lotus/chain/stmgr/execute.go
Steven Allen efb9422c01
fix: eth: re-execute tipsets on missing events (#11588)
This will re-execute tipsets to forcibly re-compute and store events
when they're missing. This is effectively lazy backfilling of events.

NOTE: This _won't_ backfill the index itself, it'll just give us the
events.

fixes #11335
2024-01-30 10:29:54 -08:00

187 lines
5.8 KiB
Go

package stmgr
import (
"context"
"fmt"
"github.com/ipfs/go-cid"
"go.opencensus.io/trace"
"github.com/filecoin-project/lotus/api"
"github.com/filecoin-project/lotus/chain/store"
"github.com/filecoin-project/lotus/chain/types"
)
func (sm *StateManager) TipSetState(ctx context.Context, ts *types.TipSet) (st cid.Cid, rec cid.Cid, err error) {
return sm.tipSetState(ctx, ts, false)
}
// Recompute the tipset state without trying to lookup a pre-computed result in the chainstore.
// Useful if we know that our local chain-state isn't complete (e.g., we've discarded the events).
func (sm *StateManager) RecomputeTipSetState(ctx context.Context, ts *types.TipSet) (st cid.Cid, rec cid.Cid, err error) {
return sm.tipSetState(ctx, ts, true)
}
func (sm *StateManager) tipSetState(ctx context.Context, ts *types.TipSet, recompute bool) (st cid.Cid, rec cid.Cid, err error) {
ctx, span := trace.StartSpan(ctx, "tipSetState")
defer span.End()
if span.IsRecordingEvents() {
span.AddAttributes(trace.StringAttribute("tipset", fmt.Sprint(ts.Cids())))
}
ck := cidsToKey(ts.Cids())
sm.stlk.Lock()
cw, cwok := sm.compWait[ck]
if cwok {
sm.stlk.Unlock()
span.AddAttributes(trace.BoolAttribute("waited", true))
select {
case <-cw:
sm.stlk.Lock()
case <-ctx.Done():
return cid.Undef, cid.Undef, ctx.Err()
}
}
cached, ok := sm.stCache[ck]
if ok {
sm.stlk.Unlock()
span.AddAttributes(trace.BoolAttribute("cache", true))
return cached[0], cached[1], nil
}
ch := make(chan struct{})
sm.compWait[ck] = ch
defer func() {
sm.stlk.Lock()
delete(sm.compWait, ck)
if st != cid.Undef {
sm.stCache[ck] = []cid.Cid{st, rec}
}
sm.stlk.Unlock()
close(ch)
}()
sm.stlk.Unlock()
if ts.Height() == 0 {
// NB: This is here because the process that executes blocks requires that the
// block miner reference a valid miner in the state tree. Unless we create some
// magical genesis miner, this won't work properly, so we short circuit here
// This avoids the question of 'who gets paid the genesis block reward'.
// This also makes us not attempt to lookup the tipset state with
// tryLookupTipsetState, which would cause a very long, very slow walk.
return ts.Blocks()[0].ParentStateRoot, ts.Blocks()[0].ParentMessageReceipts, nil
}
// First, try to find the tipset in the current chain. If found, we can avoid re-executing
// it.
if !recompute {
if st, rec, found := tryLookupTipsetState(ctx, sm.cs, ts); found {
return st, rec, nil
}
}
st, rec, err = sm.tsExec.ExecuteTipSet(ctx, sm, ts, sm.tsExecMonitor, false)
if err != nil {
return cid.Undef, cid.Undef, err
}
return st, rec, nil
}
// Try to lookup a state & receipt CID for a given tipset by walking the chain instead of executing
// it. This will only successfully return the state/receipt CIDs if they're found in the state
// store.
//
// NOTE: This _won't_ recursively walk the receipt/state trees. It assumes that having the root
// implies having the rest of the tree. However, lotus generally makes that assumption anyways.
func tryLookupTipsetState(ctx context.Context, cs *store.ChainStore, ts *types.TipSet) (cid.Cid, cid.Cid, bool) {
nextTs, err := cs.GetTipsetByHeight(ctx, ts.Height()+1, nil, false)
if err != nil {
// Nothing to see here. The requested height may be beyond the current head.
return cid.Undef, cid.Undef, false
}
// Make sure we're on the correct fork.
if nextTs.Parents() != ts.Key() {
// Also nothing to see here. This just means that the requested tipset is on a
// different fork.
return cid.Undef, cid.Undef, false
}
stateCid := nextTs.ParentState()
receiptCid := nextTs.ParentMessageReceipts()
// Make sure we have the parent state.
if hasState, err := cs.StateBlockstore().Has(ctx, stateCid); err != nil {
log.Errorw("failed to lookup state-root in blockstore", "cid", stateCid, "error", err)
return cid.Undef, cid.Undef, false
} else if !hasState {
// We have the chain but don't have the state. It looks like we need to try
// executing?
return cid.Undef, cid.Undef, false
}
// Make sure we have the receipts.
if hasReceipts, err := cs.ChainBlockstore().Has(ctx, receiptCid); err != nil {
log.Errorw("failed to lookup receipts in blockstore", "cid", receiptCid, "error", err)
return cid.Undef, cid.Undef, false
} else if !hasReceipts {
// If we don't have the receipts, re-execute and try again.
return cid.Undef, cid.Undef, false
}
return stateCid, receiptCid, true
}
func (sm *StateManager) ExecutionTraceWithMonitor(ctx context.Context, ts *types.TipSet, em ExecMonitor) (cid.Cid, error) {
st, _, err := sm.tsExec.ExecuteTipSet(ctx, sm, ts, em, true)
return st, err
}
func (sm *StateManager) ExecutionTrace(ctx context.Context, ts *types.TipSet) (cid.Cid, []*api.InvocResult, error) {
tsKey := ts.Key()
// check if we have the trace for this tipset in the cache
if execTraceCacheSize > 0 {
sm.execTraceCacheLock.Lock()
if entry, ok := sm.execTraceCache.Get(tsKey); ok {
// we have to make a deep copy since caller can modify the invocTrace
// and we don't want that to change what we store in cache
invocTraceCopy := makeDeepCopy(entry.invocTrace)
sm.execTraceCacheLock.Unlock()
return entry.postStateRoot, invocTraceCopy, nil
}
sm.execTraceCacheLock.Unlock()
}
var invocTrace []*api.InvocResult
st, err := sm.ExecutionTraceWithMonitor(ctx, ts, &InvocationTracer{trace: &invocTrace})
if err != nil {
return cid.Undef, nil, err
}
if execTraceCacheSize > 0 {
invocTraceCopy := makeDeepCopy(invocTrace)
sm.execTraceCacheLock.Lock()
sm.execTraceCache.Add(tsKey, tipSetCacheEntry{st, invocTraceCopy})
sm.execTraceCacheLock.Unlock()
}
return st, invocTrace, nil
}
func makeDeepCopy(invocTrace []*api.InvocResult) []*api.InvocResult {
c := make([]*api.InvocResult, len(invocTrace))
for i, ir := range invocTrace {
if ir == nil {
continue
}
tmp := *ir
c[i] = &tmp
}
return c
}