Merge pull request #8982 from filecoin-project/feat/post-maxpart-config

feat: wdpost: Config for maximum partition count per message

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

@@ -375,6 +375,24 @@
   # env var: LOTUS_PROVING_DISABLEWDPOSTPRECHECKS
   #DisableWDPoStPreChecks = false
 
+  # Maximum number of partitions to prove in a single SubmitWindowPoSt messace. 0 = network limit (10 in nv16)
+  # 
+  # 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
+  # 
+  # In some cases when submitting PoSt messages which are recovering sectors, the default network limit may still be
+  # 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
+  #
+  # type: int
+  # env var: LOTUS_PROVING_MAXPARTITIONSPERMESSAGE
+  #MaxPartitionsPerMessage = 0
+
 
 [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

@@ -689,6 +689,24 @@ sent to the chain
 After changing this option, confirm that the new value works in your setup by invoking
 'lotus-miner proving compute window-post 0'`,
 		},
+		{
+			Name: "MaxPartitionsPerMessage",
+			Type: "int",
+
+			Comment: `Maximum number of partitions to prove in a single SubmitWindowPoSt messace. 0 = network limit (10 in nv16)
+
+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
+
+In some cases when submitting PoSt messages which are recovering sectors, the default network limit may still be
+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`,
+		},
 	},
 	"Pubsub": []DocField{
 		{

node/config/types.go

@@ -270,6 +270,21 @@ type ProvingConfig struct {
 	// After changing this option, confirm that the new value works in your setup by invoking
 	// 'lotus-miner proving compute window-post 0'
 	DisableWDPoStPreChecks bool
+
+	// Maximum number of partitions to prove in a single SubmitWindowPoSt messace. 0 = network limit (10 in nv16)
+	//
+	// 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
+	//
+	// In some cases when submitting PoSt messages which are recovering sectors, the default network limit may still be
+	// 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
+	MaxPartitionsPerMessage int
 }
 
 type SealingConfig struct {

storage/wdpost/wdpost_run.go

@@ -517,6 +517,13 @@ func (s *WindowPoStScheduler) batchPartitions(partitions []api.Partition, nv net
 		partitionsPerMsg = declMax
 	}
 
+	// respect user config if set
+	if s.maxPartitionsPerMessage > 0 {
+		if partitionsPerMsg > s.maxPartitionsPerMessage {
+			partitionsPerMsg = s.maxPartitionsPerMessage
+		}
+	}
+
 	// The number of messages will be:
 	// ceiling(number of partitions / partitions per message)
 	batchCount := len(partitions) / partitionsPerMsg

storage/wdpost/wdpost_run_test.go

@@ -268,6 +268,105 @@ func TestWDPostDoPost(t *testing.T) {
 	}
 }
 
+// TestWDPostDoPost verifies that doPost will send the correct number of window
+// PoST messages for a given number of partitions based on user config
+func TestWDPostDoPostPartLimitConfig(t *testing.T) {
+	//stm: @CHAIN_SYNCER_LOAD_GENESIS_001, @CHAIN_SYNCER_FETCH_TIPSET_001,
+	//stm: @CHAIN_SYNCER_START_001, @CHAIN_SYNCER_SYNC_001, @BLOCKCHAIN_BEACON_VALIDATE_BLOCK_VALUES_01
+	//stm: @CHAIN_SYNCER_COLLECT_CHAIN_001, @CHAIN_SYNCER_COLLECT_HEADERS_001, @CHAIN_SYNCER_VALIDATE_TIPSET_001
+	//stm: @CHAIN_SYNCER_NEW_PEER_HEAD_001, @CHAIN_SYNCER_VALIDATE_MESSAGE_META_001, @CHAIN_SYNCER_STOP_001
+	ctx := context.Background()
+	expectedMsgCount := 364
+
+	proofType := abi.RegisteredPoStProof_StackedDrgWindow2KiBV1
+	postAct := tutils.NewIDAddr(t, 100)
+
+	mockStgMinerAPI := newMockStorageMinerAPI()
+
+	// Get the number of sectors allowed in a partition for this proof type
+	sectorsPerPartition, err := builtin.PoStProofWindowPoStPartitionSectors(proofType)
+	require.NoError(t, err)
+	// Work out the number of partitions that can be included in a message
+	// without exceeding the message sector limit
+
+	//stm: @BLOCKCHAIN_POLICY_GET_MAX_POST_PARTITIONS_001
+	partitionsPerMsg, err := policy.GetMaxPoStPartitions(network.Version13, proofType)
+	require.NoError(t, err)
+	if partitionsPerMsg > minertypes.AddressedPartitionsMax {
+		partitionsPerMsg = minertypes.AddressedPartitionsMax
+	}
+
+	partitionCount := 4 * partitionsPerMsg
+
+	// Assert that user config is less than network limit
+	userPartLimit := 33
+	lastMsgParts := 21
+	require.Greater(t, partitionCount, userPartLimit)
+
+	// Assert that we consts are correct
+	require.Equal(t, (expectedMsgCount-1)*userPartLimit+lastMsgParts, 4*partitionsPerMsg)
+
+	var partitions []api.Partition
+	for p := 0; p < partitionCount; p++ {
+		sectors := bitfield.New()
+		for s := uint64(0); s < sectorsPerPartition; s++ {
+			sectors.Set(s)
+		}
+		partitions = append(partitions, api.Partition{
+			AllSectors:        sectors,
+			FaultySectors:     bitfield.New(),
+			RecoveringSectors: bitfield.New(),
+			LiveSectors:       sectors,
+			ActiveSectors:     sectors,
+		})
+	}
+	mockStgMinerAPI.setPartitions(partitions)
+
+	// Run window PoST
+	scheduler := &WindowPoStScheduler{
+		api:          mockStgMinerAPI,
+		prover:       &mockProver{},
+		verifier:     &mockVerif{},
+		faultTracker: &mockFaultTracker{},
+		proofType:    proofType,
+		actor:        postAct,
+		journal:      journal.NilJournal(),
+		addrSel:      &ctladdr.AddressSelector{},
+
+		maxPartitionsPerMessage: userPartLimit,
+	}
+
+	di := &dline.Info{
+		WPoStPeriodDeadlines:   minertypes.WPoStPeriodDeadlines,
+		WPoStProvingPeriod:     minertypes.WPoStProvingPeriod,
+		WPoStChallengeWindow:   minertypes.WPoStChallengeWindow,
+		WPoStChallengeLookback: minertypes.WPoStChallengeLookback,
+		FaultDeclarationCutoff: minertypes.FaultDeclarationCutoff,
+	}
+	ts := mockTipSet(t)
+
+	scheduler.startGeneratePoST(ctx, ts, di, func(posts []minertypes.SubmitWindowedPoStParams, err error) {
+		scheduler.startSubmitPoST(ctx, ts, di, posts, func(err error) {})
+	})
+
+	// Read the window PoST messages
+	for i := 0; i < expectedMsgCount; i++ {
+		msg := <-mockStgMinerAPI.pushedMessages
+		require.Equal(t, builtin.MethodsMiner.SubmitWindowedPoSt, msg.Method)
+		var params minertypes.SubmitWindowedPoStParams
+		err := params.UnmarshalCBOR(bytes.NewReader(msg.Params))
+		require.NoError(t, err)
+
+		if i == expectedMsgCount-1 {
+			// In the last message we only included a 21 partitions
+			require.Len(t, params.Partitions, lastMsgParts)
+		} else {
+			// All previous messages should include the full number of partitions
+			require.Len(t, params.Partitions, userPartLimit)
+		}
+	}
+}
+
 func mockTipSet(t *testing.T) *types.TipSet {
 	minerAct := tutils.NewActorAddr(t, "miner")
 	c, err := cid.Decode("QmbFMke1KXqnYyBBWxB74N4c5SBnJMVAiMNRcGu6x1AwQH")

storage/wdpost/wdpost_sched.go

@@ -62,16 +62,17 @@ type NodeAPI interface {
 // WindowPoStScheduler watches the chain though the changeHandler, which in turn
 // turn calls the scheduler when the time arrives to do work.
 type WindowPoStScheduler struct {
-	api              NodeAPI
-	feeCfg           config.MinerFeeConfig
-	addrSel          *ctladdr.AddressSelector
-	prover           storiface.ProverPoSt
-	verifier         storiface.Verifier
-	faultTracker     sealer.FaultTracker
-	proofType        abi.RegisteredPoStProof
-	partitionSectors uint64
-	disablePreChecks bool
-	ch               *changeHandler
+	api                     NodeAPI
+	feeCfg                  config.MinerFeeConfig
+	addrSel                 *ctladdr.AddressSelector
+	prover                  storiface.ProverPoSt
+	verifier                storiface.Verifier
+	faultTracker            sealer.FaultTracker
+	proofType               abi.RegisteredPoStProof
+	partitionSectors        uint64
+	disablePreChecks        bool
+	maxPartitionsPerMessage int
+	ch                      *changeHandler
 
 	actor address.Address
 
@@ -98,15 +99,16 @@ func NewWindowedPoStScheduler(api NodeAPI,
 	}
 
 	return &WindowPoStScheduler{
-		api:              api,
-		feeCfg:           cfg,
-		addrSel:          as,
-		prover:           sp,
-		verifier:         verif,
-		faultTracker:     ft,
-		proofType:        mi.WindowPoStProofType,
-		partitionSectors: mi.WindowPoStPartitionSectors,
-		disablePreChecks: pcfg.DisableWDPoStPreChecks,
+		api:                     api,
+		feeCfg:                  cfg,
+		addrSel:                 as,
+		prover:                  sp,
+		verifier:                verif,
+		faultTracker:            ft,
+		proofType:               mi.WindowPoStProofType,
+		partitionSectors:        mi.WindowPoStPartitionSectors,
+		disablePreChecks:        pcfg.DisableWDPoStPreChecks,
+		maxPartitionsPerMessage: pcfg.MaxPartitionsPerMessage,
 
 		actor: actor,
 		evtTypes: [...]journal.EventType{