Re-enable optimistic sync for Gloas payload envelopes by dapplion · Pull Request #9934 · sigp/lighthouse

dapplion · 2026-08-29T12:34:42Z

Breaking change

Can we force existing Gloas nodes to resync when updating to this new version? Then we can simplify the migrations as the property execution_status has continuity in mainnet from Fulu to Gloas. Otherwise there's an awkward partial disappearance between v29 and v30.

Issue Addressed

Re-enable optimistic sync for Gloas payload envelopes

Proposed Changes

Track the status of the payload (FULL) node in ProtoNode::execution_status

eserilev · 2026-08-29T13:42:31Z

Yes we can force a resync for all Gloas nodes

pawanjay176 · 2026-09-02T00:25:36Z

I think we can easily force resync devnet nodes. Not an issue. Halfway through the review. Will complete tomorrow

michaelsproul · 2026-09-02T09:42:26Z

+                // The payload is revealed later, in an envelope.
+                execution_status: ExecutionStatus::Irrelevant(false),


Almost need another variant (like NotYetRevealed or Pending) for the Gloas case?

Added NotYetRevealed

michaelsproul · 2026-09-02T09:43:34Z

            // Execution status tracking only exists on V17 (pre-Gloas) nodes.
            if let Ok(v17) = parent.as_v17()
                && v17.execution_status.is_invalid()
            {
                return Err(Error::ParentExecutionStatusIsInvalid {
                    block_root: block.root,
                    parent_root: parent.root(),
                });
            }


Out of date. Probably need to handle parent payload status here to know if the parent's invalidity affects this new block (if we build on the empty variant of a parent whose payload is invalid, that's ok).

-> Good test case.

michaelsproul · 2026-09-02T09:57:25Z

                // If the node has an invalid execution payload, reduce its weight to zero.
                0_i64
                    .checked_sub(node.weight() as i64)
                    .ok_or(Error::InvalidExecutionDeltaOverflow(node_index))?


I think we need to not do this for Gloas. I think we should just do it for the full weight?

michaelsproul · 2026-09-02T09:57:44Z

                // Invalid nodes always have a weight of 0.
                *node.weight_mut() = 0;


I think we need to tweak this for Gloas too.

michaelsproul · 2026-09-02T10:07:28Z

            && matches!(block.execution_status, ExecutionStatus::Valid(_))
        {
-            self.propagate_execution_payload_validation_by_index(parent_index)?;
+            let parent_status = node.parent_payload_status().unwrap_or(PayloadStatus::Full);


Conflicting convention with get_parent_payload_status:

https://github.com/dapplion/lighthouse/blob/8d1aa2a09853d6df7efbcd6b63186d2833bacd61/consensus/proto_array/src/proto_array.rs#L185-L189

Maybe we should be consistent and treat pre-Gloas blocks as Empty? Or add a new enum with a pre-Gloas status to this codepath.

Added ParentPayloadStatus and removed all defaults

michaelsproul · 2026-09-02T10:10:46Z

+            PayloadStatus::Empty | PayloadStatus::Pending => self
+                .proto_array
+                .empty_node_execution_status(*block_root)
+                .ok(),


Nit: probably shouldn't swallow this error (we don't expect this to error, I imagine?)

michaelsproul · 2026-09-02T10:23:24Z

                }
-                Ok(ExecutionStatus::Irrelevant(_)) => break,
-                Err(_) => break,
+                ExecutionStatus::Irrelevant(_) => break,


Once we add the new status, can probably return an error for that variant (we should not be able to invalidate a head block with an unrevealed payload, and the block should not have any Full ancestors with unrevealed payloads)

michaelsproul · 2026-09-02T10:28:31Z

            }

            if let Some(parent_index) = node.parent() {
+                status = node.parent_payload_status().unwrap_or(PayloadStatus::Full);


To exercise this path, need a test case for the first payload after the Gloas fork being invalid

michaelsproul · 2026-09-02T10:33:32Z

Could simplify this loop using ProtoArray::children field

michaelsproul · 2026-09-02T10:40:42Z

+                // A Gloas descendant becomes invalid only when it built on the payload of the
+                // ancestor. A descendant that took the `EMPTY` edge stays viable.
+                if let ProtoNode::V29(gloas_node) = node
+                    && gloas_node.parent_payload_status != PayloadStatus::Full
+                {
+                    continue;
+                }


It's incorrect to continue here. A block (node) with parent_payload_status Empty could descend from some other Full block which is invalid because it descends from the latest valid hash.

michaelsproul · 2026-09-02T10:41:12Z

+                        // In Gloas this means only that the payload is not revealed yet. The block did
+                        // commit to the invalid ancestry. Pre-Gloas this state is a contradiction.
+                        match node {
+                            ProtoNode::V29(gloas_node) => {
+                                gloas_node.execution_status = ExecutionStatus::Invalid(
+                                    gloas_node.execution_payload_block_hash,
+                                );
+                            }


Looks like a good use for the separate variant

Added NotYetRevealed

dapplion · 2026-08-31T13:03:56Z

-                            proto_block.execution_status.is_optimistic_or_invalid();
+                        let new_head_is_optimistic = fork_choice
+                            .get_node_execution_status(&block_root, head_payload_status)
+                            .is_some_and(|status| status.is_optimistic_or_invalid());


Needs to walk backwards to find the first full ancestor which is the head's payload status

dapplion · 2026-09-08T15:26:20Z

+                // The payload is revealed later, in an envelope.
+                execution_status: ExecutionStatus::Irrelevant(false),


Added NotYetRevealed

dapplion · 2026-09-08T15:26:41Z

            && matches!(block.execution_status, ExecutionStatus::Valid(_))
        {
-            self.propagate_execution_payload_validation_by_index(parent_index)?;
+            let parent_status = node.parent_payload_status().unwrap_or(PayloadStatus::Full);


Added ParentPayloadStatus and removed all defaults

dapplion · 2026-09-08T15:29:10Z

+                        // In Gloas this means only that the payload is not revealed yet. The block did
+                        // commit to the invalid ancestry. Pre-Gloas this state is a contradiction.
+                        match node {
+                            ProtoNode::V29(gloas_node) => {
+                                gloas_node.execution_status = ExecutionStatus::Invalid(
+                                    gloas_node.execution_payload_block_hash,
+                                );
+                            }


Added NotYetRevealed

dapplion · 2026-09-08T15:30:51Z

dapplion · 2026-09-08T15:31:23Z

+                // A Gloas descendant becomes invalid only when it built on the payload of the
+                // ancestor. A descendant that took the `EMPTY` edge stays viable.
+                if let ProtoNode::V29(gloas_node) = node
+                    && gloas_node.parent_payload_status != PayloadStatus::Full
+                {
+                    continue;
+                }


dapplion force-pushed the gloas-optimistic-payload-status branch from 105919b to 142e3f1 Compare August 29, 2026 12:44

dapplion requested review from eserilev and pawanjay176 August 29, 2026 12:45

dapplion force-pushed the gloas-optimistic-payload-status branch from 142e3f1 to 9bc89e2 Compare August 29, 2026 14:16

Re-enable optimistic sync for Gloas payload envelopes

8d1aa2a

dapplion force-pushed the gloas-optimistic-payload-status branch from 9bc89e2 to 8d1aa2a Compare August 29, 2026 15:53

dapplion mentioned this pull request Sep 1, 2026

Gate payload import on optional execution proofs #9808

Merged

pawanjay176 added the under-review A reviewer has only partially completed a review. label Sep 2, 2026

michaelsproul added the backwards-incompat Backwards-incompatible API change label Sep 2, 2026

michaelsproul reviewed Sep 2, 2026

View reviewed changes

Comment thread consensus/proto_array/src/proto_array.rs

michaelsproul reviewed Sep 2, 2026

View reviewed changes

dapplion mentioned this pull request Sep 2, 2026

🤖 Suggested fixes for the sigp#9934 review findings dapplion/lighthouse#145

Open

dapplion force-pushed the gloas-optimistic-payload-status branch from 700056b to 8d1aa2a Compare September 4, 2026 09:31

dapplion mentioned this pull request Sep 7, 2026

🤖 Gloas optimistic sync, final form dapplion/lighthouse#149

Open

dapplion added 4 commits September 8, 2026 13:35

Parent payload edge enum with the invalidation body reset to unstable

daa578f

Rework payload invalidation around find_deepest_node_to_invalidate

ac4c23c

Sweep descendants through the children index in one loop

9d45dce

Add a NotYetRevealed execution status for unrevealed Gloas payloads

9e0ca91

dapplion added 3 commits September 8, 2026 16:41

Preserve settled payload verdicts when an envelope arrives

17c6e30

Fix execution typo in the unrevealed-head log

dea8696

Resolve unrevealed payload heads to their empty-side execution status

d87098a

dapplion commented Sep 8, 2026

View reviewed changes

		// The payload is revealed later, in an envelope.
		execution_status: ExecutionStatus::Irrelevant(false),

		// Invalid nodes always have a weight of 0.
		*node.weight_mut() = 0;

Uh oh!

Conversation

dapplion commented Aug 29, 2026

Breaking change

Issue Addressed

Proposed Changes

Uh oh!

eserilev commented Aug 29, 2026

Uh oh!

pawanjay176 commented Sep 2, 2026

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

michaelsproul Sep 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

michaelsproul Sep 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

michaelsproul Sep 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

michaelsproul Sep 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

michaelsproul Sep 2, 2026 •

edited

Loading

michaelsproul Sep 2, 2026 •

edited

Loading

michaelsproul Sep 2, 2026 •

edited

Loading

michaelsproul Sep 2, 2026 •

edited

Loading