fsm: Call ReleaseSectorKey after WinningPoSt lookback period

This commit is contained in:
Łukasz Magiera 2022-02-09 12:41:39 +01:00
parent 67e3f182f5
commit 947000c94f
6 changed files with 124 additions and 2 deletions

View File

@ -466,6 +466,7 @@ var stateOrder = map[sealing.SectorState]stateMeta{}
var stateList = []stateMeta{
{col: 39, state: "Total"},
{col: color.FgGreen, state: sealing.Proving},
{col: color.FgGreen, state: sealing.UpdateActivating},
{col: color.FgBlue, state: sealing.Empty},
{col: color.FgBlue, state: sealing.WaitDeals},
@ -496,6 +497,7 @@ var stateList = []stateMeta{
{col: color.FgYellow, state: sealing.SubmitReplicaUpdate},
{col: color.FgYellow, state: sealing.ReplicaUpdateWait},
{col: color.FgYellow, state: sealing.FinalizeReplicaUpdate},
{col: color.FgYellow, state: sealing.ReleaseSectorKey},
{col: color.FgCyan, state: sealing.Terminating},
{col: color.FgCyan, state: sealing.TerminateWait},
@ -524,6 +526,7 @@ var stateList = []stateMeta{
{col: color.FgRed, state: sealing.SnapDealsAddPieceFailed},
{col: color.FgRed, state: sealing.SnapDealsDealsExpired},
{col: color.FgRed, state: sealing.ReplicaUpdateFailed},
{col: color.FgRed, state: sealing.ReleaseSectorKeyFailed},
}
func init() {

View File

@ -169,7 +169,14 @@ var fsmPlanners = map[SectorState]func(events []statemachine.Event, state *Secto
on(SectorAbortUpgrade{}, AbortUpgrade),
),
FinalizeReplicaUpdate: planOne(
on(SectorFinalized{}, Proving),
on(SectorFinalized{}, UpdateActivating),
),
UpdateActivating: planOne(
on(SectorUpdateActive{}, ReleaseSectorKey),
),
ReleaseSectorKey: planOne(
on(SectorKeyReleased{}, Proving),
on(SectorReleaseKeyFailed{}, ReleaseSectorKeyFailed),
),
// Sealing errors
@ -250,6 +257,9 @@ var fsmPlanners = map[SectorState]func(events []statemachine.Event, state *Secto
on(SectorInvalidDealIDs{}, SnapDealsRecoverDealIDs),
on(SectorDealsExpired{}, SnapDealsDealsExpired),
),
ReleaseSectorKeyFailed: planOne(
on(SectorUpdateActive{}, ReleaseSectorKey),
),
// Post-seal
@ -477,6 +487,10 @@ func (m *Sealing) plan(events []statemachine.Event, state *SectorInfo) (func(sta
return m.handleReplicaUpdateWait, processed, nil
case FinalizeReplicaUpdate:
return m.handleFinalizeReplicaUpdate, processed, nil
case UpdateActivating:
return m.handleUpdateActivating, processed, nil
case ReleaseSectorKey:
return m.handleReleaseSectorKey, processed, nil
// Handled failure modes
case AddPieceFailed:
@ -513,6 +527,8 @@ func (m *Sealing) plan(events []statemachine.Event, state *SectorInfo) (func(sta
return m.handleSnapDealsRecoverDealIDs, processed, nil
case ReplicaUpdateFailed:
return m.handleSubmitReplicaUpdateFailed, processed, nil
case ReleaseSectorKeyFailed:
return m.handleReleaseSectorKeyFailed, 0, err
case AbortUpgrade:
return m.handleAbortUpgrade, processed, nil

View File

@ -335,6 +335,14 @@ type SectorReplicaUpdateLanded struct{}
func (evt SectorReplicaUpdateLanded) apply(state *SectorInfo) {}
type SectorUpdateActive struct{}
func (evt SectorUpdateActive) apply(state *SectorInfo) {}
type SectorKeyReleased struct{}
func (evt SectorKeyReleased) apply(state *SectorInfo) {}
// Failed state recovery
type SectorRetrySealPreCommit1 struct{}
@ -445,6 +453,13 @@ type SectorSubmitReplicaUpdateFailed struct{}
func (evt SectorSubmitReplicaUpdateFailed) apply(state *SectorInfo) {}
type SectorReleaseKeyFailed struct{ error }
func (evt SectorReleaseKeyFailed) FormatError(xerrors.Printer) (next error) {
return evt.error
}
func (evt SectorReleaseKeyFailed) apply(state *SectorInfo) {}
// Faults
type SectorFaulty struct{}

View File

@ -52,11 +52,14 @@ var ExistSectorStateList = map[SectorState]struct{}{
ProveReplicaUpdate: {},
SubmitReplicaUpdate: {},
ReplicaUpdateWait: {},
UpdateActivating: {},
ReleaseSectorKey: {},
FinalizeReplicaUpdate: {},
SnapDealsAddPieceFailed: {},
SnapDealsDealsExpired: {},
SnapDealsRecoverDealIDs: {},
ReplicaUpdateFailed: {},
ReleaseSectorKeyFailed: {},
AbortUpgrade: {},
}
@ -104,6 +107,8 @@ const (
SubmitReplicaUpdate SectorState = "SubmitReplicaUpdate"
ReplicaUpdateWait SectorState = "ReplicaUpdateWait"
FinalizeReplicaUpdate SectorState = "FinalizeReplicaUpdate"
UpdateActivating SectorState = "UpdateActivating"
ReleaseSectorKey SectorState = "ReleaseSectorKey"
// error modes
FailedUnrecoverable SectorState = "FailedUnrecoverable"
@ -124,6 +129,7 @@ const (
SnapDealsRecoverDealIDs SectorState = "SnapDealsRecoverDealIDs"
AbortUpgrade SectorState = "AbortUpgrade"
ReplicaUpdateFailed SectorState = "ReplicaUpdateFailed"
ReleaseSectorKeyFailed SectorState = "ReleaseSectorKeyFailed"
Faulty SectorState = "Faulty" // sector is corrupted or gone for some reason
FaultReported SectorState = "FaultReported" // sector has been declared as a fault on chain
@ -153,7 +159,7 @@ func toStatState(st SectorState, finEarly bool) statSectorState {
return sstProving
}
return sstSealing
case Proving, Removed, Removing, Terminating, TerminateWait, TerminateFinality, TerminateFailed:
case Proving, UpdateActivating, ReleaseSectorKey, Removed, Removing, Terminating, TerminateWait, TerminateFinality, TerminateFailed:
return sstProving
}

View File

@ -244,6 +244,16 @@ func (m *Sealing) handleSubmitReplicaUpdateFailed(ctx statemachine.Context, sect
return ctx.Send(SectorRetrySubmitReplicaUpdate{})
}
func (m *Sealing) handleReleaseSectorKeyFailed(ctx statemachine.Context, sector SectorInfo) error {
// not much we can do, wait for a bit and try again
if err := failedCooldown(ctx, sector); err != nil {
return err
}
return ctx.Send(SectorUpdateActive{})
}
func (m *Sealing) handleCommitFailed(ctx statemachine.Context, sector SectorInfo) error {
tok, _, err := m.Api.ChainHead(ctx.Context())
if err != nil {

View File

@ -2,12 +2,15 @@ package sealing
import (
"bytes"
"context"
"time"
"github.com/filecoin-project/go-state-types/big"
"github.com/filecoin-project/go-state-types/exitcode"
statemachine "github.com/filecoin-project/go-statemachine"
api "github.com/filecoin-project/lotus/api"
"github.com/filecoin-project/lotus/chain/actors/builtin/miner"
"github.com/filecoin-project/lotus/chain/actors/policy"
"golang.org/x/xerrors"
)
@ -215,6 +218,75 @@ func (m *Sealing) handleFinalizeReplicaUpdate(ctx statemachine.Context, sector S
return ctx.Send(SectorFinalized{})
}
func (m *Sealing) handleUpdateActivating(ctx statemachine.Context, sector SectorInfo) error {
try := func() error {
mw, err := m.Api.StateWaitMsg(ctx.Context(), *sector.ReplicaUpdateMessage)
if err != nil {
return err
}
tok, height, err := m.Api.ChainHead(ctx.Context())
if err != nil {
return err
}
nv, err := m.Api.StateNetworkVersion(ctx.Context(), tok)
if err != nil {
return err
}
lb := policy.GetWinningPoStSectorSetLookback(nv)
targetHeight := mw.Height + lb + InteractivePoRepConfidence
delay := 50 * time.Millisecond
for {
if height >= targetHeight {
break
}
wctx, cancel := context.WithTimeout(ctx.Context(), delay)
<-wctx.Done()
cancel()
// increasing backoff; can't just calculate the correct time because integration tests do funny things with time
delay = delay * 10 / 3
if delay > 5*time.Minute {
delay = 5 * time.Minute
}
_, height, err = m.Api.ChainHead(ctx.Context())
if err != nil {
return err
}
}
return nil
}
for {
err := try()
if err == nil {
break
}
log.Errorw("error in handleUpdateActivating", "error", err)
// likely an API issue, sleep for a bit and retry
time.Sleep(time.Minute)
}
return ctx.Send(SectorUpdateActive{})
}
func (m *Sealing) handleReleaseSectorKey(ctx statemachine.Context, sector SectorInfo) error {
if err := m.sealer.ReleaseSectorKey(sector.sealingCtx(ctx.Context()), m.minerSector(sector.SectorType, sector.SectorNumber)); err != nil {
return ctx.Send(SectorReleaseKeyFailed{err})
}
return ctx.Send(SectorKeyReleased{})
}
func handleErrors(ctx statemachine.Context, err error, sector SectorInfo) error {
switch err.(type) {
case *ErrApi: