lotus/cmd/lotus-chainwatch/sync.go

614 lines
15 KiB
Go
Raw Normal View History

2019-11-15 16:38:56 +00:00
package main
import (
"bytes"
2019-11-15 16:38:56 +00:00
"container/list"
"context"
2020-01-19 16:18:47 +00:00
"encoding/json"
"fmt"
2020-01-08 16:29:46 +00:00
"math"
"sync"
"time"
2020-02-26 22:54:34 +00:00
"github.com/filecoin-project/go-address"
"github.com/ipfs/go-cid"
cbg "github.com/whyrusleeping/cbor-gen"
"github.com/filecoin-project/specs-actors/actors/abi"
2020-02-26 22:54:34 +00:00
"github.com/filecoin-project/specs-actors/actors/abi/big"
"github.com/filecoin-project/specs-actors/actors/builtin"
"github.com/filecoin-project/specs-actors/actors/builtin/miner"
"github.com/filecoin-project/specs-actors/actors/builtin/power"
"github.com/filecoin-project/specs-actors/actors/util/adt"
2019-11-15 16:38:56 +00:00
"github.com/filecoin-project/lotus/api"
"github.com/filecoin-project/lotus/chain/store"
"github.com/filecoin-project/lotus/chain/types"
parmap "github.com/filecoin-project/lotus/lib/parmap"
2019-11-15 16:38:56 +00:00
)
func runSyncer(ctx context.Context, api api.FullNode, st *storage, maxBatch int) {
2019-11-15 16:38:56 +00:00
notifs, err := api.ChainNotify(ctx)
if err != nil {
panic(err)
}
go func() {
for notif := range notifs {
for _, change := range notif {
switch change.Type {
case store.HCCurrent:
fallthrough
case store.HCApply:
syncHead(ctx, api, st, change.Val, maxBatch)
2019-11-15 16:38:56 +00:00
case store.HCRevert:
log.Warnf("revert todo")
}
if change.Type == store.HCCurrent {
go subMpool(ctx, api, st)
go subBlocks(ctx, api, st)
}
2019-11-15 16:38:56 +00:00
}
}
}()
}
2020-06-30 20:22:58 +00:00
type minerStateInfo struct {
// common
addr address.Address
act types.Actor
stateroot cid.Cid
2020-07-02 04:08:45 +00:00
// calculating changes
tsKey types.TipSetKey
parentTsKey types.TipSetKey
// miner specific
2020-02-25 21:09:22 +00:00
state miner.State
info miner.MinerInfo
2019-12-11 23:31:59 +00:00
// tracked by power actor
rawPower big.Int
qalPower big.Int
ssize uint64
psize uint64
}
2020-01-19 16:18:47 +00:00
type actorInfo struct {
2020-07-02 04:08:45 +00:00
stateroot cid.Cid
tsKey types.TipSetKey
parentTsKey types.TipSetKey
state string
2020-01-19 16:18:47 +00:00
}
func syncHead(ctx context.Context, api api.FullNode, st *storage, headTs *types.TipSet, maxBatch int) {
2019-11-15 18:37:57 +00:00
var alk sync.Mutex
2019-12-11 22:17:44 +00:00
log.Infof("Getting synced block list")
hazlist := st.hasList()
2019-11-15 18:37:57 +00:00
log.Infof("Getting headers / actors")
// global list of all blocks that need to be synced
2020-01-08 16:29:46 +00:00
allToSync := map[cid.Cid]*types.BlockHeader{}
// a stack
2019-11-15 16:38:56 +00:00
toVisit := list.New()
for _, header := range headTs.Blocks() {
2019-11-15 16:38:56 +00:00
toVisit.PushBack(header)
}
// TODO consider making a db query to check where syncing left off at in the case of a restart and avoid reprocessing
// those entries, or write value to file on shutdown
// walk the entire chain starting from headTS
2019-11-15 16:38:56 +00:00
for toVisit.Len() > 0 {
bh := toVisit.Remove(toVisit.Back()).(*types.BlockHeader)
2019-12-11 22:17:44 +00:00
_, has := hazlist[bh.Cid()]
2020-01-08 16:29:46 +00:00
if _, seen := allToSync[bh.Cid()]; seen || has {
2019-11-15 18:37:57 +00:00
continue
2019-11-15 16:38:56 +00:00
}
2019-11-15 18:37:57 +00:00
2020-01-08 16:29:46 +00:00
allToSync[bh.Cid()] = bh
if len(allToSync)%500 == 10 {
log.Debugf("to visit: (%d) %s @%d", len(allToSync), bh.Cid(), bh.Height)
2019-11-15 16:38:56 +00:00
}
if len(bh.Parents) == 0 {
2019-11-15 18:37:57 +00:00
continue
2019-11-15 16:38:56 +00:00
}
pts, err := api.ChainGetTipSet(ctx, types.NewTipSetKey(bh.Parents...))
if err != nil {
log.Error(err)
continue
}
for _, header := range pts.Blocks() {
toVisit.PushBack(header)
}
}
// Main worker loop, this loop runs until all tipse from headTS to genesis have been processed.
2020-01-08 16:29:46 +00:00
for len(allToSync) > 0 {
// first map is addresses -> common actors states (head, code, balance, nonce)
// second map common actor states -> chain state (tipset, stateroot) & unique actor state (deserialization of their head CID) represented as json.
actors := map[address.Address]map[types.Actor]actorInfo{}
// map of actor public key address to ID address
addressToID := map[address.Address]address.Address{}
2020-02-26 09:05:22 +00:00
minH := abi.ChainEpoch(math.MaxInt64)
2019-11-15 16:38:56 +00:00
// find the blockheader with the lowest height
2020-01-08 16:29:46 +00:00
for _, header := range allToSync {
if header.Height < minH {
minH = header.Height
}
2019-11-15 18:37:57 +00:00
}
// toSync maps block cids to their headers and contains all block headers that will be synced in this batch
// `maxBatch` is a tunable parameter to control how many blocks we sync per iteration.
2020-01-08 16:29:46 +00:00
toSync := map[cid.Cid]*types.BlockHeader{}
for c, header := range allToSync {
2020-02-26 09:05:22 +00:00
if header.Height < minH+abi.ChainEpoch(maxBatch) {
2020-01-08 16:29:46 +00:00
toSync[c] = header
addressToID[header.Miner] = address.Undef
2020-01-08 16:29:46 +00:00
}
}
// remove everything we are syncing this round from the global list of blocks to sync
2020-01-08 16:29:46 +00:00
for c := range toSync {
delete(allToSync, c)
}
log.Infow("Starting Sync", "height", minH, "numBlocks", len(toSync), "maxBatch", maxBatch)
2020-01-08 16:29:46 +00:00
// map of addresses to changed actors
var changes map[string]types.Actor
// collect all actor state that has changes between block headers
2020-01-08 16:29:46 +00:00
paDone := 0
parmap.Par(50, parmap.MapArr(toSync), func(bh *types.BlockHeader) {
2020-01-08 16:29:46 +00:00
paDone++
if paDone%100 == 0 {
log.Infof("pa: %d %d%%", paDone, (paDone*100)/len(toSync))
}
if len(bh.Parents) == 0 { // genesis case
genesisTs, _ := types.NewTipSet([]*types.BlockHeader{bh})
2020-07-02 04:08:45 +00:00
st.genesisTs = genesisTs
aadrs, err := api.StateListActors(ctx, genesisTs.Key())
2020-01-08 16:29:46 +00:00
if err != nil {
log.Error(err)
return
}
// TODO suspicious there is not a lot to be gained by doing this in parallel since the genesis state
// is unlikely to contain a lot of actors, why not for loop here?
parmap.Par(50, aadrs, func(addr address.Address) {
act, err := api.StateGetActor(ctx, addr, genesisTs.Key())
2020-01-08 16:29:46 +00:00
if err != nil {
log.Error(err)
return
}
ast, err := api.StateReadState(ctx, addr, genesisTs.Key())
2020-01-19 16:18:47 +00:00
if err != nil {
log.Error(err)
return
}
state, err := json.Marshal(ast.State)
if err != nil {
log.Error(err)
return
}
2020-01-08 16:29:46 +00:00
alk.Lock()
_, ok := actors[addr]
if !ok {
2020-01-19 16:18:47 +00:00
actors[addr] = map[types.Actor]actorInfo{}
}
actors[addr][*act] = actorInfo{
2020-07-02 04:08:45 +00:00
stateroot: bh.ParentStateRoot,
tsKey: genesisTs.Key(),
parentTsKey: genesisTs.Key(),
state: string(state),
2020-01-08 16:29:46 +00:00
}
addressToID[addr] = address.Undef
2020-01-08 16:29:46 +00:00
alk.Unlock()
})
return
}
pts, err := api.ChainGetTipSet(ctx, types.NewTipSetKey(bh.Parents...))
if err != nil {
log.Error(err)
return
}
// TODO Does this return actors that have been deleted between states?
// collect all actors that had state changes between the blockheader parent-state and its grandparent-state.
changes, err = api.StateChangedActors(ctx, pts.ParentState(), bh.ParentStateRoot)
2020-01-08 16:29:46 +00:00
if err != nil {
log.Error(err)
return
}
// record the state of all actors that have changed
2020-01-08 16:29:46 +00:00
for a, act := range changes {
act := act
2020-01-08 16:29:46 +00:00
addr, err := address.NewFromString(a)
if err != nil {
log.Error(err)
return
}
ast, err := api.StateReadState(ctx, addr, pts.Key())
2020-01-19 16:18:47 +00:00
if err != nil {
log.Error(err)
return
}
2020-01-19 16:18:47 +00:00
state, err := json.Marshal(ast.State)
if err != nil {
log.Error(err)
return
}
2020-01-08 16:29:46 +00:00
alk.Lock()
_, ok := actors[addr]
if !ok {
2020-01-19 16:18:47 +00:00
actors[addr] = map[types.Actor]actorInfo{}
}
// a change occurred for the actor with address `addr` and state `act` at tipset `pts`.
2020-01-19 16:18:47 +00:00
actors[addr][act] = actorInfo{
2020-07-02 04:08:45 +00:00
stateroot: bh.ParentStateRoot,
state: string(state),
tsKey: pts.Key(),
parentTsKey: pts.Parents(),
}
addressToID[addr] = address.Undef
alk.Unlock()
2020-01-08 16:29:46 +00:00
}
})
// map of tipset to all miners that had a head-change at that tipset.
2020-06-30 20:22:58 +00:00
minerTips := make(map[types.TipSetKey][]*minerStateInfo, len(changes))
// heads we've seen, im being paranoid
headsSeen := make(map[cid.Cid]struct{}, len(actors))
2020-01-08 16:29:46 +00:00
log.Infof("Getting messages")
2020-01-08 16:29:46 +00:00
msgs, incls := fetchMessages(ctx, api, toSync)
2020-01-08 16:29:46 +00:00
log.Infof("Resolving addresses")
for _, message := range msgs {
addressToID[message.To] = address.Undef
addressToID[message.From] = address.Undef
2019-11-15 18:37:57 +00:00
}
parmap.Par(50, parmap.KMapArr(addressToID), func(addr address.Address) {
// FIXME: cannot use EmptyTSK here since actorID's can change during reorgs, need to use the corresponding tipset.
// TODO: figure out a way to get the corresponding tipset...
raddr, err := api.StateLookupID(ctx, addr, types.EmptyTSK)
2019-11-15 18:37:57 +00:00
if err != nil {
2020-01-08 16:29:46 +00:00
log.Warn(err)
2019-11-15 18:37:57 +00:00
return
}
alk.Lock()
addressToID[addr] = raddr
2019-11-15 18:37:57 +00:00
alk.Unlock()
2020-01-08 16:29:46 +00:00
})
2019-12-12 18:34:28 +00:00
2020-01-08 16:29:46 +00:00
log.Infof("Getting miner info")
2019-12-12 18:34:28 +00:00
minerChanges := 0
2020-01-08 16:29:46 +00:00
for addr, m := range actors {
for actor, c := range m {
2020-02-25 21:09:22 +00:00
if actor.Code != builtin.StorageMinerActorCodeID {
2020-01-08 16:29:46 +00:00
continue
}
2019-11-15 18:37:57 +00:00
// only want miner actors with head change events
if _, found := headsSeen[actor.Head]; found {
continue
}
minerChanges++
2020-06-30 20:22:58 +00:00
minerTips[c.tsKey] = append(minerTips[c.tsKey], &minerStateInfo{
2020-01-08 16:29:46 +00:00
addr: addr,
act: actor,
2020-01-19 16:18:47 +00:00
stateroot: c.stateroot,
2019-12-12 18:34:28 +00:00
2020-07-02 04:08:45 +00:00
tsKey: c.tsKey,
parentTsKey: c.parentTsKey,
state: miner.State{},
info: miner.MinerInfo{},
rawPower: big.Zero(),
qalPower: big.Zero(),
})
2020-02-26 22:54:34 +00:00
headsSeen[actor.Head] = struct{}{}
2020-01-08 16:29:46 +00:00
}
}
minerProcessingStartedAt := time.Now()
log.Infow("Processing miners", "numTips", len(minerTips), "numMinerChanges", minerChanges)
// extract the power actor state at each tipset, loop over all miners that changed at said tipset and extract their
// claims from the power actor state. This ensures we only fetch the power actors state once for each tipset.
2020-06-30 20:22:58 +00:00
parmap.Par(50, parmap.KVMapArr(minerTips), func(it func() (types.TipSetKey, []*minerStateInfo)) {
tsKey, minerInfo := it()
// get the power actors claims map
mp, err := getPowerActorClaimsMap(ctx, api, tsKey)
2020-01-08 16:29:46 +00:00
if err != nil {
log.Error(err)
return
}
// Get miner raw and quality power
for _, mi := range minerInfo {
var claim power.Claim
// get miner claim from power actors claim map and store if found, else the miner had no claim at
// this tipset
found, err := mp.Get(adt.AddrKey(mi.addr), &claim)
if err != nil {
log.Error(err)
}
if found {
mi.qalPower = claim.QualityAdjPower
mi.rawPower = claim.RawBytePower
}
// Get the miner state info
astb, err := api.ChainReadObj(ctx, mi.act.Head)
if err != nil {
log.Error(err)
return
}
if err := mi.state.UnmarshalCBOR(bytes.NewReader(astb)); err != nil {
log.Error(err)
return
}
mi.info = mi.state.Info
2020-01-08 16:29:46 +00:00
}
// TODO Get the Sector Count
// FIXME this is returning a lot of "address not found" errors, which is strange given that StateChangedActors
// retruns all actors that had a state change at tipset `k.tsKey`, maybe its returning deleted miners too??
/*
sszs, err := api.StateMinerSectorCount(ctx, k.addr, k.tsKey)
if err != nil {
info.psize = 0
info.ssize = 0
} else {
info.psize = sszs.Pset
info.ssize = sszs.Sset
}
*/
2020-01-08 16:29:46 +00:00
})
log.Infow("Completed Miner Processing", "duration", time.Since(minerProcessingStartedAt).String(), "processed", minerChanges)
2020-01-08 16:29:46 +00:00
log.Info("Getting receipts")
receipts := fetchParentReceipts(ctx, api, toSync)
log.Info("Storing headers")
if err := st.storeHeaders(toSync, true); err != nil {
log.Errorf("%+v", err)
2019-12-11 23:31:59 +00:00
return
}
2020-01-08 16:29:46 +00:00
log.Info("Storing address mapping")
if err := st.storeAddressMap(addressToID); err != nil {
log.Error(err)
return
}
2020-01-08 16:29:46 +00:00
log.Info("Storing actors")
if err := st.storeActors(actors); err != nil {
log.Error(err)
return
}
2020-01-08 16:29:46 +00:00
log.Info("Storing miners")
if err := st.storeMiners(minerTips); err != nil {
log.Error(err)
return
}
2020-01-08 16:29:46 +00:00
log.Info("Storing miner sectors")
sectorStart := time.Now()
if err := st.storeSectors(minerTips, api); err != nil {
log.Error(err)
return
}
log.Infow("Finished storing miner sectors", "duration", time.Since(sectorStart).String())
2020-06-30 20:22:58 +00:00
log.Info("Storing miner sectors heads")
if err := st.storeMinerSectorsHeads(minerTips, api); err != nil {
log.Error(err)
return
}
2020-07-02 04:08:45 +00:00
log.Info("updating miner sectors heads")
if err := st.updateMinerSectors(minerTips, api); err != nil {
log.Error(err)
return
}
2020-01-08 16:29:46 +00:00
log.Infof("Storing messages")
if err := st.storeMessages(msgs); err != nil {
log.Error(err)
return
}
2019-12-12 18:34:28 +00:00
2020-01-08 16:29:46 +00:00
log.Info("Storing message inclusions")
2019-12-12 18:34:28 +00:00
2020-01-08 16:29:46 +00:00
if err := st.storeMsgInclusions(incls); err != nil {
log.Error(err)
return
}
2019-12-10 23:42:36 +00:00
2020-01-08 16:29:46 +00:00
log.Infof("Storing parent receipts")
2019-12-10 23:42:36 +00:00
2020-01-08 16:29:46 +00:00
if err := st.storeReceipts(receipts); err != nil {
log.Error(err)
return
}
log.Infof("Sync stage done")
2019-12-10 23:42:36 +00:00
}
2020-01-20 00:49:52 +00:00
log.Infof("Get deals")
// TODO: incremental, gather expired
deals, err := api.StateMarketDeals(ctx, headTs.Key())
2020-01-20 00:49:52 +00:00
if err != nil {
log.Error(err)
return
}
log.Infof("Store deals")
if err := st.storeDeals(deals); err != nil {
log.Error(err)
return
}
2020-01-22 15:10:22 +00:00
log.Infof("Refresh views")
if err := st.refreshViews(); err != nil {
log.Error(err)
return
}
2019-11-15 16:38:56 +00:00
log.Infof("Sync done")
}
2019-11-15 18:37:57 +00:00
func fetchMessages(ctx context.Context, api api.FullNode, toSync map[cid.Cid]*types.BlockHeader) (map[cid.Cid]*types.Message, map[cid.Cid][]cid.Cid) {
2019-11-15 16:38:56 +00:00
var lk sync.Mutex
messages := map[cid.Cid]*types.Message{}
inclusions := map[cid.Cid][]cid.Cid{} // block -> msgs
parmap.Par(50, parmap.MapArr(toSync), func(header *types.BlockHeader) {
2019-11-15 18:37:57 +00:00
msgs, err := api.ChainGetBlockMessages(ctx, header.Cid())
if err != nil {
log.Error(err)
return
2019-11-15 16:38:56 +00:00
}
2019-11-15 18:37:57 +00:00
vmm := make([]*types.Message, 0, len(msgs.Cids))
for _, m := range msgs.BlsMessages {
vmm = append(vmm, m)
}
2019-11-15 16:38:56 +00:00
2019-11-15 18:37:57 +00:00
for _, m := range msgs.SecpkMessages {
vmm = append(vmm, &m.Message)
}
2019-11-15 16:38:56 +00:00
2019-11-15 18:37:57 +00:00
lk.Lock()
for _, message := range vmm {
messages[message.Cid()] = message
inclusions[header.Cid()] = append(inclusions[header.Cid()], message.Cid())
}
lk.Unlock()
})
2019-11-15 16:38:56 +00:00
return messages, inclusions
}
2019-12-03 11:05:12 +00:00
type mrec struct {
msg cid.Cid
state cid.Cid
idx int
}
func fetchParentReceipts(ctx context.Context, api api.FullNode, toSync map[cid.Cid]*types.BlockHeader) map[mrec]*types.MessageReceipt {
var lk sync.Mutex
out := map[mrec]*types.MessageReceipt{}
parmap.Par(50, parmap.MapArr(toSync), func(header *types.BlockHeader) {
2019-12-03 11:05:12 +00:00
recs, err := api.ChainGetParentReceipts(ctx, header.Cid())
if err != nil {
log.Error(err)
return
}
msgs, err := api.ChainGetParentMessages(ctx, header.Cid())
if err != nil {
log.Error(err)
return
}
lk.Lock()
for i, r := range recs {
out[mrec{
msg: msgs[i].Cid,
state: header.ParentStateRoot,
idx: i,
}] = r
}
lk.Unlock()
})
return out
}
// load the power actor state clam as an adt.Map at the tipset `ts`.
func getPowerActorClaimsMap(ctx context.Context, api api.FullNode, ts types.TipSetKey) (*adt.Map, error) {
powerActor, err := api.StateGetActor(ctx, builtin.StoragePowerActorAddr, ts)
if err != nil {
return nil, err
}
powerRaw, err := api.ChainReadObj(ctx, powerActor.Head)
if err != nil {
return nil, err
}
var powerActorState power.State
if err := powerActorState.UnmarshalCBOR(bytes.NewReader(powerRaw)); err != nil {
return nil, fmt.Errorf("failed to unmarshal power actor state: %w", err)
}
s := &apiIpldStore{ctx, api}
return adt.AsMap(s, powerActorState.Claims)
}
// require for AMT and HAMT access
// TODO extract this to a common location in lotus and reuse the code
type apiIpldStore struct {
ctx context.Context
api api.FullNode
}
func (ht *apiIpldStore) Context() context.Context {
return ht.ctx
}
func (ht *apiIpldStore) Get(ctx context.Context, c cid.Cid, out interface{}) error {
raw, err := ht.api.ChainReadObj(ctx, c)
if err != nil {
return err
}
cu, ok := out.(cbg.CBORUnmarshaler)
if ok {
if err := cu.UnmarshalCBOR(bytes.NewReader(raw)); err != nil {
return err
}
return nil
}
return fmt.Errorf("Object does not implement CBORUnmarshaler: %T", out)
}
func (ht *apiIpldStore) Put(ctx context.Context, v interface{}) (cid.Cid, error) {
return cid.Undef, fmt.Errorf("Put is not implemented on apiIpldStore")
}