auxiliary info disemination msm only by samliok · Pull Request #524 · ava-labs/Simplex

samliok · 2026-08-18T20:21:26Z

Blocks now embed an AuxiliaryInfoBatch holding one entry per node. The builder collects legal entries from a new auxInfoStore mempool (populated by HandleAuxiliaryInfo) instead of calling Generate itself.
SignApproval is exported so the transition listener (next PR) can sign and broadcast approvals, and GetAuxiliaryHistory exposes the aux history traversal.

In future prs, the HandleAuxiliaryInfo and HandleApproval methods will be called by the instance that feeds in messages.

samliok · 2026-08-26T15:16:16Z

 )

 func TestInstanceMixedNodeType(t *testing.T) {
+	t.Skip("skipping until test instance refactor")


im skipping these because i didn't want to spend time updating the tests since i have them passing here

#508

yacovm

Made a first pass.

The most important comment is regarding resetting the aux info store between epochs.

yacovm · 2026-08-26T18:37:24Z

+	// Can be used for backward-compatibility and upgrade purposes.
+	Version VersionID `canoto:"uint,1"`
+
+	// Info is opaque bytes that can be used by applications to encode any information that describes


Info is opaque bytes that can be used by applications to encode any information that describes the current state for the application.

That's the most ambiguous definition I've read for quite some time, but I can only blame myself.

yacovm · 2026-08-26T19:22:52Z

+
+// AuxiliaryInfo defines application-specific information for applications that might care about epoch change,
+// such as threshold distributed public key generation.
+type AuxiliaryInfo struct {


We should add an epoch field here, otherwise we may get a misunderstanding if a node is lagging and sends information regarding the wrong epoch.

shouldn't this be handled by Data and the auxiliary info application?

I think the only source of truth for whether AuxiliaryInfo is valid should be the IsLegalAppend function call. It shouldn't be up to the msm/instance to filter aux infos. If the application cared about epochs, they can encode that information in Data or maybe encode a timestamp.

But there is no point in putting stuff into memory that if that stuff belongs to the wrong epoch.

The framework should do its best to provide reliable data delivery.

If the node is correct but it's just behind, we should exclude its contribution at the protocol layer, not at the app layer.

ok, will add

i still think a malicious node can bypass this by encoding a different epoch. Therefore, we would still need some sort of check in the application layer

added the epoch to the struct, but I'll set up blocking old auxiliary info for when I wire it into the Instance since the MSM isn't aware of the current epoch.

i still think a malicious node can bypass this by encoding a different epoch. Therefore, we would still need some sort of check in the application layer

Yeah but we should offload protection of malicious input to the application, but not let correct but stale input be regarded as malicious

yacovm · 2026-08-26T20:02:28Z

+	a.lock.Lock()
+	defer a.lock.Unlock()
+
+	// Iterate in node ID order so the returned entries are deterministic.


why do we need it to be deterministic? So that anyone verifying this would iterate in the same order when applying IsLegalAppend?

it's helpful for testing. i don't think it matters for production, because this is only called by the block builder.

yacovm · 2026-08-26T20:09:23Z

 	return asn1.Marshal(signedMsg)
 }

+func SignApproval(signer common.Signer, nextPChainReferenceHeight uint64, auxInfoDigest [32]byte) ([]byte, error) {


what's the reason this is exported? This was createSelfApproval before.

it's used in a future PR

https://github.com/ava-labs/Simplex/pull/507/changes

yacovm · 2026-08-26T20:12:17Z

 	VerifiedBlockMessage        *VerifiedBlockMessage
 	VerifiedReplicationResponse *VerifiedReplicationResponse
+
+	// Epoch Transition Messages


I'm missing the part where we send these two...?

also in a future PR 😉

https://github.com/ava-labs/Simplex/pull/507/changes

yacovm

Some more comments. Will make another pass after they're addressed.

yacovm · 2026-08-27T14:43:31Z

+		}
+
+		if err := a.app.IsLegalAppend(info.Version, validators, legalHistory, info.Data); err != nil {
+			// we don't remove this info from the mempool. maybe it can be added in a different block


This is a good decision, because DKG material is linear in the size of the validator set, so we may not be able to add more than a single entry per block.

yacovm · 2026-08-27T14:54:53Z

+		if auxInfoHistory.OldestVersionID != info.Version {
+			return nil, [32]byte{}, false, fmt.Errorf("proposed auxiliary info does not have the proper version %d: %w", auxInfoHistory.OldestVersionID, err)
+		}
+		if err := sm.AuxiliaryInfoApp.IsLegalAppend(auxInfoHistory.OldestVersionID, validators, auxInfoHistory.Data, info.Data); err != nil {


This just checks that adding info is allowed if we just add this info.

But it might be that adding the first info is allowed on its own, adding the second info is allowed on its own, but adding the first info and then the second info is forbidden.

For example, if proposedAuxInfos contains several info that are the same one, we should only be able to add the first one, not the rest.

We need to accumulate the history as we go through the auxiliary info we check are legal appends.

ah good catch, we need to have some better testing with the auxiliary info stuff. maybe in some future prs or when DKG is added we can create some better aux tests

yacovm · 2026-08-27T19:50:05Z

+	if ai.data != nil {
+		cloned.data = make([]common.AuxiliaryInfo, len(ai.data))
+		for i, entry := range ai.data {
+			entry.Data = slices.Clone(entry.Data)


There is no reason to clone the data. We are cloning to avoid cloning the canoto fields.

yacovm · 2026-08-27T19:50:29Z

+	}
+	if ai.data != nil {
+		cloned.data = make([]common.AuxiliaryInfo, len(ai.data))
+		for i, entry := range ai.data {


for i, entry := range ai.data { - this is racy.

We should only use i to access otherwise we copy canotoData too.

yacovm

Two important comments and then LGTM

samliok force-pushed the aux-dis-1 branch from e7d30f5 to 8655e67 Compare August 21, 2026 15:39

samliok force-pushed the aux-dis-1 branch from 8655e67 to e360eab Compare August 21, 2026 15:43

samliok force-pushed the aux-dis-1 branch from e360eab to 1bb93d6 Compare August 21, 2026 15:56

samliok force-pushed the aux-dis-1 branch from 1bb93d6 to ed1ba0c Compare August 21, 2026 16:01

samliok force-pushed the aux-dis-1 branch 2 times, most recently from 68a788d to d8290b2 Compare August 26, 2026 14:13

Base automatically changed from move-wal-function to main August 26, 2026 14:20

samliok force-pushed the aux-dis-1 branch from d8290b2 to ad0a6a8 Compare August 26, 2026 14:20

samliok commented Aug 26, 2026

View reviewed changes

yacovm reviewed Aug 26, 2026

View reviewed changes

samliok force-pushed the aux-dis-1 branch from 9be6503 to 0edb2d7 Compare August 26, 2026 21:06

samliok added 5 commits August 26, 2026 17:17

auxiliary info disemination msm only

67a50f5

add skips

12b2793

nits

d3405b5

add encoding

ee510a7

add epoch to aux info

fa46341

samliok force-pushed the aux-dis-1 branch from e1af316 to fa46341 Compare August 26, 2026 21:17

yacovm reviewed Aug 27, 2026

View reviewed changes

samliok added 3 commits August 27, 2026 13:09

add msm tests, aux info batching and logging

ef86451

update test

daea1db

reduce diff

338b53e

yacovm reviewed Aug 27, 2026

View reviewed changes

clone

a52a00f

yacovm approved these changes Aug 27, 2026

View reviewed changes

samliok merged commit 652700b into main Aug 27, 2026
7 checks passed

Uh oh!

Conversation

samliok commented Aug 18, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

yacovm left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

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!

yacovm left a comment

Choose a reason for hiding this comment

Uh oh!

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!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

yacovm left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants