updated duties using split endpoints

[james-prysm]

What type of PR is this?

Feature

What does this PR do? Why is it needed?

validator client calls split endpoints for grpc instead of the combined call(getdutiesv2) and has some optimizations swapping current and next duties instead of calling both each epoch. the split duties properly fixes the attestation vs proposer dependent roots for lookahead.

best way to test is to remove the fork if statement and call the split directly , rebuild the image then run in kurtosis

How to reproduce the reorg test for this PR

End-to-end test recipe for validating the split-duty-endpoint path on a local kurtosis devnet, including a forced dependent-root mismatch that exercises the checkDependentRoots update branches (equivalent to a reorg).

1. Kurtosis config

~/git/kurtosis/gloas-config-4node.yml:

participants:
  - el_type: ethrex
    el_image: ethpandaops/ethrex:glamsterdam-devnet-4
    el_extra_params:
      - --http.api=eth,net,web3,admin
    cl_type: prysm
    cl_image: gcr.io/offchainlabs/prysm/beacon-chain:latest
    vc_image: gcr.io/offchainlabs/prysm/validator:latest
    supernode: true
    cl_extra_params:
      - --verbosity=debug
    vc_extra_params:
      - --verbosity=debug
    count: 4

network_params:
  fulu_fork_epoch: 0
  gloas_fork_epoch: 1
  seconds_per_slot: 6
  genesis_delay: 30

additional_services:
  - dora
  - tx_fuzz

global_log_level: debug

2. Build local images (arm64)

bazel build //cmd/beacon-chain:oci_image_tarball \
  --platforms=@io_bazel_rules_go//go/toolchain:linux_arm64_cgo \
  --config=release --nouse_action_cache && \
docker rmi gcr.io/offchainlabs/prysm/beacon-chain || true && \
docker load -i bazel-bin/cmd/beacon-chain/oci_image_tarball/tarball.tar && \
bazel build //cmd/validator:oci_image_tarball \
  --platforms=@io_bazel_rules_go//go/toolchain:linux_arm64_cgo \
  --config=release --nouse_action_cache && \
docker rmi gcr.io/offchainlabs/prysm/validator || true && \
docker load -i bazel-bin/cmd/validator/oci_image_tarball/tarball.tar

Switch linux_arm64_cgo → linux_amd64_cgo for x86.

Precondition: kill any running enclave before rebuilding (docker rmi fails if the image is in use):

kurtosis enclave rm -f gloas-grpc

3. Start enclave

The latest ethereum-package main has a Starlark error (GpuConfig undefined in zkboost_launcher.star). Pin to the last good commit:

kurtosis run --enclave gloas-grpc \
  github.com/ethpandaops/ethereum-package@555e9b72f25bd42a9c13640faef4f2866e62bc21 \
  --args-file ~/git/kurtosis/gloas-config-4node.yml

4. Inducing a dependent-root mismatch

On a clean 4-node network, natural reorgs don't happen, so the mismatch branch in checkDependentRoots doesn't get exercised. Two options to force it:

Option A — temporary code hook (deterministic)

Add a one-shot force to validator/client/duties.go's checkDependentRoots (do NOT commit this):

import "strconv"

// At package level:
var (
    debugForcedPrev bool
    debugForcedCurr bool
)

// Inside checkDependentRoots, right after computing needsPrevUpdate/needsCurrUpdate:
if headSlotNum, parseErr := strconv.ParseUint(head.Slot, 10, 64); parseErr == nil {
    if primitives.Slot(headSlotNum) >= 96 && headSlotNum%32 == 20 {
        if !debugForcedPrev {
            debugForcedPrev = true
            needsPrevUpdate = true
            log.WithField("headSlot", head.Slot).Warn("DEBUG: forcing needsPrevUpdate=true")
        } else if !debugForcedCurr {
            debugForcedCurr = true
            needsCurrUpdate = true
            log.WithField("headSlot", head.Slot).Warn("DEBUG: forcing needsCurrUpdate=true")
        }
    }
}

This fires the prev-update branch once at slot 116 and the curr-update branch once at slot 148 — both post-gloas (fork epoch = 2, so slot 64+).

Rebuild validator, restart enclave, then monitor.

Option B — network partition (non-deterministic but closer to real)

Pause a beacon container for ~1 epoch across an epoch boundary:

docker pause cl-2-prysm-geth--<suffix>
sleep 180
docker unpause cl-2-prysm-geth--<suffix>

With only 4 validators this may not actually produce a reorg — the paused node usually just catches up to the same canonical chain. A genuine fork needs a proper network partition:

docker network disconnect <kurtosis-net> cl-1-prysm-geth--<suffix>
docker network disconnect <kurtosis-net> cl-2-prysm-geth--<suffix>
# wait 1-2 epochs
docker network connect <kurtosis-net> cl-1-prysm-geth--<suffix>
docker network connect <kurtosis-net> cl-2-prysm-geth--<suffix>

Option A is the recommended path for regression testing — deterministic and ~14 minutes to both branches firing.

5. Monitoring

After the enclave is running past slot ~148 (≈15 min post-start), grep vc-1-geth-prysm logs:

# Forced triggers and resulting UpdateDuties info logs (option A)
kurtosis service logs gloas-grpc vc-1-geth-prysm -a \
  | sed 's/\x1b\[[0-9;]*m//g' \
  | grep -E "DEBUG: forcing|Updated duties due to"

# The bug this PR fixes: count "no duties for validators" errors.
# Expect 0 after fix. Pre-fix produces ~130 errors per forced trigger.
kurtosis service logs gloas-grpc vc-1-geth-prysm -a \
  | sed 's/\x1b\[[0-9;]*m//g' \
  | grep -c "no duties for validators"

# Split endpoint call counts — should rise on every forced/real trigger.
kurtosis service logs gloas-grpc vc-1-geth-prysm -a \
  | sed 's/\x1b\[[0-9;]*m//g' \
  | grep -oE "BeaconNodeValidator/(GetAttesterDuties|GetProposerDutiesV2|GetSyncCommitteeDuties|GetPTCDuties|GetDutiesV2)" \
  | sort | uniq -c

6. Pass criteria

• Two DEBUG: forcing WARN lines at slots 116 and 148.
• Two Updated duties due to … dependent root change INFO lines, each ~1–1.6s after the corresponding force.
• "no duties for validators" error count: 0.
• Split endpoint counts increase by one full set per trigger (AttesterDuties / ProposerDutiesV2 / SyncCommitteeDuties / PTCDuties).
• No unrelated validator errors (ignore the HEZE_/EIP7928_/VIEW_FREEZE_* yaml config-parse warnings — those are unknown future-fork fields in the kurtosis preset, not caused by this PR).

7. Cleanup

Revert the debug hook from Option A before committing. The only intended changes on this PR are the two v.clearDuties() removals in checkDependentRoots plus the new test.

Which issues(s) does this PR fix?

Fixes #

Other notes for review

Acknowledgements

• I have read CONTRIBUTING.md.
• I have included a uniquely named changelog fragment file.
• I have added a description with sufficient context for reviewers to understand this PR.
• I have tested that my changes work as expected and I added a testing plan to the PR description (if applicable).

[nalepae]

Approved without re-running the kurtosis devnet on the latest state of this PR.