we..." Yes it’s done 🙂 please take a look at the PR shared
- terence
Reacted to "Yes it’s done 🙂 ple..." with ❤️
- Łukasz Rozmej
Still working on it :(
- Stefan Starflinger
Replying to "Can you hear me?" reth, besu, geth have it Nimbusel,ethrex and nethermind are missing the flags
- Stefan Starflinger
Replying to "Can you hear me?" And erigon missing
- nflaig
Reacted to "FYI, I've closed #49..." with ❤️
- Barnabas
Reacted to "Yes it’s done 🙂 ple..." with ❤️
- Karim T. (matkt)
Should we try with and without state root in parallel also ?
- nflaig
Replying to "https://github.com/e..." https://github.com/ethereum/beacon-APIs/pull/592 need to re-add next epoch lookahead again
- Mario Vega
Reacted to "FYI, I've closed #49..." with ❤️
- Jihoon
Reacted to "https://github.com/e..." with 🔥
- Karim T. (matkt)
@Rafael Matias (skylenet) where we can find your result ?
- Ameziane Hamlat
Reacted to "@Rafael Matias (skyl..." with ➕
- Soubhik
Reacted to "@Rafael Matias (skyl..." with ➕
- Karim T. (matkt)
Replying to "@Rafael Matias (skyl..." thanks
- lightclient
https://hackmd.io/@matt/frame-mempool https://github.com/ethereum/EIPs/pull/11415
- Rafael Matias (skylenet)
Replying to "@Rafael Matias (sk..." Example of a besu comparison: https://pop-os-shiny.taile8f668.ts.net/compare?runs=1774000622_2696dbe8_besu-bal-sequential%2C1774002127_64a46368_besu-bal-optimized%2C1774005008_106db2cd_besu-bal-noprefetch&labels=instance-id&tableBase=%220%22&filter=memory_ (Might be a bit slow to load)
- David (dionysuz)
Reacted to "https://hackmd.io/@m..." with 🔥
- Mario Vega
Reacted to "https://hackmd.io/@m..." with 🔥
- felipe
Reacted to "https://hackmd.io/..." with 🔥
- wolovim
Reacted to "https://hackmd.io/@m..." with 🔥
- Barnabas
Replying to "https://hackmd.io/@m..." how is the PoC coming?
- jvn
Reacted to "https://hackmd.io/..." with 🔥
- Barnabas
Reacted to "https://hackmd.io/@m..." with 🔥
- Karim T. (matkt)
Reacted to "Example of a besu co..." with 👀
- nixo
Reacted to "https://hackmd.io/@m..." with 🔥
- Karim T. (matkt)
Replying to "@Rafael Matias (skyl..." @Rafael Matias (skylenet) A (besu-bal-sequential) do you have conflict to be sure it’s really sequential ?
- jochem-brouwer
Heeft gereageerd op "https://hackmd.io/..." met 🔥
- Potuz
I pushed the minimum change to add 3 epochs to 4979, lint fails locally who knows why (Python's package managers sucks) And I still need to check if the computation will succeed at that stage of epoch processing (before updating the slot in the state) but the idea should be clear to align on this and just ship devnet 1
- lightclient
Replying to "https://hackmd.io/@m..." it’s coming
- lightclient
Replying to "https://hackmd.io/@m..." sorry have to decide if i’m going to join a nazi suicide cult first
Call summary
Targets
- •ePBS devnet-1 spec release targeting end of week - 00:27:54
Decisions
- •Store 3 epochs of PTC in consensus state (previous + next 2 epochs) - 00:24:27
- •Variable PTC deadline deferred; can add post-fork if needed - 00:29:23
- •Defer minor 8037 spec changes to bal-devnet-4 to stabilize devnet-3 - 00:13:20
- •ePBS devnet-1 focuses on structure testing, not builder API or benchmarking - 00:28:51
Highlights
- Bal Benchmarking:
- ·Batch I/O flag urgently needed from Nethermind and Erigon for benchmarking - 00:32:06
- ·Flag must isolate prefetch optimization, not disable all BAL optimizations - 00:32:43
- ·Geth showing 3X speedup from execution parallelization (limited to 4 CPUs) - 00:40:49
- ·Nethermind block-building memory leak blocking stateful test generation - 00:35:50
- Network Planning:
- ·Proposal: Replace bal-devnet-4 with Glamsterdam-devnet-0 - 00:16:01
- Epbs Specification:
- ·Consensus reached: Store 3 epochs of PTC in state (prev + next 2) - 00:18:11
- ·PTC in state enables testability and eliminates caching bugs across clients - 00:20:41
- ·State size increase: 256-384KB, negligible compared to hundreds of MB state - 00:21:40
- ·Variable PTC deadline postponed; non-consensus, can add later if needed - 00:28:51
- Account Abstraction:
- ·Frame transaction mempool: Strategy 2 (canonical paymaster) preferred approach - 00:42:02
- ·Strategy 1 (self-relaying only) fallback if Strategy 2 faces issues - 00:42:48
- ·Full ERC-7562 permissionless paymasters deemed too complex for initial deployment - 00:43:18
- Bal Devnet Progress:
- ·bal-devnet-3: Nimbus EL and Etherax nearly done with testing - 00:11:44
- ·Besu 19 failures, Geth 49 failures, Nethermind/Reth 400+ failures remaining - 00:12:00
- ·Many failures are exception mapping issues, not logic bugs - 00:12:47
- ·Minor spec changes deferred to bal-devnet-4 to stabilize current implementation - 00:13:20
- Blob Devnet Updates:
- ·blob-devnet-0: Nethermind and Erigon failing to stay at head on trunk branches - 00:08:10
- ·Need stable releases with GetBlobV3 instead of unstable trunk branches - 00:08:34
Action Items
- •Potuz - Update consensus-specs PR #4979 with 3-epoch PTC storage - 00:27:17
- •Nethermind and Erigon teams - Add batch I/O prefetch flag (isolate from other optimizations) - 00:32:06
- •Kamil - Use alternative client (Besu/Geth) for stateful test generation - 00:35:50
- •lightclient - Schedule breakout Wednesday for Frame mempool discussion - 00:42:24
Key decisions
Store 3 epochs of PTC in consensus state (previous + next 2 epochs)
enables testability and eliminates caching bugs across clientsVariable PTC deadline deferred; can add post-fork if needed
non-consensus, can add later if neededDefer minor 8037 spec changes to bal-devnet-4 to stabilize devnet-3
ePBS devnet-1 focuses on structure testing, not builder API or benchmarking