Golang engine.MVCCGetProto函数代码示例

// TestRemoveRangeWithoutGC ensures that we do not panic when a
// replica has been removed but not yet GC'd (and therefore
// does not have an active raft group).
func TestRemoveRangeWithoutGC(t *testing.T) {
	defer leaktest.AfterTest(t)

	mtc := startMultiTestContext(t, 2)
	defer mtc.Stop()
	// Disable the GC queue and move the range from store 0 to 1.
	mtc.stores[0].DisableReplicaGCQueue(true)
	const rangeID roachpb.RangeID = 1
	mtc.replicateRange(rangeID, 1)
	mtc.unreplicateRange(rangeID, 0)

	// Wait for store 0 to process the removal.
	util.SucceedsWithin(t, time.Second, func() error {
		rep, err := mtc.stores[0].GetReplica(rangeID)
		if err != nil {
			return err
		}
		desc := rep.Desc()
		if len(desc.Replicas) != 1 {
			return util.Errorf("range has %d replicas", len(desc.Replicas))
		}
		return nil
	})

	// The replica's data is still on disk even though the Replica
	// object is removed.
	var desc roachpb.RangeDescriptor
	descKey := keys.RangeDescriptorKey(roachpb.RKeyMin)
	if ok, err := engine.MVCCGetProto(mtc.stores[0].Engine(), descKey,
		mtc.stores[0].Clock().Now(), true, nil, &desc); err != nil {
		t.Fatal(err)
	} else if !ok {
		t.Fatal("expected range descriptor to be present")
	}

	// Stop and restart the store to reset the replica's raftGroup
	// pointer to nil. As long as the store has not been restarted it
	// can continue to use its last known replica ID.
	mtc.stopStore(0)
	mtc.restartStore(0)
	// Turn off the GC queue to ensure that the replica is deleted at
	// startup instead of by the scanner. This is not 100% guaranteed
	// since the scanner could have already run at this point, but it
	// should be enough to prevent us from accidentally relying on the
	// scanner.
	mtc.stores[0].DisableReplicaGCQueue(true)

	// The Replica object is not recreated.
	if _, err := mtc.stores[0].GetReplica(rangeID); err == nil {
		t.Fatalf("expected replica to be missing")
	}

	// And the data is no longer on disk.
	if ok, err := engine.MVCCGetProto(mtc.stores[0].Engine(), descKey,
		mtc.stores[0].Clock().Now(), true, nil, &desc); err != nil {
		t.Fatal(err)
	} else if ok {
		t.Fatal("expected range descriptor to be absent")
	}
}

// TestReplicateRange verifies basic replication functionality by creating two stores
// and a range, replicating the range to the second store, and reading its data there.
func TestReplicateRange(t *testing.T) {
	defer leaktest.AfterTest(t)
	mtc := multiTestContext{}
	mtc.Start(t, 2)
	defer mtc.Stop()

	// Issue a command on the first node before replicating.
	incArgs, incResp := incrementArgs([]byte("a"), 5, 1, mtc.stores[0].StoreID())
	if err := mtc.stores[0].ExecuteCmd(context.Background(), proto.Call{Args: incArgs, Reply: incResp}); err != nil {
		t.Fatal(err)
	}

	rng, err := mtc.stores[0].GetRange(1)
	if err != nil {
		t.Fatal(err)
	}

	if err := rng.ChangeReplicas(proto.ADD_REPLICA,
		proto.Replica{
			NodeID:  mtc.stores[1].Ident.NodeID,
			StoreID: mtc.stores[1].Ident.StoreID,
		}); err != nil {
		t.Fatal(err)
	}
	// Verify no intent remains on range descriptor key.
	key := keys.RangeDescriptorKey(rng.Desc().StartKey)
	desc := proto.RangeDescriptor{}
	if ok, err := engine.MVCCGetProto(mtc.stores[0].Engine(), key, mtc.stores[0].Clock().Now(), true, nil, &desc); !ok || err != nil {
		t.Fatalf("fetching range descriptor yielded %t, %s", ok, err)
	}
	// Verify that in time, no intents remain on meta addressing
	// keys, and that range descriptor on the meta records is correct.
	util.SucceedsWithin(t, 1*time.Second, func() error {
		meta2 := keys.RangeMetaKey(proto.KeyMax)
		meta1 := keys.RangeMetaKey(meta2)
		for _, key := range []proto.Key{meta2, meta1} {
			metaDesc := proto.RangeDescriptor{}
			if ok, err := engine.MVCCGetProto(mtc.stores[0].Engine(), key, mtc.stores[0].Clock().Now(), true, nil, &metaDesc); !ok || err != nil {
				return util.Errorf("failed to resolve %s", key)
			}
			if !reflect.DeepEqual(metaDesc, desc) {
				return util.Errorf("descs not equal: %+v != %+v", metaDesc, desc)
			}
		}
		return nil
	})

	// Verify that the same data is available on the replica.
	util.SucceedsWithin(t, 1*time.Second, func() error {
		getArgs, getResp := getArgs([]byte("a"), 1, mtc.stores[1].StoreID())
		getArgs.ReadConsistency = proto.INCONSISTENT
		if err := mtc.stores[1].ExecuteCmd(context.Background(), proto.Call{Args: getArgs, Reply: getResp}); err != nil {
			return util.Errorf("failed to read data")
		}
		if v := mustGetInteger(getResp.Value); v != 5 {
			return util.Errorf("failed to read correct data: %d", v)
		}
		return nil
	})
}

// TestReplicateRange verifies basic replication functionality by creating two stores
// and a range, replicating the range to the second store, and reading its data there.
func TestReplicateRange(t *testing.T) {
	defer leaktest.AfterTest(t)
	mtc := startMultiTestContext(t, 2)
	defer mtc.Stop()

	// Issue a command on the first node before replicating.
	incArgs := incrementArgs([]byte("a"), 5)
	if _, err := client.SendWrapped(rg1(mtc.stores[0]), nil, &incArgs); err != nil {
		t.Fatal(err)
	}

	rng, err := mtc.stores[0].GetReplica(1)
	if err != nil {
		t.Fatal(err)
	}

	if err := rng.ChangeReplicas(roachpb.ADD_REPLICA,
		roachpb.ReplicaDescriptor{
			NodeID:  mtc.stores[1].Ident.NodeID,
			StoreID: mtc.stores[1].Ident.StoreID,
		}, rng.Desc()); err != nil {
		t.Fatal(err)
	}
	// Verify no intent remains on range descriptor key.
	key := keys.RangeDescriptorKey(rng.Desc().StartKey)
	desc := roachpb.RangeDescriptor{}
	if ok, err := engine.MVCCGetProto(mtc.stores[0].Engine(), key, mtc.stores[0].Clock().Now(), true, nil, &desc); !ok || err != nil {
		t.Fatalf("fetching range descriptor yielded %t, %s", ok, err)
	}
	// Verify that in time, no intents remain on meta addressing
	// keys, and that range descriptor on the meta records is correct.
	util.SucceedsWithin(t, 1*time.Second, func() error {
		meta2 := keys.Addr(keys.RangeMetaKey(roachpb.RKeyMax))
		meta1 := keys.Addr(keys.RangeMetaKey(meta2))
		for _, key := range []roachpb.RKey{meta2, meta1} {
			metaDesc := roachpb.RangeDescriptor{}
			if ok, err := engine.MVCCGetProto(mtc.stores[0].Engine(), key.AsRawKey(), mtc.stores[0].Clock().Now(), true, nil, &metaDesc); !ok || err != nil {
				return util.Errorf("failed to resolve %s", key.AsRawKey())
			}
			if !reflect.DeepEqual(metaDesc, desc) {
				return util.Errorf("descs not equal: %+v != %+v", metaDesc, desc)
			}
		}
		return nil
	})

	// Verify that the same data is available on the replica.
	util.SucceedsWithin(t, replicaReadTimeout, func() error {
		getArgs := getArgs([]byte("a"))
		if reply, err := client.SendWrappedWith(rg1(mtc.stores[1]), nil, roachpb.Header{
			ReadConsistency: roachpb.INCONSISTENT,
		}, &getArgs); err != nil {
			return util.Errorf("failed to read data: %s", err)
		} else if e, v := int64(5), mustGetInt(reply.(*roachpb.GetResponse).Value); v != e {
			return util.Errorf("failed to read correct data: expected %d, got %d", e, v)
		}
		return nil
	})
}

// InternalHeartbeatTxn updates the transaction status and heartbeat
// timestamp after receiving transaction heartbeat messages from
// coordinator. Returns the updated transaction.
func (r *Range) InternalHeartbeatTxn(batch engine.Engine, ms *engine.MVCCStats, args proto.InternalHeartbeatTxnRequest) (proto.InternalHeartbeatTxnResponse, error) {
	var reply proto.InternalHeartbeatTxnResponse

	key := keys.TransactionKey(args.Txn.Key, args.Txn.ID)

	var txn proto.Transaction
	if ok, err := engine.MVCCGetProto(batch, key, proto.ZeroTimestamp, true, nil, &txn); err != nil {
		return reply, err
	} else if !ok {
		// If no existing transaction record was found, initialize to a
		// shallow copy of the transaction in the request header. We copy
		// to avoid mutating the original below.
		txn = *args.Txn
	}

	if txn.Status == proto.PENDING {
		if txn.LastHeartbeat == nil {
			txn.LastHeartbeat = &proto.Timestamp{}
		}
		if txn.LastHeartbeat.Less(args.Header().Timestamp) {
			*txn.LastHeartbeat = args.Header().Timestamp
		}
		if err := engine.MVCCPutProto(batch, ms, key, proto.ZeroTimestamp, nil, &txn); err != nil {
			return reply, err
		}
	}

	reply.Txn = &txn
	return reply, nil
}

// InternalHeartbeatTxn updates the transaction status and heartbeat
// timestamp after receiving transaction heartbeat messages from
// coordinator. Returns the updated transaction.
func (r *Range) InternalHeartbeatTxn(batch engine.Engine, ms *engine.MVCCStats,
	args *proto.InternalHeartbeatTxnRequest, reply *proto.InternalHeartbeatTxnResponse) {
	key := keys.TransactionKey(args.Txn.Key, args.Txn.ID)

	var txn proto.Transaction
	ok, err := engine.MVCCGetProto(batch, key, proto.ZeroTimestamp, true, nil, &txn)
	if err != nil {
		reply.SetGoError(err)
		return
	}
	// If no existing transaction record was found, initialize
	// to the transaction in the request header.
	if !ok {
		gogoproto.Merge(&txn, args.Txn)
	}
	if txn.Status == proto.PENDING {
		if txn.LastHeartbeat == nil {
			txn.LastHeartbeat = &proto.Timestamp{}
		}
		if txn.LastHeartbeat.Less(args.Header().Timestamp) {
			*txn.LastHeartbeat = args.Header().Timestamp
		}
		if err := engine.MVCCPutProto(batch, ms, key, proto.ZeroTimestamp, nil, &txn); err != nil {
			reply.SetGoError(err)
			return
		}
	}
	reply.Txn = &txn
}

// InitialState implements the raft.Storage interface.
func (r *Range) InitialState() (raftpb.HardState, raftpb.ConfState, error) {
	var hs raftpb.HardState
	found, err := engine.MVCCGetProto(r.rm.Engine(), keys.RaftHardStateKey(r.Desc().RaftID),
		proto.ZeroTimestamp, true, nil, &hs)
	if err != nil {
		return raftpb.HardState{}, raftpb.ConfState{}, err
	}
	if !found {
		// We don't have a saved HardState, so set up the defaults.
		if r.isInitialized() {
			// Set the initial log term.
			hs.Term = raftInitialLogTerm
			hs.Commit = raftInitialLogIndex

			atomic.StoreUint64(&r.lastIndex, raftInitialLogIndex)
		} else {
			// This is a new range we are receiving from another node. Start
			// from zero so we will receive a snapshot.
			atomic.StoreUint64(&r.lastIndex, 0)
		}
	}

	var cs raftpb.ConfState
	// For uninitalized ranges, membership is unknown at this point.
	if found || r.isInitialized() {
		for _, rep := range r.Desc().Replicas {
			cs.Nodes = append(cs.Nodes, uint64(proto.MakeRaftNodeID(rep.NodeID, rep.StoreID)))
		}
	}

	return hs, cs, nil
}

func loadLeaderLease(eng engine.Engine, raftID proto.RaftID) (*proto.Lease, error) {
	lease := &proto.Lease{}
	if _, err := engine.MVCCGetProto(eng, keys.RaftLeaderLeaseKey(raftID), proto.ZeroTimestamp, true, nil, lease); err != nil {
		return nil, err
	}
	return lease, nil
}

func (ls *Stores) readBootstrapInfoLocked(bi *gossip.BootstrapInfo) error {
	latestTS := roachpb.ZeroTimestamp
	timestamps := map[roachpb.StoreID]roachpb.Timestamp{}

	// Find the most recent bootstrap info, collecting timestamps for
	// each store along the way.
	for id, s := range ls.storeMap {
		var storeBI gossip.BootstrapInfo
		ok, err := engine.MVCCGetProto(s.engine, keys.StoreGossipKey(), roachpb.ZeroTimestamp, true, nil, &storeBI)
		if err != nil {
			return err
		}
		timestamps[id] = storeBI.Timestamp
		if ok && latestTS.Less(storeBI.Timestamp) {
			latestTS = storeBI.Timestamp
			*bi = storeBI
		}
	}

	// Update all stores with an earlier timestamp.
	for id, s := range ls.storeMap {
		if timestamps[id].Less(latestTS) {
			if err := engine.MVCCPutProto(s.engine, nil, keys.StoreGossipKey(), roachpb.ZeroTimestamp, nil, bi); err != nil {
				return err
			}
			log.Infof("updated gossip bootstrap info to %s", s)
		}
	}

	ls.biLatestTS = latestTS
	return nil
}

// raftTruncatedStateLocked returns metadata about the log that preceded the
// first current entry. This includes both entries that have been compacted away
// and the dummy entries that make up the starting point of an empty log.
// raftTruncatedStateLocked requires that the replica lock be held.
func (r *Replica) raftTruncatedStateLocked() (roachpb.RaftTruncatedState, error) {
	if r.mu.truncatedState != nil {
		return *r.mu.truncatedState, nil
	}
	ts := roachpb.RaftTruncatedState{}
	ok, err := engine.MVCCGetProto(r.store.Engine(), keys.RaftTruncatedStateKey(r.RangeID),
		roachpb.ZeroTimestamp, true, nil, &ts)
	if err != nil {
		return ts, err
	}
	if !ok {
		if r.isInitializedLocked() {
			// If we created this range, set the initial log index/term.
			ts.Index = raftInitialLogIndex
			ts.Term = raftInitialLogTerm
		} else {
			// This is a new range we are receiving from another node. Start
			// from zero so we will receive a snapshot.
			ts.Index = 0
			ts.Term = 0
		}
	}

	if ts.Index != 0 {
		r.mu.truncatedState = &ts
	}
	return ts, nil
}

// raftTruncatedState returns metadata about the log that preceded the first
// current entry. This includes both entries that have been compacted away
// and the dummy entries that make up the starting point of an empty log.
func (r *Replica) raftTruncatedState() (proto.RaftTruncatedState, error) {
	if ts := r.getCachedTruncatedState(); ts != nil {
		return *ts, nil
	}
	ts := proto.RaftTruncatedState{}
	ok, err := engine.MVCCGetProto(r.rm.Engine(), keys.RaftTruncatedStateKey(r.Desc().RangeID),
		proto.ZeroTimestamp, true, nil, &ts)
	if err != nil {
		return ts, err
	}
	if !ok {
		if r.isInitialized() {
			// If we created this range, set the initial log index/term.
			ts.Index = raftInitialLogIndex
			ts.Term = raftInitialLogTerm
		} else {
			// This is a new range we are receiving from another node. Start
			// from zero so we will receive a snapshot.
			ts.Index = 0
			ts.Term = 0
		}
	}

	if ts.Index != 0 {
		r.setCachedTruncatedState(&ts)
	}
	return ts, nil
}

// Snapshot implements the raft.Storage interface.
func (r *Replica) Snapshot() (raftpb.Snapshot, error) {
	// Copy all the data from a consistent RocksDB snapshot into a RaftSnapshotData.
	snap := r.rm.NewSnapshot()
	defer snap.Close()
	var snapData proto.RaftSnapshotData

	// Read the range metadata from the snapshot instead of the members
	// of the Range struct because they might be changed concurrently.
	appliedIndex, err := r.loadAppliedIndex(snap)
	if err != nil {
		return raftpb.Snapshot{}, err
	}
	var desc proto.RangeDescriptor
	// We ignore intents on the range descriptor (consistent=false) because we
	// know they cannot be committed yet; operations that modify range
	// descriptors resolve their own intents when they commit.
	ok, err := engine.MVCCGetProto(snap, keys.RangeDescriptorKey(r.Desc().StartKey),
		r.rm.Clock().Now(), false /* !consistent */, nil, &desc)
	if err != nil {
		return raftpb.Snapshot{}, util.Errorf("failed to get desc: %s", err)
	}
	if !ok {
		return raftpb.Snapshot{}, util.Errorf("couldn't find range descriptor")
	}

	// Store RangeDescriptor as metadata, it will be retrieved by ApplySnapshot()
	snapData.RangeDescriptor = desc

	// Iterate over all the data in the range, including local-only data like
	// the response cache.
	for iter := newRangeDataIterator(r.Desc(), snap); iter.Valid(); iter.Next() {
		snapData.KV = append(snapData.KV,
			&proto.RaftSnapshotData_KeyValue{Key: iter.Key(), Value: iter.Value()})
	}

	data, err := gogoproto.Marshal(&snapData)
	if err != nil {
		return raftpb.Snapshot{}, err
	}

	// Synthesize our raftpb.ConfState from desc.
	var cs raftpb.ConfState
	for _, rep := range desc.Replicas {
		cs.Nodes = append(cs.Nodes, uint64(proto.MakeRaftNodeID(rep.NodeID, rep.StoreID)))
	}

	term, err := r.Term(appliedIndex)
	if err != nil {
		return raftpb.Snapshot{}, util.Errorf("failed to fetch term of %d: %s", appliedIndex, err)
	}

	return raftpb.Snapshot{
		Data: data,
		Metadata: raftpb.SnapshotMetadata{
			Index:     appliedIndex,
			Term:      term,
			ConfState: cs,
		},
	}, nil
}

func raftTruncatedState(
	eng engine.Reader, rangeID roachpb.RangeID,
) (roachpb.RaftTruncatedState, error) {
	ts := roachpb.RaftTruncatedState{}
	_, err := engine.MVCCGetProto(context.Background(), eng, keys.RaftTruncatedStateKey(rangeID),
		hlc.ZeroTimestamp, true, nil, &ts)
	return ts /* zero if not found */, err
}

func loadGCThreshold(
	ctx context.Context, reader engine.Reader, rangeID roachpb.RangeID,
) (hlc.Timestamp, error) {
	var t hlc.Timestamp
	_, err := engine.MVCCGetProto(ctx, reader, keys.RangeLastGCKey(rangeID),
		hlc.ZeroTimestamp, true, nil, &t)
	return t, err
}

// GetGCMetadata reads the latest GC metadata for this range.
func (r *Range) GetGCMetadata() (*proto.GCMetadata, error) {
	key := keys.RangeGCMetadataKey(r.Desc().RaftID)
	gcMeta := &proto.GCMetadata{}
	_, err := engine.MVCCGetProto(r.rm.Engine(), key, proto.ZeroTimestamp, true, nil, gcMeta)
	if err != nil {
		return nil, err
	}
	return gcMeta, nil
}

// GetLastVerificationTimestamp reads the timestamp at which the range's
// data was last verified.
func (r *Range) GetLastVerificationTimestamp() (proto.Timestamp, error) {
	key := keys.RangeLastVerificationTimestampKey(r.Desc().RaftID)
	timestamp := proto.Timestamp{}
	_, err := engine.MVCCGetProto(r.rm.Engine(), key, proto.ZeroTimestamp, true, nil, &timestamp)
	if err != nil {
		return proto.ZeroTimestamp, err
	}
	return timestamp, nil
}