StateProofs: fix fast catchup bug (#4346)

* fix fast catchup bug

* fix partition issue

* fix CR comments.

catchup/catchpointService.go

@@ -31,6 +31,7 @@ import (
 	"github.com/algorand/go-algorand/ledger/ledgercore"
 	"github.com/algorand/go-algorand/logging"
 	"github.com/algorand/go-algorand/network"
+	"github.com/algorand/go-algorand/protocol"
 )
 
 const (
@@ -464,6 +465,27 @@ func (cs *CatchpointCatchupService) processStageLastestBlockDownload() (err erro
 	return nil
 }
 
+// lookbackForStateproofSupport calculates the lookback (from topblock round) needed to be downloaded
+// in order to support state proofs verification.
+func lookbackForStateproofsSupport(topBlock *bookkeeping.Block) uint64 {
+	proto := config.Consensus[topBlock.CurrentProtocol]
+	if proto.StateProofInterval == 0 {
+		return 0
+	}
+	//calculate the lowest round needed for verify the upcoming state proof txn.
+	expectedStateProofRound := topBlock.StateProofTracking[protocol.StateProofBasic].StateProofNextRound
+	stateproofVerificationRound := expectedStateProofRound.SubSaturate(basics.Round(proto.StateProofInterval))
+
+	lowestRecoveryRound := topBlock.Round().SubSaturate(topBlock.Round() % basics.Round(proto.StateProofInterval))
+	lowestRecoveryRound = lowestRecoveryRound.SubSaturate(basics.Round(proto.StateProofInterval * (proto.StateProofMaxRecoveryIntervals + 1)))
+
+	if lowestRecoveryRound > stateproofVerificationRound {
+		return uint64(topBlock.Round().SubSaturate(lowestRecoveryRound))
+	}
+
+	return uint64(topBlock.Round().SubSaturate(stateproofVerificationRound))
+}
+
 // processStageBlocksDownload is the fourth catchpoint catchup stage. It downloads all the reminder of the blocks, verifying each one of them against it's predecessor.
 func (cs *CatchpointCatchupService) processStageBlocksDownload() (err error) {
 	topBlock, err := cs.ledgerAccessor.EnsureFirstBlock(cs.ctx)
@@ -478,6 +500,11 @@ func (cs *CatchpointCatchupService) processStageBlocksDownload() (err error) {
 	if lookback < proto.MaxBalLookback {
 		lookback = proto.MaxBalLookback
 	}
+
+	lookbackForStateProofSupport := lookbackForStateproofsSupport(&topBlock)
+	if lookback < lookbackForStateProofSupport {
+		lookback = lookbackForStateProofSupport
+	}
 	// in case the effective lookback is going before our rounds count, trim it there.
 	// ( a catchpoint is generated starting round MaxBalLookback, and this is a possible in any round in the range of MaxBalLookback..MaxTxnLife)
 	if lookback >= uint64(topBlock.Round()) {

catchup/service_test.go

@@ -26,6 +26,7 @@ import (
 	"time"
 
 	"github.com/algorand/go-deadlock"
+	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 
 	"github.com/algorand/go-algorand/agreement"
@@ -956,7 +957,7 @@ func TestServiceStartStop(t *testing.T) {
 	s := MakeService(logging.Base(), cfg, &httpTestPeerSource{}, ledger, &mockedAuthenticator{errorRound: int(0 + 1)}, nil, nil)
 	s.Start()
 	s.Stop()
-	_, ok := (<-s.done)
+	_, ok := <-s.done
 	require.False(t, ok)
 }
 
@@ -972,3 +973,39 @@ func TestSynchronizingTime(t *testing.T) {
 	atomic.StoreInt64(&s.syncStartNS, 1000000)
 	require.NotEqual(t, time.Duration(0), s.SynchronizingTime())
 }
+
+func TestDownloadBlocksToSupportStateProofs(t *testing.T) {
+	partitiontest.PartitionTest(t)
+	// make sure we download enough blocks to verify state proof 512
+	topBlk := bookkeeping.Block{}
+	topBlk.BlockHeader.Round = 1500
+	topBlk.BlockHeader.CurrentProtocol = protocol.ConsensusFuture
+	trackingData := bookkeeping.StateProofTrackingData{StateProofNextRound: 512}
+	topBlk.BlockHeader.StateProofTracking = make(map[protocol.StateProofType]bookkeeping.StateProofTrackingData)
+	topBlk.BlockHeader.StateProofTracking[protocol.StateProofBasic] = trackingData
+
+	lookback := lookbackForStateproofsSupport(&topBlk)
+	oldestRound := topBlk.BlockHeader.Round.SubSaturate(basics.Round(lookback))
+	assert.Equal(t, uint64(oldestRound), 512-config.Consensus[protocol.ConsensusFuture].StateProofInterval)
+
+	// the network has made progress and now it is on round 8000. in this case we would not download blocks to cover 512.
+	// instead, we will download blocks to confirm only the recovery period lookback.
+	topBlk = bookkeeping.Block{}
+	topBlk.BlockHeader.Round = 8000
+	topBlk.BlockHeader.CurrentProtocol = protocol.ConsensusFuture
+	trackingData = bookkeeping.StateProofTrackingData{StateProofNextRound: 512}
+	topBlk.BlockHeader.StateProofTracking = make(map[protocol.StateProofType]bookkeeping.StateProofTrackingData)
+	topBlk.BlockHeader.StateProofTracking[protocol.StateProofBasic] = trackingData
+
+	lookback = lookbackForStateproofsSupport(&topBlk)
+	oldestRound = topBlk.BlockHeader.Round.SubSaturate(basics.Round(lookback))
+
+	lowestRoundToRetain := 8000 - (8000 % 256) - (config.Consensus[protocol.ConsensusFuture].StateProofInterval * (config.Consensus[protocol.ConsensusFuture].StateProofMaxRecoveryIntervals + 1))
+	assert.Equal(t, uint64(oldestRound), lowestRoundToRetain)
+
+	topBlk = bookkeeping.Block{}
+	topBlk.BlockHeader.Round = 8000
+	topBlk.BlockHeader.CurrentProtocol = protocol.ConsensusV32
+	lookback = lookbackForStateproofsSupport(&topBlk)
+	assert.Equal(t, uint64(0), lookback)
+}