Merge remote-tracking branch 'origin/main' into hm/snapshot-export

Author: hanabi1224

README.md

@@ -24,10 +24,9 @@ consensus protocol to finalize tipsets when voted for by more than two-thirds of
 
 ## Status
 
-**:test_tube: Under Passive Testing**
+**🚀 Live on Mainnet**
 
-The implementation of Go-F3 is complete and is currently undergoing passive testing in nv23. It has also been integrated
-into Lotus.
+Go-F3 was successfully activated on Filecoin mainnet on April 29, 2025 on epoch 4920480.
 
 ## Project Structure
 

bootstrap_delay_test.go

@@ -64,7 +64,7 @@ func TestComputeBootstrapDelay(t *testing.T) {
 			name: "out of sync - way after bootstrap",
 			time: genesis.Add(time.Duration(bootstrapEpoch+100)*period + 1*time.Second),
 			ts:   tipset{genesis: genesis, epoch: int64(bootstrapEpoch - 100), period: period},
-			want: 0 * time.Second,
+			want: 1 * time.Nanosecond, // we don't start immediately as the tipset we need is not available yet
 		},
 		{
 			name: "out of sync - way before bootstrap",

f3.go

@@ -136,6 +136,9 @@ func (m *F3) GetPowerTableByInstance(ctx context.Context, instance uint64) (gpbf
 
 // computeBootstrapDelay returns the time at which the F3 instance specified by
 // the passed manifest should be started.
+// It will return 0 if the manifest bootstrap epoch is greater than the current epoch.
+// It will also return 1ns if the manifest bootstrap epoch is equal to the current epoch but by
+// the time calculation, we should have already received the bootstrap tipset.
 func computeBootstrapDelay(ts ec.TipSet, clock clock.Clock, mfst manifest.Manifest) time.Duration {
 	currentEpoch := ts.Epoch()
 	if currentEpoch >= mfst.BootstrapEpoch {
@@ -144,7 +147,9 @@ func computeBootstrapDelay(ts ec.TipSet, clock clock.Clock, mfst manifest.Manife
 	epochDelay := mfst.BootstrapEpoch - currentEpoch
 	start := ts.Timestamp().Add(time.Duration(epochDelay) * mfst.EC.Period)
 	delay := clock.Until(start)
-	delay = max(delay, 0)
+	// ensure that we don't start immediately
+	// to trigger waiting for the bootstrap tipset to exist
+	delay = max(delay, 1*time.Nanosecond)
 	return delay
 }
 
@@ -185,6 +190,8 @@ func (m *F3) Start(startCtx context.Context) (_err error) {
 				delay := computeBootstrapDelay(ts, m.clock, m.mfst)
 				if delay > 0 {
 					log.Infow("waiting for bootstrap epoch", "duration", delay.String())
+					// reduce hot-looping by waiting for at least 20ms
+					delay = max(delay, 20*time.Millisecond)
 					startTimer.Reset(delay)
 				} else {
 					err = m.startInternal(m.runningCtx)
@@ -201,9 +208,9 @@ func (m *F3) Start(startCtx context.Context) (_err error) {
 }
 
 // Stop F3.
-func (m *F3) Stop(context.Context) (_err error) {
+func (m *F3) Stop(ctx context.Context) (_err error) {
 	m.cancelCtx()
-	return nil
+	return m.stopInternal(ctx)
 }
 
 func (s *f3State) stop(ctx context.Context) (err error) {

f3_helpers_test.go

@@ -94,7 +94,7 @@ func TestF3WithLookback(t *testing.T) {
 	env.requireInstanceEventually(5, eventualCheckTimeout, true)
 }
 
-func TestF3PauseResumeCatchup(t *testing.T) {
+func TestF3StopStartCatchup(t *testing.T) {
 	// Quiet down the logs since the test asserts a scenario that triggers
 	// OhShitStore ERROR level logs.
 	_ = logging.SetLogLevel("f3/ohshitstore", "DPANIC")
@@ -104,8 +104,8 @@ func TestF3PauseResumeCatchup(t *testing.T) {
 	env.requireEpochFinalizedEventually(env.manifest.BootstrapEpoch, eventualCheckTimeout)
 
 	// Pausing two nodes should pause the network.
-	env.pauseNode(1)
-	env.pauseNode(2)
+	env.stopNode(1)
+	env.stopNode(2)
 
 	env.requireF3NotRunningEventually(eventualCheckTimeout, nodeMatchers.byID(1, 2))
 
@@ -118,22 +118,23 @@ func TestF3PauseResumeCatchup(t *testing.T) {
 	}, eventualCheckTimeout, eventualCheckInterval)
 
 	// Resuming node 1 should continue agreeing on instances.
-	env.resumeNode(1)
+	env.startNode(1)
+	env.connectAll()
 	env.requireF3RunningEventually(eventualCheckTimeout, nodeMatchers.byID(1))
 
 	// Wait until we're far enough that pure GPBFT catchup should be impossible.
 	targetInstance := env.nodes[1].currentGpbftInstance() + env.manifest.CommitteeLookback + 1
 	env.requireInstanceEventually(targetInstance, eventualCheckTimeout, false)
 
-	env.resumeNode(2)
+	env.startNode(2)
 	env.requireF3RunningEventually(eventualCheckTimeout, nodeMatchers.byID(2))
 
 	// Everyone should catch up eventually
 	env.requireInstanceEventually(targetInstance, eventualCheckTimeout, true)
 
 	// Pause the "good" node.
 	node0failInstance := env.nodes[0].currentGpbftInstance()
-	env.pauseNode(0)
+	env.stopNode(0)
 	env.requireF3NotRunningEventually(eventualCheckTimeout, nodeMatchers.byID(0))
 
 	// We should be able to make progress with the remaining nodes.
@@ -337,12 +338,12 @@ func (n *testNode) init() *f3.F3 {
 	return n.f3
 }
 
-func (n *testNode) pause() {
-	require.NoError(n.e.t, n.f3.Pause(n.e.testCtx))
+func (n *testNode) stop() {
+	require.NoError(n.e.t, n.f3.Stop(n.e.testCtx))
 }
 
-func (n *testNode) resume() {
-	require.NoError(n.e.t, n.f3.Resume(n.e.testCtx))
+func (n *testNode) start() {
+	require.NoError(n.e.t, n.f3.Start(n.e.testCtx))
 }
 
 type testNodeStatus struct {
@@ -629,12 +630,12 @@ func (e *testEnv) withManifest(m manifest.Manifest) *testEnv {
 	return e
 }
 
-func (e *testEnv) pauseNode(i int) {
-	e.nodes[i].pause()
+func (e *testEnv) stopNode(i int) {
+	e.nodes[i].stop()
 }
 
-func (e *testEnv) resumeNode(i int) {
-	e.nodes[i].resume()
+func (e *testEnv) startNode(i int) {
+	e.nodes[i].start()
 }
 
 func (e *testEnv) injectDatastoreFailures(i int, fn func(op string) error) {

f3_test.go

@@ -1,3 +1,3 @@
 {
-  "version": "v0.8.6"
+  "version": "v0.8.7"
 }