curio: Cleanup proving config (#11751)

curiosrc/builder.go

@@ -19,7 +19,7 @@ import (
 
 //var log = logging.Logger("provider")
 
-func WindowPostScheduler(ctx context.Context, fc config.CurioFees, pc config.ProvingConfig,
+func WindowPostScheduler(ctx context.Context, fc config.CurioFees, pc config.CurioProvingConfig,
 	api api.FullNode, verif storiface.Verifier, lw *sealer.LocalWorker, sender *message.Sender, chainSched *chainsched.CurioChainSched,
 	as *multictladdr.MultiAddressSelector, addresses map[dtypes.MinerAddress]bool, db *harmonydb.DB,
 	stor paths.Store, idx paths.SectorIndex, max int) (*window.WdPostTask, *window.WdPostSubmitTask, *window.WdPostRecoverDeclareTask, error) {

documentation/en/default-curio-config.toml

@@ -235,25 +235,6 @@
   # type: Duration
   #PartitionCheckTimeout = "20m0s"
 
-  # Disable Window PoSt computation on the lotus-miner process even if no window PoSt workers are present.
-  # 
-  # WARNING: If no windowPoSt workers are connected, window PoSt WILL FAIL resulting in faulty sectors which will need
-  # to be recovered. Before enabling this option, make sure your PoSt workers work correctly.
-  # 
-  # After changing this option, confirm that the new value works in your setup by invoking
-  # 'lotus-miner proving compute window-post 0'
-  #
-  # type: bool
-  #DisableBuiltinWindowPoSt = false
-
-  # Disable Winning PoSt computation on the lotus-miner process even if no winning PoSt workers are present.
-  # 
-  # WARNING: If no WinningPoSt workers are connected, Winning PoSt WILL FAIL resulting in lost block rewards.
-  # Before enabling this option, make sure your PoSt workers work correctly.
-  #
-  # type: bool
-  #DisableBuiltinWinningPoSt = false
-
   # Disable WindowPoSt provable sector readability checks.
   # 
   # In normal operation, when preparing to compute WindowPoSt, lotus-miner will perform a round of reading challenges

node/config/def.go

@@ -356,7 +356,7 @@ func DefaultCurioConfig() *CurioConfig {
 			TerminateControl: []string{},
 			MinerAddresses:   []string{},
 		}},
-		Proving: ProvingConfig{
+		Proving: CurioProvingConfig{
 			ParallelCheckLimit:    32,
 			PartitionCheckTimeout: Duration(20 * time.Minute),
 			SingleCheckTimeout:    Duration(10 * time.Minute),

node/config/doc_gen.go

@@ -228,7 +228,7 @@ over the worker address if this flag is set.`,
 		},
 		{
 			Name: "Proving",
-			Type: "ProvingConfig",
+			Type: "CurioProvingConfig",
 
 			Comment: ``,
 		},
@@ -295,6 +295,109 @@ over the worker address if this flag is set.`,
 			Comment: ``,
 		},
 	},
+	"CurioProvingConfig": {
+		{
+			Name: "ParallelCheckLimit",
+			Type: "int",
+
+			Comment: `Maximum number of sector checks to run in parallel. (0 = unlimited)
+
+WARNING: Setting this value too high may make the node crash by running out of stack
+WARNING: Setting this value too low may make sector challenge reading much slower, resulting in failed PoSt due
+to late submission.
+
+After changing this option, confirm that the new value works in your setup by invoking
+'lotus-miner proving compute window-post 0'`,
+		},
+		{
+			Name: "SingleCheckTimeout",
+			Type: "Duration",
+
+			Comment: `Maximum amount of time a proving pre-check can take for a sector. If the check times out the sector will be skipped
+
+WARNING: Setting this value too low risks in sectors being skipped even though they are accessible, just reading the
+test challenge took longer than this timeout
+WARNING: Setting this value too high risks missing PoSt deadline in case IO operations related to this sector are
+blocked (e.g. in case of disconnected NFS mount)`,
+		},
+		{
+			Name: "PartitionCheckTimeout",
+			Type: "Duration",
+
+			Comment: `Maximum amount of time a proving pre-check can take for an entire partition. If the check times out, sectors in
+the partition which didn't get checked on time will be skipped
+
+WARNING: Setting this value too low risks in sectors being skipped even though they are accessible, just reading the
+test challenge took longer than this timeout
+WARNING: Setting this value too high risks missing PoSt deadline in case IO operations related to this partition are
+blocked or slow`,
+		},
+		{
+			Name: "DisableWDPoStPreChecks",
+			Type: "bool",
+
+			Comment: `Disable WindowPoSt provable sector readability checks.
+
+In normal operation, when preparing to compute WindowPoSt, lotus-miner will perform a round of reading challenges
+from all sectors to confirm that those sectors can be proven. Challenges read in this process are discarded, as
+we're only interested in checking that sector data can be read.
+
+When using builtin proof computation (no PoSt workers, and DisableBuiltinWindowPoSt is set to false), this process
+can save a lot of time and compute resources in the case that some sectors are not readable - this is caused by
+the builtin logic not skipping snark computation when some sectors need to be skipped.
+
+When using PoSt workers, this process is mostly redundant, with PoSt workers challenges will be read once, and
+if challenges for some sectors aren't readable, those sectors will just get skipped.
+
+Disabling sector pre-checks will slightly reduce IO load when proving sectors, possibly resulting in shorter
+time to produce window PoSt. In setups with good IO capabilities the effect of this option on proving time should
+be negligible.
+
+NOTE: It likely is a bad idea to disable sector pre-checks in setups with no PoSt workers.
+
+NOTE: Even when this option is enabled, recovering sectors will be checked before recovery declaration message is
+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: "MaxPartitionsPerPoStMessage",
+			Type: "int",
+
+			Comment: `Maximum number of partitions to prove in a single SubmitWindowPoSt messace. 0 = network limit (3 in nv21)
+
+A single partition may contain up to 2349 32GiB sectors, or 2300 64GiB sectors.
+//
+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`,
+		},
+		{
+			Name: "MaxPartitionsPerRecoveryMessage",
+			Type: "int",
+
+			Comment: `In some cases when submitting DeclareFaultsRecovered messages,
+there may be too many recoveries to fit in a BlockGasLimit.
+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)`,
+		},
+	},
 	"CurioSubsystemsConfig": {
 		{
 			Name: "EnableWindowPost",

node/config/types.go

@@ -73,7 +73,7 @@ type CurioConfig struct {
 
 	// Addresses of wallets per MinerAddress (one of the fields).
 	Addresses []CurioAddresses
-	Proving   ProvingConfig
+	Proving   CurioProvingConfig
 	Journal   JournalConfig
 	Apis      ApisConfig
 }
@@ -723,6 +723,90 @@ type CurioAddresses struct {
 	MinerAddresses []string
 }
 
+type CurioProvingConfig struct {
+	// Maximum number of sector checks to run in parallel. (0 = unlimited)
+	//
+	// WARNING: Setting this value too high may make the node crash by running out of stack
+	// WARNING: Setting this value too low may make sector challenge reading much slower, resulting in failed PoSt due
+	// to late submission.
+	//
+	// After changing this option, confirm that the new value works in your setup by invoking
+	// 'lotus-miner proving compute window-post 0'
+	ParallelCheckLimit int
+
+	// Maximum amount of time a proving pre-check can take for a sector. If the check times out the sector will be skipped
+	//
+	// WARNING: Setting this value too low risks in sectors being skipped even though they are accessible, just reading the
+	// test challenge took longer than this timeout
+	// WARNING: Setting this value too high risks missing PoSt deadline in case IO operations related to this sector are
+	// blocked (e.g. in case of disconnected NFS mount)
+	SingleCheckTimeout Duration
+
+	// Maximum amount of time a proving pre-check can take for an entire partition. If the check times out, sectors in
+	// the partition which didn't get checked on time will be skipped
+	//
+	// WARNING: Setting this value too low risks in sectors being skipped even though they are accessible, just reading the
+	// test challenge took longer than this timeout
+	// WARNING: Setting this value too high risks missing PoSt deadline in case IO operations related to this partition are
+	// blocked or slow
+	PartitionCheckTimeout Duration
+
+	// Disable WindowPoSt provable sector readability checks.
+	//
+	// In normal operation, when preparing to compute WindowPoSt, lotus-miner will perform a round of reading challenges
+	// from all sectors to confirm that those sectors can be proven. Challenges read in this process are discarded, as
+	// we're only interested in checking that sector data can be read.
+	//
+	// When using builtin proof computation (no PoSt workers, and DisableBuiltinWindowPoSt is set to false), this process
+	// can save a lot of time and compute resources in the case that some sectors are not readable - this is caused by
+	// the builtin logic not skipping snark computation when some sectors need to be skipped.
+	//
+	// When using PoSt workers, this process is mostly redundant, with PoSt workers challenges will be read once, and
+	// if challenges for some sectors aren't readable, those sectors will just get skipped.
+	//
+	// Disabling sector pre-checks will slightly reduce IO load when proving sectors, possibly resulting in shorter
+	// time to produce window PoSt. In setups with good IO capabilities the effect of this option on proving time should
+	// be negligible.
+	//
+	// NOTE: It likely is a bad idea to disable sector pre-checks in setups with no PoSt workers.
+	//
+	// NOTE: Even when this option is enabled, recovering sectors will be checked before recovery declaration message is
+	// 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'
+	DisableWDPoStPreChecks bool
+
+	// Maximum number of partitions to prove in a single SubmitWindowPoSt messace. 0 = network limit (3 in nv21)
+	//
+	// A single partition may contain up to 2349 32GiB sectors, or 2300 64GiB sectors.
+	//	//
+	// 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
+	MaxPartitionsPerPoStMessage int
+
+	// Maximum number of partitions to declare in a single DeclareFaultsRecovered message. 0 = no limit.
+
+	// In some cases when submitting DeclareFaultsRecovered messages,
+	// there may be too many recoveries to fit in a BlockGasLimit.
+	// 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)
+	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
+}
+
 // API contains configs for API endpoint
 type API struct {
 	// Binding address for the Lotus API