Gloas (ePBS) testnet profiles for Anchor epbs testing

[shane-moore]

What this adds

Two Gloas (EIP-7732 / ePBS) testnet profiles for ssv-mini, so Anchor's epbs branch can be
exercised end to end on a local Glamsterdam-style chain with its own genesis and SSV contracts:
block proposal through the ePBS block structure, and attestation against the new 3s Gloas deadline.
A 4-operator all-Anchor cluster does real duties; no go-SSV nodes involved.

Built for Anchor devs: point it at your epbs work and watch the cluster propose and attest on a
Gloas chain.

Quick start

ANCHOR_COMMIT=epbs make prepare-anchor   # build node/anchor from sigp/anchor epbs
make run-gloas                           # Gloas at genesis, 4-operator all-Anchor cluster
make logs SERVICE=anchor-node-0          # tail an operator

ePBS is gated on the Ethereum Gloas fork (gloas_fork_epoch); there is no SSV-side fork knob. The
epbs branch already carries this model, so a plain ANCHOR_COMMIT=epbs build is all you need.

Requires Docker and a recent Kurtosis CLI (1.19.x; 1.15.x rejects a GpuConfig builtin in the
pinned ethereum-package). Full walkthrough and edge cases are in the new "Gloas (ePBS) testnet"
section of the README.

Testing your own Anchor changes

prepare-anchor builds node/anchor from any branch/tag/commit on your ../anchor clone's origin:

ANCHOR_COMMIT=my-epbs-feature make prepare-anchor && make run-gloas

Or build the image straight from your working tree and skip prepare entirely:

cd ../anchor && docker build -f Dockerfile.devnet -t node/anchor .
cd ../ssv-mini && make reset PARAMS_FILE=params-gloas.yaml   # scoped teardown + re-run

Either way the profile consumes whatever is tagged node/anchor. There is no hot-swap for Anchor
nodes (unlike restart-ssv-nodes), so the iteration loop is: rebuild the image, then make reset.

The two profiles

Profile	Command	Exercises
params-gloas.yaml	make run-gloas	Gloas at genesis, 4 Anchor operators, 10 keyshare validators registered on-chain. Proposal duty (attestation timing needs the #1061 fix in the image).
params-gloas-builders.yaml	make run-gloas-builders	Above plus 2 genesis ePBS builders and buildoor: external bids, payload reveals, chain-level PTC (payload_attestations).

Reference behavior (tell a regression from a known gap)

Measured live on an epbs image without #1061, local devnet:

• Proposals work. Cluster publishes Gloas blocks ~+0.13s into the slot, canonical (verified on-chain: multiple cluster-proposed slots, all canonical=true, "Successfully published block" on every operator).
• Attestations: broken without #1061; working only once #1061 is also applied. Measured on an epbs image without #1061: 0 cluster attestations published, even though the attestation QBFT reaches COMMIT consensus. The sync-committee half stalls and takes submission down with it (see the #1061 note below). With #1061 also in the image, attestations publish ~+4.09s vs the 3.0s Gloas deadline (QBFT itself ~15ms; the remaining gate is the committee partial-signature batch waiting on the sync-message flow, which fires on Lighthouse's static 4s sync timer; head-event pipeline healthy at ~+0.12s). That ~+4.09s is an Anchor/LH timing-wiring gap, not a cluster failure, and is identical on every operator.
• Cluster-proposed slots are execution-empty by design. Anchor self-builds and never reveals the payload (SIP-94 envelope stub), so PTC votes payload_present=false on them and the EL does not advance that slot. Expected, not a bug.
• Gloas is gated on the Ethereum fork; SSV attestations need the #1061 fix in your anchor image. ePBS activates purely from gloas_fork_epoch; there is no SSV-side fork knob. With Gloas active the committee runs GloasBeaconVote for attestations, but the sync-committee signing path still runs BeaconVote (feat(validator_store): migrate sign_committee_sync_committee_signatures to GloasBeaconVote at CStar sigp/anchor#1061, open), so the two resolve to different QBFT instances instead of sharing one. Observed mechanism: the attestation instance reaches COMMIT, but the sync instance round-changes every slot and never completes, and because the committee's post-consensus partial-signature batch is sized to attestations + sync messages, it never fills, so the agreed attestation is never submitted (0 published, every operator). So attestation testing needs an anchor image carrying #1061; proposals, external-builder bids, and chain-level PTC work regardless. (Consensus succeeds; only submission is blocked.)

Related Anchor issues: sigp/anchor#1061 (sync GloasBeaconVote migration; required for SSV attestations), sigp/anchor#1082 / #1078 (PTC duties).

Mechanics worth knowing

• Imports a forked ethereum-package pin (shane-moore/ethereum-package, by commit), fetched automatically by Kurtosis. Needed because Gloas genesis requires ethereum-genesis-generator >= 6.0.0 (no tagged release ships it) plus a one-line fix so genesis ePBS builders do not collide with the preregistered keyshare validators (indices 64-73).
• Uses nethermind + lighthouse glamsterdam-devnet-5 images (geth has no Glamsterdam build).
• Registration scripts carry two live-verified fixes: explicit gas limits (nethermind's eth_estimateGas underestimates SSVNetwork delegatecall paths ~3x) and msg.value on the payable v2.0.0 bulkRegisterValidator.
• register_validators (default false) gates on-chain registration so the default aetheria flow is untouched; the Gloas profiles set it true so operators get duties.
• All-Anchor cluster (ssv.count: 0): no node/ssv image needed, and monitor is disabled.

Why draft

This is a testing tool for the in-flight epbs work, not a finished feature. Opening as draft for
visibility and so Anchor devs can use it now; happy to adjust scope or take out of draft per review.

[iurii-ssv]

Hey @shane-moore, I think most of this functionality has already been implemented in main branch, aside from:

1. params-gloas-builders.yaml — the buildoor external-builder profile: ePBS bids, payload reveals, chain-level PTC (payload_attestations). This is the one real net-new capability; nothing in main exercises the builder/payload/PTC side.
2. Anchor Prometheus metrics (--metrics on port 5164 in nodes/anchor/node.star) — small, self-contained, useful regardless of Gloas.

Maybe you could rebase this PR as needed and we can merge it into main ?

[shane-moore]

Thanks @iurii-ssv, agreed on both.

Metrics: split out into #46. The --metrics flags plus the published port, verified live on a 4-operator run (all four nodes serve /metrics on 5164).

Builders profile: this one needs an upstream fix first. builder_count requires an ethereum-package newer than our 6.1.0 pin, and its current implementation starts builder keys at the participants' validator sum. For our layout that is index 64, exactly where the static keyshares live (64-73), so genesis aborts with duplicate public key in validator set. This PR worked around it by pinning main.star to a patched fork, but that pin is global rather than per-profile and moves every profile 60 commits off the pinned release, so it is not something to carry on main. Fixed at the source instead: ethpandaops/ethereum-package#1470. Once that lands and we bump the pin I will rebuild the profile on current main and open it separately.

Closing this rather than rebasing, since the rest is superseded by #34, #35, #36, #37, #39, #40, #41. The sigp/anchor#1061, #1078 and #1082 caveats in the description are stale too, all three closed in June/July.