fix: determine F3 participants relative to current network name

CHANGELOG.md

@@ -12,6 +12,7 @@
 - Fix a bug in the `lotus-shed indexes backfill-events` command that may result in either duplicate events being backfilled where there are existing events (such an operation *should* be idempotent) or events erroneously having duplicate `logIndex` values when queried via ETH APIs. ([filecoin-project/lotus#12567](https://github.com/filecoin-project/lotus/pull/12567))
 - Event APIs (Eth events and actor events) should only return reverted events if client queries by specific block hash / tipset. Eth and actor event subscription APIs should always return reverted events to enable accurate observation of real-time changes. ([filecoin-project/lotus#12585](https://github.com/filecoin-project/lotus/pull/12585))
 - Add logic to check if the miner's owner address is delegated (f4 address). If it is delegated, the `lotus-shed sectors termination-estimate` command now sends the termination state call using the worker ID. This fix resolves the issue where termination-estimate did not function correctly for miners with delegated owner addresses. ([filecoin-project/lotus#12569](https://github.com/filecoin-project/lotus/pull/12569))
+- Fix a bug in F3 participation API where valid leases may get removed due to dynamic manifest update. ([filecoin-project/lotus#12597](https://github.com/filecoin-project/lotus/pull/12597))
 
 ## Improvements
 

chain/lf3/f3.go

@@ -155,7 +155,7 @@ func (fff *F3) runSigningLoop(ctx context.Context) {
 			clear(alreadyParticipated)
 		}
 
-		participants := fff.leaser.getParticipantsByInstance(mb.Payload.Instance)
+		participants := fff.leaser.getParticipantsByInstance(mb.NetworkName, mb.Payload.Instance)
 		for _, id := range participants {
 			if _, ok := alreadyParticipated[id]; ok {
 				continue

chain/lf3/participation_lease.go

@@ -112,15 +112,26 @@ func (l *leaser) participate(ticket api.F3ParticipationTicket) (api.F3Participat
 	return newLease, nil
 }
 
-func (l *leaser) getParticipantsByInstance(instance uint64) []uint64 {
+func (l *leaser) getParticipantsByInstance(network gpbft.NetworkName, instance uint64) []uint64 {
 	l.mutex.Lock()
 	defer l.mutex.Unlock()
+	currentManifest, _ := l.status()
+	currentNetwork := currentManifest.NetworkName
+	if currentNetwork != network {
+		log.Warnf("no participants for network: current network (%s) does not match requested network (%s) at instance %d", currentNetwork, network, instance)
+		return nil
+	}
 	var participants []uint64
 	for id, lease := range l.leases {
-		if instance > lease.ToInstance() {
+		if currentNetwork != lease.Network {
+			// Lazily delete any lease that does not belong to network, likely acquired from
+			// prior manifests.
+			delete(l.leases, id)
+			log.Warnf("lost F3 participation lease for miner %d at instance %d due to network mismatch: %s != %s", id, instance, currentNetwork, lease.Network)
+		} else if instance > lease.ToInstance() {
 			// Lazily delete the expired leases.
 			delete(l.leases, id)
-			log.Warnf("lost F3 participation lease for miner %d", id)
+			log.Warnf("lost F3 participation lease for miner %d due to instance (%d) > lease to instance (%d)", id, instance, lease.ToInstance())
 		} else {
 			participants = append(participants, id)
 		}

chain/lf3/participation_lease_test.go

@@ -42,18 +42,18 @@ func TestLeaser(t *testing.T) {
 		require.NoError(t, err)
 
 		// Both participants should still be valid.
-		participants := subject.getParticipantsByInstance(11)
+		participants := subject.getParticipantsByInstance(testManifest.NetworkName, 11)
 		require.Len(t, participants, 2)
 		require.Contains(t, participants, uint64(123))
 		require.Contains(t, participants, uint64(456))
 
 		// After instance 16, only participant 456 should be valid.
-		participants = subject.getParticipantsByInstance(16)
+		participants = subject.getParticipantsByInstance(testManifest.NetworkName, 16)
 		require.Len(t, participants, 1)
 		require.Contains(t, participants, uint64(456))
 
 		// After instance 17, no participant must have a lease.
-		participants = subject.getParticipantsByInstance(17)
+		participants = subject.getParticipantsByInstance(testManifest.NetworkName, 17)
 		require.Empty(t, participants)
 	})
 	t.Run("expired ticket", func(t *testing.T) {

itests/f3_test.go

@@ -2,6 +2,7 @@ package itests
 
 import (
 	"context"
+	"sync"
 	"testing"
 	"time"
 
@@ -36,6 +37,7 @@ type testEnv struct {
 	m       *manifest.Manifest
 	t       *testing.T
 	testCtx context.Context
+	debug   bool
 }
 
 // Test that checks that F3 is enabled successfully,
@@ -194,6 +196,24 @@ func (e *testEnv) waitFor(f func(n *kit.TestFullNode) bool, timeout time.Duratio
 	e.t.Helper()
 	require.Eventually(e.t, func() bool {
 		e.t.Helper()
+		defer func() {
+			if e.debug {
+				var wg sync.WaitGroup
+				printProgress := func(index int, n *kit.TestFullNode) {
+					defer wg.Done()
+					if progress, err := n.F3GetProgress(e.testCtx); err != nil {
+						e.t.Logf("Node #%d progress: err: %v", index, err)
+					} else {
+						e.t.Logf("Node #%d progress: %v", index, progress)
+					}
+				}
+				for i, n := range e.minerFullNodes {
+					wg.Add(1)
+					go printProgress(i, n)
+				}
+				wg.Wait()
+			}
+		}()
 		for _, n := range e.minerFullNodes {
 			if !f(n) {
 				return false
@@ -210,6 +230,12 @@ func (e *testEnv) waitFor(f func(n *kit.TestFullNode) bool, timeout time.Duratio
 // a miner. The last return value is the manifest sender for the network.
 func setup(t *testing.T, blocktime time.Duration) *testEnv {
 	manif := lf3.NewManifest(BaseNetworkName+"/1", DefaultFinality, DefaultBootstrapEpoch, blocktime, cid.Undef)
+	manif.Gpbft.Delta = 250 * time.Millisecond
+	manif.Gpbft.DeltaBackOffExponent = 1.3
+	manif.Gpbft.RebroadcastBackoffBase = manif.Gpbft.Delta * 2
+	manif.Gpbft.RebroadcastBackoffMax = manif.Gpbft.RebroadcastBackoffBase * 2
+	manif.Gpbft.RebroadcastBackoffExponent = manif.Gpbft.DeltaBackOffExponent
+
 	return setupWithStaticManifest(t, manif, false)
 }
 
@@ -275,14 +301,23 @@ func setupWithStaticManifest(t *testing.T, manif *manifest.Manifest, testBootstr
 		err = n.NetConnect(ctx, e.ms.PeerInfo())
 		require.NoError(t, err)
 	}
-
 	errgrp.Go(func() error {
 		defer func() {
 			require.NoError(t, manifestServerHost.Close())
 		}()
 		return e.ms.Run(ctx)
 	})
 
+	// Update initial manifest params to shorten the timeouts and backoff for
+	// testing, and assert it is consistently applied to all nodes.
+	e.m.Gpbft.Delta = 250 * time.Millisecond
+	e.m.Gpbft.DeltaBackOffExponent = 1.3
+	e.m.Gpbft.RebroadcastBackoffBase = manif.Gpbft.Delta * 2
+	e.m.Gpbft.RebroadcastBackoffMax = manif.Gpbft.RebroadcastBackoffBase * 2
+	e.m.Gpbft.RebroadcastBackoffExponent = manif.Gpbft.DeltaBackOffExponent
+	e.ms.UpdateManifest(m)
+	e.waitTillManifestChange(m, 20*time.Second)
+
 	return e
 }