lotus/extern/sector-storage/manager.go
2021-07-12 11:30:26 +02:00

726 lines
22 KiB
Go

package sectorstorage
import (
"context"
"errors"
"io"
"net/http"
"sync"
"github.com/google/uuid"
"github.com/hashicorp/go-multierror"
"github.com/ipfs/go-cid"
logging "github.com/ipfs/go-log/v2"
"github.com/mitchellh/go-homedir"
"golang.org/x/xerrors"
"github.com/filecoin-project/go-state-types/abi"
"github.com/filecoin-project/go-statestore"
"github.com/filecoin-project/specs-storage/storage"
"github.com/filecoin-project/lotus/extern/sector-storage/ffiwrapper"
"github.com/filecoin-project/lotus/extern/sector-storage/fsutil"
"github.com/filecoin-project/lotus/extern/sector-storage/sealtasks"
"github.com/filecoin-project/lotus/extern/sector-storage/stores"
"github.com/filecoin-project/lotus/extern/sector-storage/storiface"
)
var log = logging.Logger("advmgr")
var ErrNoWorkers = errors.New("no suitable workers found")
type Worker interface {
storiface.WorkerCalls
TaskTypes(context.Context) (map[sealtasks.TaskType]struct{}, error)
// Returns paths accessible to the worker
Paths(context.Context) ([]stores.StoragePath, error)
Info(context.Context) (storiface.WorkerInfo, error)
Session(context.Context) (uuid.UUID, error)
Close() error // TODO: do we need this?
}
type SectorManager interface {
ffiwrapper.StorageSealer
storage.Prover
storiface.WorkerReturn
FaultTracker
}
type WorkerID uuid.UUID // worker session UUID
var ClosedWorkerID = uuid.UUID{}
func (w WorkerID) String() string {
return uuid.UUID(w).String()
}
type Manager struct {
ls stores.LocalStorage
storage *stores.Remote
localStore *stores.Local
remoteHnd *stores.FetchHandler
index stores.SectorIndex
sched *scheduler
storage.Prover
workLk sync.Mutex
work *statestore.StateStore
callToWork map[storiface.CallID]WorkID
// used when we get an early return and there's no callToWork mapping
callRes map[storiface.CallID]chan result
results map[WorkID]result
waitRes map[WorkID]chan struct{}
}
type result struct {
r interface{}
err error
}
// ResourceFilteringStrategy is an enum indicating the kinds of resource
// filtering strategies that can be configured for workers.
type ResourceFilteringStrategy string
const (
// ResourceFilteringHardware specifies that available hardware resources
// should be evaluated when scheduling a task against the worker.
ResourceFilteringHardware = ResourceFilteringStrategy("hardware")
// ResourceFilteringDisabled disables resource filtering against this
// worker. The scheduler may assign any task to this worker.
ResourceFilteringDisabled = ResourceFilteringStrategy("disabled")
)
type SealerConfig struct {
ParallelFetchLimit int
// Local worker config
AllowAddPiece bool
AllowPreCommit1 bool
AllowPreCommit2 bool
AllowCommit bool
AllowUnseal bool
// ResourceFiltering instructs the system which resource filtering strategy
// to use when evaluating tasks against this worker. An empty value defaults
// to "hardware".
ResourceFiltering ResourceFilteringStrategy
}
type StorageAuth http.Header
type WorkerStateStore *statestore.StateStore
type ManagerStateStore *statestore.StateStore
func New(ctx context.Context, lstor *stores.Local, stor *stores.Remote, ls stores.LocalStorage, si stores.SectorIndex, sc SealerConfig, wss WorkerStateStore, mss ManagerStateStore) (*Manager, error) {
prover, err := ffiwrapper.New(&readonlyProvider{stor: lstor, index: si})
if err != nil {
return nil, xerrors.Errorf("creating prover instance: %w", err)
}
m := &Manager{
ls: ls,
storage: stor,
localStore: lstor,
remoteHnd: &stores.FetchHandler{Local: lstor, PfHandler: &stores.DefaultPartialFileHandler{}},
index: si,
sched: newScheduler(),
Prover: prover,
work: mss,
callToWork: map[storiface.CallID]WorkID{},
callRes: map[storiface.CallID]chan result{},
results: map[WorkID]result{},
waitRes: map[WorkID]chan struct{}{},
}
m.setupWorkTracker()
go m.sched.runSched()
localTasks := []sealtasks.TaskType{
sealtasks.TTCommit1, sealtasks.TTFinalize, sealtasks.TTFetch,
}
if sc.AllowAddPiece {
localTasks = append(localTasks, sealtasks.TTAddPiece)
}
if sc.AllowPreCommit1 {
localTasks = append(localTasks, sealtasks.TTPreCommit1)
}
if sc.AllowPreCommit2 {
localTasks = append(localTasks, sealtasks.TTPreCommit2)
}
if sc.AllowCommit {
localTasks = append(localTasks, sealtasks.TTCommit2)
}
if sc.AllowUnseal {
localTasks = append(localTasks, sealtasks.TTUnseal)
}
wcfg := WorkerConfig{
IgnoreResourceFiltering: sc.ResourceFiltering == ResourceFilteringDisabled,
TaskTypes: localTasks,
}
worker := NewLocalWorker(wcfg, stor, lstor, si, m, wss)
err = m.AddWorker(ctx, worker)
if err != nil {
return nil, xerrors.Errorf("adding local worker: %w", err)
}
return m, nil
}
func (m *Manager) AddLocalStorage(ctx context.Context, path string) error {
path, err := homedir.Expand(path)
if err != nil {
return xerrors.Errorf("expanding local path: %w", err)
}
if err := m.localStore.OpenPath(ctx, path); err != nil {
return xerrors.Errorf("opening local path: %w", err)
}
if err := m.ls.SetStorage(func(sc *stores.StorageConfig) {
sc.StoragePaths = append(sc.StoragePaths, stores.LocalPath{Path: path})
}); err != nil {
return xerrors.Errorf("get storage config: %w", err)
}
return nil
}
func (m *Manager) AddWorker(ctx context.Context, w Worker) error {
return m.sched.runWorker(ctx, w)
}
func (m *Manager) ServeHTTP(w http.ResponseWriter, r *http.Request) {
m.remoteHnd.ServeHTTP(w, r)
}
func schedNop(context.Context, Worker) error {
return nil
}
func (m *Manager) schedFetch(sector storage.SectorRef, ft storiface.SectorFileType, ptype storiface.PathType, am storiface.AcquireMode) func(context.Context, Worker) error {
return func(ctx context.Context, worker Worker) error {
_, err := m.waitSimpleCall(ctx)(worker.Fetch(ctx, sector, ft, ptype, am))
return err
}
}
// SectorsUnsealPiece will Unseal the Sealed sector file for the given sector.
// It will schedule the Unsealing task on a worker that either already has the sealed sector files or has space in
// one of it's sealing scratch spaces to store them after fetching them from another worker.
// If the chosen worker already has the Unsealed sector file, we will NOT Unseal the sealed sector file again.
func (m *Manager) SectorsUnsealPiece(ctx context.Context, sector storage.SectorRef, offset storiface.UnpaddedByteIndex, size abi.UnpaddedPieceSize, ticket abi.SealRandomness, unsealed *cid.Cid) error {
ctx, cancel := context.WithCancel(ctx)
defer cancel()
log.Debugf("acquire unseal sector lock for sector %d", sector.ID)
if err := m.index.StorageLock(ctx, sector.ID, storiface.FTSealed|storiface.FTCache, storiface.FTUnsealed); err != nil {
return xerrors.Errorf("acquiring unseal sector lock: %w", err)
}
// if the selected worker does NOT have the sealed files for the sector, instruct it to fetch it from a worker that has them and
// put it in the sealing scratch space.
sealFetch := func(ctx context.Context, worker Worker) error {
log.Debugf("copy sealed/cache sector data for sector %d", sector.ID)
if _, err := m.waitSimpleCall(ctx)(worker.Fetch(ctx, sector, storiface.FTSealed|storiface.FTCache, storiface.PathSealing, storiface.AcquireCopy)); err != nil {
return xerrors.Errorf("copy sealed/cache sector data: %w", err)
}
return nil
}
if unsealed == nil {
return xerrors.Errorf("cannot unseal piece (sector: %d, offset: %d size: %d) - unsealed cid is undefined", sector, offset, size)
}
ssize, err := sector.ProofType.SectorSize()
if err != nil {
return xerrors.Errorf("getting sector size: %w", err)
}
// selector will schedule the Unseal task on a worker that either already has the sealed sector files or has space in
// one of it's sealing scratch spaces to store them after fetching them from another worker.
selector := newExistingSelector(m.index, sector.ID, storiface.FTSealed|storiface.FTCache, true)
log.Debugf("will schedule unseal for sector %d", sector.ID)
err = m.sched.Schedule(ctx, sector, sealtasks.TTUnseal, selector, sealFetch, func(ctx context.Context, w Worker) error {
// TODO: make restartable
// NOTE: we're unsealing the whole sector here as with SDR we can't really
// unseal the sector partially. Requesting the whole sector here can
// save us some work in case another piece is requested from here
log.Debugf("calling unseal sector on worker, sectoID=%d", sector.ID)
// Note: This unseal piece call will essentially become a no-op if the worker already has an Unsealed sector file for the given sector.
_, err := m.waitSimpleCall(ctx)(w.UnsealPiece(ctx, sector, 0, abi.PaddedPieceSize(ssize).Unpadded(), ticket, *unsealed))
log.Debugf("completed unseal sector %d", sector.ID)
return err
})
if err != nil {
return xerrors.Errorf("worker UnsealPiece call: %s", err)
}
return nil
}
func (m *Manager) NewSector(ctx context.Context, sector storage.SectorRef) error {
log.Warnf("stub NewSector")
return nil
}
func (m *Manager) AddPiece(ctx context.Context, sector storage.SectorRef, existingPieces []abi.UnpaddedPieceSize, sz abi.UnpaddedPieceSize, r io.Reader) (abi.PieceInfo, error) {
ctx, cancel := context.WithCancel(ctx)
defer cancel()
if err := m.index.StorageLock(ctx, sector.ID, storiface.FTNone, storiface.FTUnsealed); err != nil {
return abi.PieceInfo{}, xerrors.Errorf("acquiring sector lock: %w", err)
}
var selector WorkerSelector
var err error
if len(existingPieces) == 0 { // new
selector = newAllocSelector(m.index, storiface.FTUnsealed, storiface.PathSealing)
} else { // use existing
selector = newExistingSelector(m.index, sector.ID, storiface.FTUnsealed, false)
}
var out abi.PieceInfo
err = m.sched.Schedule(ctx, sector, sealtasks.TTAddPiece, selector, schedNop, func(ctx context.Context, w Worker) error {
p, err := m.waitSimpleCall(ctx)(w.AddPiece(ctx, sector, existingPieces, sz, r))
if err != nil {
return err
}
if p != nil {
out = p.(abi.PieceInfo)
}
return nil
})
return out, err
}
func (m *Manager) SealPreCommit1(ctx context.Context, sector storage.SectorRef, ticket abi.SealRandomness, pieces []abi.PieceInfo) (out storage.PreCommit1Out, err error) {
ctx, cancel := context.WithCancel(ctx)
defer cancel()
wk, wait, cancel, err := m.getWork(ctx, sealtasks.TTPreCommit1, sector, ticket, pieces)
if err != nil {
return nil, xerrors.Errorf("getWork: %w", err)
}
defer cancel()
var waitErr error
waitRes := func() {
p, werr := m.waitWork(ctx, wk)
if werr != nil {
waitErr = werr
return
}
if p != nil {
out = p.(storage.PreCommit1Out)
}
}
if wait { // already in progress
waitRes()
return out, waitErr
}
if err := m.index.StorageLock(ctx, sector.ID, storiface.FTUnsealed, storiface.FTSealed|storiface.FTCache); err != nil {
return nil, xerrors.Errorf("acquiring sector lock: %w", err)
}
// TODO: also consider where the unsealed data sits
selector := newAllocSelector(m.index, storiface.FTCache|storiface.FTSealed, storiface.PathSealing)
err = m.sched.Schedule(ctx, sector, sealtasks.TTPreCommit1, selector, m.schedFetch(sector, storiface.FTUnsealed, storiface.PathSealing, storiface.AcquireMove), func(ctx context.Context, w Worker) error {
err := m.startWork(ctx, w, wk)(w.SealPreCommit1(ctx, sector, ticket, pieces))
if err != nil {
return err
}
waitRes()
return nil
})
if err != nil {
return nil, err
}
return out, waitErr
}
func (m *Manager) SealPreCommit2(ctx context.Context, sector storage.SectorRef, phase1Out storage.PreCommit1Out) (out storage.SectorCids, err error) {
ctx, cancel := context.WithCancel(ctx)
defer cancel()
wk, wait, cancel, err := m.getWork(ctx, sealtasks.TTPreCommit2, sector, phase1Out)
if err != nil {
return storage.SectorCids{}, xerrors.Errorf("getWork: %w", err)
}
defer cancel()
var waitErr error
waitRes := func() {
p, werr := m.waitWork(ctx, wk)
if werr != nil {
waitErr = werr
return
}
if p != nil {
out = p.(storage.SectorCids)
}
}
if wait { // already in progress
waitRes()
return out, waitErr
}
if err := m.index.StorageLock(ctx, sector.ID, storiface.FTSealed, storiface.FTCache); err != nil {
return storage.SectorCids{}, xerrors.Errorf("acquiring sector lock: %w", err)
}
selector := newExistingSelector(m.index, sector.ID, storiface.FTCache|storiface.FTSealed, true)
err = m.sched.Schedule(ctx, sector, sealtasks.TTPreCommit2, selector, m.schedFetch(sector, storiface.FTCache|storiface.FTSealed, storiface.PathSealing, storiface.AcquireMove), func(ctx context.Context, w Worker) error {
err := m.startWork(ctx, w, wk)(w.SealPreCommit2(ctx, sector, phase1Out))
if err != nil {
return err
}
waitRes()
return nil
})
if err != nil {
return storage.SectorCids{}, err
}
return out, waitErr
}
func (m *Manager) SealCommit1(ctx context.Context, sector storage.SectorRef, ticket abi.SealRandomness, seed abi.InteractiveSealRandomness, pieces []abi.PieceInfo, cids storage.SectorCids) (out storage.Commit1Out, err error) {
ctx, cancel := context.WithCancel(ctx)
defer cancel()
wk, wait, cancel, err := m.getWork(ctx, sealtasks.TTCommit1, sector, ticket, seed, pieces, cids)
if err != nil {
return storage.Commit1Out{}, xerrors.Errorf("getWork: %w", err)
}
defer cancel()
var waitErr error
waitRes := func() {
p, werr := m.waitWork(ctx, wk)
if werr != nil {
waitErr = werr
return
}
if p != nil {
out = p.(storage.Commit1Out)
}
}
if wait { // already in progress
waitRes()
return out, waitErr
}
if err := m.index.StorageLock(ctx, sector.ID, storiface.FTSealed, storiface.FTCache); err != nil {
return storage.Commit1Out{}, xerrors.Errorf("acquiring sector lock: %w", err)
}
// NOTE: We set allowFetch to false in so that we always execute on a worker
// with direct access to the data. We want to do that because this step is
// generally very cheap / fast, and transferring data is not worth the effort
selector := newExistingSelector(m.index, sector.ID, storiface.FTCache|storiface.FTSealed, false)
err = m.sched.Schedule(ctx, sector, sealtasks.TTCommit1, selector, m.schedFetch(sector, storiface.FTCache|storiface.FTSealed, storiface.PathSealing, storiface.AcquireMove), func(ctx context.Context, w Worker) error {
err := m.startWork(ctx, w, wk)(w.SealCommit1(ctx, sector, ticket, seed, pieces, cids))
if err != nil {
return err
}
waitRes()
return nil
})
if err != nil {
return nil, err
}
return out, waitErr
}
func (m *Manager) SealCommit2(ctx context.Context, sector storage.SectorRef, phase1Out storage.Commit1Out) (out storage.Proof, err error) {
wk, wait, cancel, err := m.getWork(ctx, sealtasks.TTCommit2, sector, phase1Out)
if err != nil {
return storage.Proof{}, xerrors.Errorf("getWork: %w", err)
}
defer cancel()
var waitErr error
waitRes := func() {
p, werr := m.waitWork(ctx, wk)
if werr != nil {
waitErr = werr
return
}
if p != nil {
out = p.(storage.Proof)
}
}
if wait { // already in progress
waitRes()
return out, waitErr
}
selector := newTaskSelector()
err = m.sched.Schedule(ctx, sector, sealtasks.TTCommit2, selector, schedNop, func(ctx context.Context, w Worker) error {
err := m.startWork(ctx, w, wk)(w.SealCommit2(ctx, sector, phase1Out))
if err != nil {
return err
}
waitRes()
return nil
})
if err != nil {
return nil, err
}
return out, waitErr
}
func (m *Manager) FinalizeSector(ctx context.Context, sector storage.SectorRef, keepUnsealed []storage.Range) error {
ctx, cancel := context.WithCancel(ctx)
defer cancel()
if err := m.index.StorageLock(ctx, sector.ID, storiface.FTNone, storiface.FTSealed|storiface.FTUnsealed|storiface.FTCache); err != nil {
return xerrors.Errorf("acquiring sector lock: %w", err)
}
unsealed := storiface.FTUnsealed
{
unsealedStores, err := m.index.StorageFindSector(ctx, sector.ID, storiface.FTUnsealed, 0, false)
if err != nil {
return xerrors.Errorf("finding unsealed sector: %w", err)
}
if len(unsealedStores) == 0 { // Is some edge-cases unsealed sector may not exist already, that's fine
unsealed = storiface.FTNone
}
}
pathType := storiface.PathStorage
{
sealedStores, err := m.index.StorageFindSector(ctx, sector.ID, storiface.FTSealed, 0, false)
if err != nil {
return xerrors.Errorf("finding sealed sector: %w", err)
}
for _, store := range sealedStores {
if store.CanSeal {
pathType = storiface.PathSealing
break
}
}
}
selector := newExistingSelector(m.index, sector.ID, storiface.FTCache|storiface.FTSealed, false)
err := m.sched.Schedule(ctx, sector, sealtasks.TTFinalize, selector,
m.schedFetch(sector, storiface.FTCache|storiface.FTSealed|unsealed, pathType, storiface.AcquireMove),
func(ctx context.Context, w Worker) error {
_, err := m.waitSimpleCall(ctx)(w.FinalizeSector(ctx, sector, keepUnsealed))
return err
})
if err != nil {
return err
}
fetchSel := newAllocSelector(m.index, storiface.FTCache|storiface.FTSealed, storiface.PathStorage)
moveUnsealed := unsealed
{
if len(keepUnsealed) == 0 {
moveUnsealed = storiface.FTNone
}
}
err = m.sched.Schedule(ctx, sector, sealtasks.TTFetch, fetchSel,
m.schedFetch(sector, storiface.FTCache|storiface.FTSealed|moveUnsealed, storiface.PathStorage, storiface.AcquireMove),
func(ctx context.Context, w Worker) error {
_, err := m.waitSimpleCall(ctx)(w.MoveStorage(ctx, sector, storiface.FTCache|storiface.FTSealed|moveUnsealed))
return err
})
if err != nil {
return xerrors.Errorf("moving sector to storage: %w", err)
}
return nil
}
func (m *Manager) ReleaseUnsealed(ctx context.Context, sector storage.SectorRef, safeToFree []storage.Range) error {
log.Warnw("ReleaseUnsealed todo")
return nil
}
func (m *Manager) Remove(ctx context.Context, sector storage.SectorRef) error {
ctx, cancel := context.WithCancel(ctx)
defer cancel()
if err := m.index.StorageLock(ctx, sector.ID, storiface.FTNone, storiface.FTSealed|storiface.FTUnsealed|storiface.FTCache); err != nil {
return xerrors.Errorf("acquiring sector lock: %w", err)
}
var err error
if rerr := m.storage.Remove(ctx, sector.ID, storiface.FTSealed, true); rerr != nil {
err = multierror.Append(err, xerrors.Errorf("removing sector (sealed): %w", rerr))
}
if rerr := m.storage.Remove(ctx, sector.ID, storiface.FTCache, true); rerr != nil {
err = multierror.Append(err, xerrors.Errorf("removing sector (cache): %w", rerr))
}
if rerr := m.storage.Remove(ctx, sector.ID, storiface.FTUnsealed, true); rerr != nil {
err = multierror.Append(err, xerrors.Errorf("removing sector (unsealed): %w", rerr))
}
return err
}
func (m *Manager) ReturnAddPiece(ctx context.Context, callID storiface.CallID, pi abi.PieceInfo, err *storiface.CallError) error {
return m.returnResult(ctx, callID, pi, err)
}
func (m *Manager) ReturnSealPreCommit1(ctx context.Context, callID storiface.CallID, p1o storage.PreCommit1Out, err *storiface.CallError) error {
return m.returnResult(ctx, callID, p1o, err)
}
func (m *Manager) ReturnSealPreCommit2(ctx context.Context, callID storiface.CallID, sealed storage.SectorCids, err *storiface.CallError) error {
return m.returnResult(ctx, callID, sealed, err)
}
func (m *Manager) ReturnSealCommit1(ctx context.Context, callID storiface.CallID, out storage.Commit1Out, err *storiface.CallError) error {
return m.returnResult(ctx, callID, out, err)
}
func (m *Manager) ReturnSealCommit2(ctx context.Context, callID storiface.CallID, proof storage.Proof, err *storiface.CallError) error {
return m.returnResult(ctx, callID, proof, err)
}
func (m *Manager) ReturnFinalizeSector(ctx context.Context, callID storiface.CallID, err *storiface.CallError) error {
return m.returnResult(ctx, callID, nil, err)
}
func (m *Manager) ReturnReleaseUnsealed(ctx context.Context, callID storiface.CallID, err *storiface.CallError) error {
return m.returnResult(ctx, callID, nil, err)
}
func (m *Manager) ReturnMoveStorage(ctx context.Context, callID storiface.CallID, err *storiface.CallError) error {
return m.returnResult(ctx, callID, nil, err)
}
func (m *Manager) ReturnUnsealPiece(ctx context.Context, callID storiface.CallID, err *storiface.CallError) error {
return m.returnResult(ctx, callID, nil, err)
}
func (m *Manager) ReturnReadPiece(ctx context.Context, callID storiface.CallID, ok bool, err *storiface.CallError) error {
return m.returnResult(ctx, callID, ok, err)
}
func (m *Manager) ReturnFetch(ctx context.Context, callID storiface.CallID, err *storiface.CallError) error {
return m.returnResult(ctx, callID, nil, err)
}
func (m *Manager) StorageLocal(ctx context.Context) (map[stores.ID]string, error) {
l, err := m.localStore.Local(ctx)
if err != nil {
return nil, err
}
out := map[stores.ID]string{}
for _, st := range l {
out[st.ID] = st.LocalPath
}
return out, nil
}
func (m *Manager) FsStat(ctx context.Context, id stores.ID) (fsutil.FsStat, error) {
return m.storage.FsStat(ctx, id)
}
func (m *Manager) SchedDiag(ctx context.Context, doSched bool) (interface{}, error) {
if doSched {
select {
case m.sched.workerChange <- struct{}{}:
case <-ctx.Done():
return nil, ctx.Err()
}
}
si, err := m.sched.Info(ctx)
if err != nil {
return nil, err
}
type SchedInfo interface{}
i := struct {
SchedInfo
ReturnedWork []string
Waiting []string
CallToWork map[string]string
EarlyRet []string
}{
SchedInfo: si,
CallToWork: map[string]string{},
}
m.workLk.Lock()
for w := range m.results {
i.ReturnedWork = append(i.ReturnedWork, w.String())
}
for id := range m.callRes {
i.EarlyRet = append(i.EarlyRet, id.String())
}
for w := range m.waitRes {
i.Waiting = append(i.Waiting, w.String())
}
for c, w := range m.callToWork {
i.CallToWork[c.String()] = w.String()
}
m.workLk.Unlock()
return i, nil
}
func (m *Manager) Close(ctx context.Context) error {
return m.sched.Close(ctx)
}
var _ Unsealer = &Manager{}
var _ SectorManager = &Manager{}