Merge pull request #9427 from filecoin-project/sbansal/backport-9413

feat: wdpost: Add ability to only have single partition per msg for partitions with recovery sectors

documentation/en/default-lotus-miner-config.toml

@@ -380,11 +380,9 @@
   # A single partition may contain up to 2349 32GiB sectors, or 2300 64GiB sectors.
   # 
   # The maximum number of sectors which can be proven in a single PoSt message is 25000 in network version 16, which
-  # means that a single message can prove at most 10 partinions
+  # means that a single message can prove at most 10 partitions
   # 
-  # In some cases when submitting PoSt messages which are recovering sectors, the default network limit may still be
+  # Note that setting this value lower may result in less efficient gas use - more messages will be sent,
-  # too high to fit in the block gas limit; In those cases it may be necessary to set this value to something lower
-  # than 10; Note that setting this value lower may result in less efficient gas use - more messages will be sent,
   # to prove each deadline, resulting in more total gas use (but each message will have lower gas limit)
   # 
   # Setting this value above the network limit has no effect
@@ -403,6 +401,19 @@
   # env var: LOTUS_PROVING_MAXPARTITIONSPERRECOVERYMESSAGE
   #MaxPartitionsPerRecoveryMessage = 0
 
+  # Enable single partition per PoSt Message for partitions containing recovery sectors
+  # 
+  # In cases when submitting PoSt messages which contain recovering sectors, the default network limit may still be
+  # too high to fit in the block gas limit. In those cases, it becomes useful to only house the single partition
+  # with recovering sectors in the post message
+  # 
+  # Note that setting this value lower may result in less efficient gas use - more messages will be sent,
+  # to prove each deadline, resulting in more total gas use (but each message will have lower gas limit)
+  #
+  # type: bool
+  # env var: LOTUS_PROVING_SINGLERECOVERINGPARTITIONPERPOSTMESSAGE
+  #SingleRecoveringPartitionPerPostMessage = false
+
 
 [Sealing]
   # Upper bound on how many sectors can be waiting for more deals to be packed in it before it begins sealing at any given time.

node/config/doc_gen.go

@@ -698,11 +698,9 @@ After changing this option, confirm that the new value works in your setup by in
 A single partition may contain up to 2349 32GiB sectors, or 2300 64GiB sectors.
 
 The maximum number of sectors which can be proven in a single PoSt message is 25000 in network version 16, which
-means that a single message can prove at most 10 partinions
+means that a single message can prove at most 10 partitions
 
-In some cases when submitting PoSt messages which are recovering sectors, the default network limit may still be
+Note that setting this value lower may result in less efficient gas use - more messages will be sent,
-too high to fit in the block gas limit; In those cases it may be necessary to set this value to something lower
-than 10; Note that setting this value lower may result in less efficient gas use - more messages will be sent,
 to prove each deadline, resulting in more total gas use (but each message will have lower gas limit)
 
 Setting this value above the network limit has no effect`,
@@ -717,6 +715,19 @@ In those cases it may be necessary to set this value to something low (eg 1);
 Note that setting this value lower may result in less efficient gas use - more messages will be sent than needed,
 resulting in more total gas use (but each message will have lower gas limit)`,
 		},
+		{
+			Name: "SingleRecoveringPartitionPerPostMessage",
+			Type: "bool",
+
+			Comment: `Enable single partition per PoSt Message for partitions containing recovery sectors
+
+In cases when submitting PoSt messages which contain recovering sectors, the default network limit may still be
+too high to fit in the block gas limit. In those cases, it becomes useful to only house the single partition
+with recovering sectors in the post message
+
+Note that setting this value lower may result in less efficient gas use - more messages will be sent,
+to prove each deadline, resulting in more total gas use (but each message will have lower gas limit)`,
+		},
 	},
 	"Pubsub": []DocField{
 		{

node/config/types.go

@@ -276,11 +276,9 @@ type ProvingConfig struct {
 	// A single partition may contain up to 2349 32GiB sectors, or 2300 64GiB sectors.
 	//
 	// The maximum number of sectors which can be proven in a single PoSt message is 25000 in network version 16, which
-	// means that a single message can prove at most 10 partinions
+	// means that a single message can prove at most 10 partitions
 	//
-	// In some cases when submitting PoSt messages which are recovering sectors, the default network limit may still be
+	// Note that setting this value lower may result in less efficient gas use - more messages will be sent,
-	// too high to fit in the block gas limit; In those cases it may be necessary to set this value to something lower
-	// than 10; Note that setting this value lower may result in less efficient gas use - more messages will be sent,
 	// to prove each deadline, resulting in more total gas use (but each message will have lower gas limit)
 	//
 	// Setting this value above the network limit has no effect
@@ -294,6 +292,16 @@ type ProvingConfig struct {
 	// Note that setting this value lower may result in less efficient gas use - more messages will be sent than needed,
 	// resulting in more total gas use (but each message will have lower gas limit)
 	MaxPartitionsPerRecoveryMessage int
+
+	// Enable single partition per PoSt Message for partitions containing recovery sectors
+	//
+	// In cases when submitting PoSt messages which contain recovering sectors, the default network limit may still be
+	// too high to fit in the block gas limit. In those cases, it becomes useful to only house the single partition
+	// with recovering sectors in the post message
+	//
+	// Note that setting this value lower may result in less efficient gas use - more messages will be sent,
+	// to prove each deadline, resulting in more total gas use (but each message will have lower gas limit)
+	SingleRecoveringPartitionPerPostMessage bool
 }
 
 type SealingConfig struct {

storage/wdpost/wdpost_run.go

@@ -286,7 +286,7 @@ func (s *WindowPoStScheduler) runPoStCycle(ctx context.Context, manual bool, di
 
 	// Split partitions into batches, so as not to exceed the number of sectors
 	// allowed in a single message
-	partitionBatches, err := s.batchPartitions(partitions, nv)
+	partitionBatches, err := s.BatchPartitions(partitions, nv)
 	if err != nil {
 		return nil, err
 	}
@@ -492,7 +492,9 @@ func (s *WindowPoStScheduler) runPoStCycle(ctx context.Context, manual bool, di
 	return posts, nil
 }
 
-func (s *WindowPoStScheduler) batchPartitions(partitions []api.Partition, nv network.Version) ([][]api.Partition, error) {
+// Note: Partition order within batches must match original partition order in order
+// for code following the user code to work
+func (s *WindowPoStScheduler) BatchPartitions(partitions []api.Partition, nv network.Version) ([][]api.Partition, error) {
 	// We don't want to exceed the number of sectors allowed in a message.
 	// So given the number of sectors in a partition, work out the number of
 	// partitions that can be in a message without exceeding sectors per
@@ -524,21 +526,33 @@ func (s *WindowPoStScheduler) batchPartitions(partitions []api.Partition, nv net
 		}
 	}
 
-	// The number of messages will be:
+	batches := [][]api.Partition{}
-	// ceiling(number of partitions / partitions per message)
+
-	batchCount := len(partitions) / partitionsPerMsg
+	currBatch := []api.Partition{}
-	if len(partitions)%partitionsPerMsg != 0 {
+	for _, partition := range partitions {
-		batchCount++
+		recSectors, err := partition.RecoveringSectors.Count()
+		if err != nil {
+			return nil, err
 		}
 
-	// Split the partitions into batches
+		// Only add single partition to a batch if it contains recovery sectors
-	batches := make([][]api.Partition, 0, batchCount)
+		// and has the below user config set
-	for i := 0; i < len(partitions); i += partitionsPerMsg {
+		if s.singleRecoveringPartitionPerPostMessage && recSectors > 0 {
-		end := i + partitionsPerMsg
+			if len(currBatch) > 0 {
-		if end > len(partitions) {
+				batches = append(batches, currBatch)
-			end = len(partitions)
+				currBatch = []api.Partition{}
 			}
-		batches = append(batches, partitions[i:end])
+			batches = append(batches, []api.Partition{partition})
+		} else {
+			if len(currBatch) >= partitionsPerMsg {
+				batches = append(batches, currBatch)
+				currBatch = []api.Partition{}
+			}
+			currBatch = append(currBatch, partition)
+		}
+	}
+	if len(currBatch) > 0 {
+		batches = append(batches, currBatch)
 	}
 
 	return batches, nil

storage/wdpost/wdpost_run_test.go

@@ -177,6 +177,26 @@ func (m mockFaultTracker) CheckProvable(ctx context.Context, pp abi.RegisteredPo
 	return map[abi.SectorID]string{}, nil
 }
 
+func generatePartition(sectorCount uint64, recoverySectorCount uint64) api.Partition {
+	var partition api.Partition
+	sectors := bitfield.New()
+	recoverySectors := bitfield.New()
+	for s := uint64(0); s < sectorCount; s++ {
+		sectors.Set(s)
+	}
+	for s := uint64(0); s < recoverySectorCount; s++ {
+		recoverySectors.Set(s)
+	}
+	partition = api.Partition{
+		AllSectors:        sectors,
+		FaultySectors:     bitfield.New(),
+		RecoveringSectors: recoverySectors,
+		LiveSectors:       sectors,
+		ActiveSectors:     sectors,
+	}
+	return partition
+}
+
 // TestWDPostDoPost verifies that doPost will send the correct number of window
 // PoST messages for a given number of partitions
 func TestWDPostDoPost(t *testing.T) {
@@ -368,6 +388,55 @@ func TestWDPostDoPostPartLimitConfig(t *testing.T) {
 	}
 }
 
+// TestBatchPartitionsRecoverySectors tests if the batches with recovery sectors
+// contain only single partitions while keeping all the partitions in order
+func TestBatchPartitionsRecoverySectors(t *testing.T) {
+
+	proofType := abi.RegisteredPoStProof_StackedDrgWindow2KiBV1
+	postAct := tutils.NewIDAddr(t, 100)
+
+	mockStgMinerAPI := newMockStorageMinerAPI()
+
+	userPartLimit := 4
+
+	scheduler := &WindowPoStScheduler{
+		api:          mockStgMinerAPI,
+		prover:       &mockProver{},
+		verifier:     &mockVerif{},
+		faultTracker: &mockFaultTracker{},
+		proofType:    proofType,
+		actor:        postAct,
+		journal:      journal.NilJournal(),
+		addrSel:      &ctladdr.AddressSelector{},
+
+		maxPartitionsPerPostMessage:             userPartLimit,
+		singleRecoveringPartitionPerPostMessage: true,
+	}
+
+	var partitions []api.Partition
+	for p := 0; p < 4; p++ {
+		partitions = append(partitions, generatePartition(100, 0))
+	}
+	for p := 0; p < 2; p++ {
+		partitions = append(partitions, generatePartition(100, 10))
+	}
+	for p := 0; p < 6; p++ {
+		partitions = append(partitions, generatePartition(100, 0))
+	}
+	partitions = append(partitions, generatePartition(100, 10))
+
+	expectedBatchLens := []int{4, 1, 1, 4, 2, 1}
+
+	batches, err := scheduler.BatchPartitions(partitions, network.Version16)
+	require.NoError(t, err)
+
+	require.Equal(t, len(batches), 6)
+
+	for i, batch := range batches {
+		require.Equal(t, len(batch), expectedBatchLens[i])
+	}
+}
+
 // TestWDPostDeclareRecoveriesPartLimitConfig verifies that declareRecoveries will send the correct number of
 // DeclareFaultsRecovered messages for a given number of partitions based on user config
 func TestWDPostDeclareRecoveriesPartLimitConfig(t *testing.T) {

storage/wdpost/wdpost_sched.go

@@ -75,6 +75,7 @@ type WindowPoStScheduler struct {
 	disablePreChecks                        bool
 	maxPartitionsPerPostMessage             int
 	maxPartitionsPerRecoveryMessage         int
+	singleRecoveringPartitionPerPostMessage bool
 	ch                                      *changeHandler
 
 	actor address.Address
@@ -113,6 +114,7 @@ func NewWindowedPoStScheduler(api NodeAPI,
 		disablePreChecks:                        pcfg.DisableWDPoStPreChecks,
 		maxPartitionsPerPostMessage:             pcfg.MaxPartitionsPerPoStMessage,
 		maxPartitionsPerRecoveryMessage:         pcfg.MaxPartitionsPerRecoveryMessage,
+		singleRecoveringPartitionPerPostMessage: pcfg.SingleRecoveringPartitionPerPostMessage,
 		actor:                                   actor,
 		evtTypes: [...]journal.EventType{
 			evtTypeWdPoStScheduler:  j.RegisterEventType("wdpost", "scheduler"),