Refactor Non-Validator and Epoch to have a transition callback by samliok · Pull Request #521 · ava-labs/Simplex

samliok · 2026-08-18T17:46:56Z

The main feature of this PR is the callback in the Epoch and NonValidator structs that signal to parent components a transition is required. This fixes the non-validator bug of restarting in-between epochs when it shouldn't be.

In the Epoch struct, when a sealing block is indexed we will now call the OnSealingBlockIndex which will notify the instance of an epoch change. Similarly, the NonValidator has a TransitionToValidator config that notifies the Instance when it should transition an epoch change.

The main benefit is now the non-validator can signal to the Instance when it is ready to transition to a validator. Before, since the Storage component was handling this change, the Storage component needed to re-call into the non-validator to confirm whether the sealing block actually meant a transition was needed. Since the non-validator has all the info, that logic should stay within the non-validator. This also helps avoiding deadlock since we don't need to be concerned about re-grabbing the non-validator lock.

This PR also simplifies and refactors some of the logic for transitioning an epoch.

No more nodeRole iota
removes some methods that i felt weren't adding much such as transitionEpochNonValidator and transitionEpochValidator
other small cleanups that were kind of hard to separate from the main changes

yacovm · 2026-08-21T14:07:39Z


+		if block.SealingBlockInfo() != nil {
+			// are we the highest validator
+			highestEpoch, highestValidatorSet := n.epochs.highestEpoch()


highestEpoch == md.Seq {

So this basically means we have replicated the sealing block of the last epoch?

And the reason this works is that we replicate backwards?

If so, can we add some comments?

yep, if we have an epoch in the epochs map, that means we have verified it's accepted by enough validators. Therefore, if the block we are indexing the sealing block that is part of the highest epoch we should check if we need to transition.

will add a comment

I'm not sure this is good enough, on second thought.

If we start the non-validator we initialize this map with the last thing we have in the ledger and then if we commit a sealing block we will transition, won't we?

I think we should peek at the P-chain and compare the validator set and only if they're the same, then trigger it.

Thoughts?

I think we should peek at the P-chain and compare the validator set and only if they're the same, then trigger it.

Hmm not sure this works either, because what if we are in the middle of a transition? and the highest pchain validator set is different than the current epoch. Maybe our validator is down, and it needs to catch up in order to sign off on the approval that will transition the epoch.

ok so i think as long as we do #530 we should be able to know for sure that we have the latest epoch for real, thoughts?

yea i think this is sufficient

yacovm · 2026-08-23T21:35:42Z

-type EpochAwareStorage struct {
+// InstanceStorage is a wrapper around Storage that skips indexing Telocks
+// and delegates post-index handling to a caller-provided onIndex hook.
+type InstanceStorage struct {


why the rename? The name doesn't say anything about what it is doing, but just says now who is consuming it.

i think EpochAware was no longer relevant, i renamed to CallbackStorage 🤷

yacovm · 2026-08-23T21:37:52Z

-	if numBlocks == 0 {
-		return metadata.StateMachineBlock{}, 0, fmt.Errorf("no genesis block found in storage")
+	switch {
+	case i.nv != nil:


can we be more explicit about what i.nv != nil and i.e != nilmean?

yacovm · 2026-08-23T21:39:39Z

 		i.e.Stop()
+		// Wipe out the WALs from the config so we won't try to load them again
+		if garbageCollectWAL {
+			i.Config.WALs = nil


isn't this a problem? How are we going to re-create the WAL now?

i think this is a problem, but i just copied the code from before. Probably should address the WAL issues in a separate PR.

I had this PR, maybe we can add it on to that

yacovm · 2026-08-23T21:46:52Z

-	// On epoch change, garbage collect the WAL to remove all entries from previous epochs.
-	if err := i.wal.GarbageCollect(math.MaxUint64); err != nil {
-		i.Config.Logger.Error("Error garbage collecting epoch config on epoch change", zap.Error(err))
+func GetHighestValidatorSet(platform PlatformChain) (common.Nodes, error) {


This function is only used once, why is it needed?

yacovm · 2026-08-25T17:00:47Z

+	tests := []struct {
+		name          string
+		setup         func(t *testing.T) (*testChain, []*messageInfo)
+		expectedCalls []transitionCall


why is this a slice and not a single element?

yacovm · 2026-08-25T17:36:49Z

+		select {
+		case i.epochChanges <- ec:
+			return
+		// The slot holds a stale epoch change: take it, keep the newer of the two and retry.


I'm not strictly opposed to this, but - do we really need it now? Do we have a test that fails if this isn't done?

I don't think it can happen because we aren't acquiring the lock so we shouldn't deadlock. I still think we should have it and maybe I can log a warning so our test fails if it ever does?

samliok force-pushed the on-index branch from 7c175b8 to f13ab23 Compare August 18, 2026 18:24

yacovm reviewed Aug 21, 2026

View reviewed changes

samliok force-pushed the on-index branch from 54027c9 to e828b64 Compare August 21, 2026 15:39

samliok force-pushed the on-index branch from e828b64 to b8e5cc1 Compare August 21, 2026 15:43

samliok force-pushed the on-index branch from b8e5cc1 to 296d7fc Compare August 21, 2026 15:56

Base automatically changed from double-start-nit to main August 21, 2026 16:01

samliok force-pushed the on-index branch from 296d7fc to 46c1175 Compare August 21, 2026 16:01

samliok linked an issue Aug 21, 2026 that may be closed by this pull request

Only start an epoch instance if has synced to the current epoch and is part of the latest validator set #445

Closed

yacovm reviewed Aug 21, 2026

View reviewed changes

Comment thread simplex/epoch.go Outdated

samliok added 4 commits August 23, 2026 21:20

update on index callback

8f91f02

update test

3db47a3

add test for index

5f547ef

add comments

918ccd6

yacovm force-pushed the on-index branch from aec2f98 to 918ccd6 Compare August 23, 2026 19:20

yacovm reviewed Aug 23, 2026

View reviewed changes

samliok added 4 commits August 24, 2026 10:24

rename to CallbackStorage

1dd4254

create non-validator config

3370131

add helpful bools

9213209

on index removed from epoch

a10f679

yacovm reviewed Aug 25, 2026

View reviewed changes

yacovm approved these changes Aug 25, 2026

View reviewed changes

make slice into single element

382904a

samliok merged commit 6b9a7f2 into main Aug 26, 2026
7 checks passed

samliok deleted the on-index branch August 31, 2026 14:06

Uh oh!

Conversation

samliok commented Aug 18, 2026

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!

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!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants