// compact compacts etcd store and returns current rev.
// It will return the current compact time and global revision if no error occurred.
// Note that CAS fail will not incur any error.
func compact(ctx context.Context, client *clientv3.Client, t, rev int64) (int64, int64, error) {
	resp, err := client.KV.Txn(ctx).If(
		clientv3.Compare(clientv3.Version(compactRevKey), "=", t),
	).Then(
		clientv3.OpPut(compactRevKey, strconv.FormatInt(rev, 10)), // Expect side effect: increment Version
	).Else(
		clientv3.OpGet(compactRevKey),
	).Commit()
	if err != nil {
		return t, rev, err
	}

	curRev := resp.Header.Revision

	if !resp.Succeeded {
		curTime := resp.Responses[0].GetResponseRange().Kvs[0].Version
		return curTime, curRev, nil
	}
	curTime := t + 1

	if rev == 0 {
		// We don't compact on bootstrap.
		return curTime, curRev, nil
	}
	if _, err = client.Compact(ctx, rev); err != nil {
		return curTime, curRev, err
	}
	glog.Infof("etcd: compacted rev (%d), endpoints (%v)", rev, client.Endpoints())
	return curTime, curRev, nil
}

func (b *blockEtcd) CreateBlockVolume(volume *models.Volume) error {
	new, err := b.AtomicModifyKey([]byte(etcd.MkKey("meta", "volumeminter")), etcd.BytesAddOne)
	volume.Id = new.(uint64)
	if err != nil {
		return err
	}
	vbytes, err := volume.Marshal()
	if err != nil {
		return err
	}
	inodeBytes := torus.NewINodeRef(torus.VolumeID(volume.Id), 1).ToBytes()

	do := b.Etcd.Client.Txn(b.getContext()).If(
		etcdv3.Compare(etcdv3.Version(etcd.MkKey("volumes", volume.Name)), "=", 0),
	).Then(
		etcdv3.OpPut(etcd.MkKey("volumes", volume.Name), string(etcd.Uint64ToBytes(volume.Id))),
		etcdv3.OpPut(etcd.MkKey("volumeid", etcd.Uint64ToHex(volume.Id)), string(vbytes)),
		etcdv3.OpPut(etcd.MkKey("volumemeta", etcd.Uint64ToHex(volume.Id), "inode"), string(etcd.Uint64ToBytes(1))),
		etcdv3.OpPut(etcd.MkKey("volumemeta", etcd.Uint64ToHex(volume.Id), "blockinode"), string(inodeBytes)),
	)
	resp, err := do.Commit()
	if err != nil {
		return err
	}
	if !resp.Succeeded {
		return torus.ErrExists
	}
	return nil
}

// Lock locks the mutex with a cancellable context. If the context is cancelled
// while trying to acquire the lock, the mutex tries to clean its stale lock entry.
func (m *Mutex) Lock(ctx context.Context) error {
	s, serr := NewSession(m.client)
	if serr != nil {
		return serr
	}

	m.myKey = fmt.Sprintf("%s/%x", m.pfx, s.Lease())
	cmp := v3.Compare(v3.CreateRevision(m.myKey), "=", 0)
	// put self in lock waiters via myKey; oldest waiter holds lock
	put := v3.OpPut(m.myKey, "", v3.WithLease(s.Lease()))
	// reuse key in case this session already holds the lock
	get := v3.OpGet(m.myKey)
	resp, err := m.client.Txn(ctx).If(cmp).Then(put).Else(get).Commit()
	if err != nil {
		return err
	}
	m.myRev = resp.Header.Revision
	if !resp.Succeeded {
		m.myRev = resp.Responses[0].GetResponseRange().Kvs[0].CreateRevision
	}

	// wait for deletion revisions prior to myKey
	err = waitDeletes(ctx, m.client, m.pfx, v3.WithPrefix(), v3.WithRev(m.myRev-1))
	// release lock key if cancelled
	select {
	case <-ctx.Done():
		m.Unlock()
	default:
	}
	return err
}

func TestTxnWriteFail(t *testing.T) {
	defer testutil.AfterTest(t)

	clus := integration.NewClusterV3(t, &integration.ClusterConfig{Size: 3})
	defer clus.Terminate(t)

	kv := clientv3.NewKV(clus.Client(0))
	clus.Members[0].Stop(t)
	<-clus.Members[0].StopNotify()

	resp, err := kv.Txn().Then(clientv3.OpPut("foo", "bar", 0)).Commit()
	if err == nil {
		t.Fatalf("expected error, got response %v", resp)
	}

	// reconnect so cluster terminate doesn't complain about double-close
	clus.Members[0].Restart(t)

	// and ensure the put didn't take
	gresp, gerr := kv.Get("foo", 0)
	if gerr != nil {
		t.Fatal(gerr)
	}
	if len(gresp.Kvs) != 0 {
		t.Fatalf("expected no keys, got %v", gresp.Kvs)
	}
}

func parseRequestUnion(line string) (*clientv3.Op, error) {
	args := argify(line)
	if len(args) < 2 {
		return nil, fmt.Errorf("invalid txn compare request: %s", line)
	}

	opc := make(chan clientv3.Op, 1)

	put := NewPutCommand()
	put.Run = func(cmd *cobra.Command, args []string) {
		key, value, opts := getPutOp(cmd, args)
		opc <- clientv3.OpPut(key, value, opts...)
	}
	get := NewGetCommand()
	get.Run = func(cmd *cobra.Command, args []string) {
		key, opts := getGetOp(cmd, args)
		opc <- clientv3.OpGet(key, opts...)
	}
	del := NewDelCommand()
	del.Run = func(cmd *cobra.Command, args []string) {
		key, opts := getDelOp(cmd, args)
		opc <- clientv3.OpDelete(key, opts...)
	}
	cmds := &cobra.Command{SilenceErrors: true}
	cmds.AddCommand(put, get, del)

	cmds.SetArgs(args)
	if err := cmds.Execute(); err != nil {
		return nil, fmt.Errorf("invalid txn request: %s", line)
	}

	op := <-opc
	return &op, nil
}

func (c *etcdCtx) SetRing(ring torus.Ring) error {
	oldr, etcdver, err := c.getRing()
	if err != nil {
		return err
	}
	if oldr.Version() != ring.Version()-1 {
		return torus.ErrNonSequentialRing
	}
	b, err := ring.Marshal()
	if err != nil {
		return err
	}
	key := MkKey("meta", "the-one-ring")
	txn := c.etcd.Client.Txn(c.getContext()).If(
		etcdv3.Compare(etcdv3.Version(key), "=", etcdver),
	).Then(
		etcdv3.OpPut(key, string(b)),
	)
	resp, err := txn.Commit()
	if err != nil {
		return err
	}
	if resp.Succeeded {
		return nil
	}
	return torus.ErrAgain
}

// Create implements storage.Interface.Create.
func (s *store) Create(ctx context.Context, key string, obj, out runtime.Object, ttl uint64) error {
	if version, err := s.versioner.ObjectResourceVersion(obj); err == nil && version != 0 {
		return errors.New("resourceVersion should not be set on objects to be created")
	}
	data, err := runtime.Encode(s.codec, obj)
	if err != nil {
		return err
	}
	key = keyWithPrefix(s.pathPrefix, key)

	opts, err := s.ttlOpts(ctx, int64(ttl))
	if err != nil {
		return err
	}

	txnResp, err := s.client.KV.Txn(ctx).If(
		notFound(key),
	).Then(
		clientv3.OpPut(key, string(data), opts...),
	).Commit()
	if err != nil {
		return err
	}
	if !txnResp.Succeeded {
		return storage.NewKeyExistsError(key, 0)
	}

	if out != nil {
		putResp := txnResp.Responses[0].GetResponsePut()
		return decode(s.codec, s.versioner, data, out, putResp.Header.Revision)
	}
	return nil
}

func TestTxnWriteFail(t *testing.T) {
	defer testutil.AfterTest(t)

	clus := integration.NewClusterV3(t, &integration.ClusterConfig{Size: 3})
	defer clus.Terminate(t)

	kv := clientv3.NewKV(clus.Client(0))
	ctx := context.TODO()

	clus.Members[0].Stop(t)
	<-clus.Members[0].StopNotify()

	donec := make(chan struct{})
	go func() {
		resp, err := kv.Txn(ctx).Then(clientv3.OpPut("foo", "bar")).Commit()
		if err == nil {
			t.Fatalf("expected error, got response %v", resp)
		}
		donec <- struct{}{}
	}()

	dialTimeout := 5 * time.Second
	select {
	case <-time.After(2*dialTimeout + time.Second):
		t.Fatalf("timed out waiting for txn to fail")
	case <-donec:
		// don't restart cluster until txn errors out
	}

	go func() {
		// reconnect so terminate doesn't complain about double-close
		clus.Members[0].Restart(t)
		// wait for etcdserver to get established (CI races and get req times out)
		time.Sleep(2 * time.Second)
		donec <- struct{}{}

		// and ensure the put didn't take
		gresp, gerr := kv.Get(ctx, "foo")
		if gerr != nil {
			t.Fatal(gerr)
		}
		if len(gresp.Kvs) != 0 {
			t.Fatalf("expected no keys, got %v", gresp.Kvs)
		}
		donec <- struct{}{}
	}()

	select {
	case <-time.After(5 * time.Second):
		t.Fatalf("timed out waiting for restart")
	case <-donec:
	}

	select {
	case <-time.After(5 * time.Second):
		t.Fatalf("timed out waiting for get")
	case <-donec:
	}
}

// TestSTMSerialize tests that serialization is honored when serializable.
func TestSTMSerialize(t *testing.T) {
	clus := NewClusterV3(t, &ClusterConfig{Size: 3})
	defer clus.Terminate(t)

	etcdc := clus.RandClient()

	// set up initial keys
	keys := make([]string, 5)
	for i := 0; i < len(keys); i++ {
		keys[i] = fmt.Sprintf("foo-%d", i)
	}

	// update keys in full batches
	updatec := make(chan struct{})
	go func() {
		defer close(updatec)
		for i := 0; i < 5; i++ {
			s := fmt.Sprintf("%d", i)
			ops := []v3.Op{}
			for _, k := range keys {
				ops = append(ops, v3.OpPut(k, s))
			}
			if _, err := etcdc.Txn(context.TODO()).Then(ops...).Commit(); err != nil {
				t.Fatalf("couldn't put keys (%v)", err)
			}
			updatec <- struct{}{}
		}
	}()

	// read all keys in txn, make sure all values match
	errc := make(chan error)
	for range updatec {
		curEtcdc := clus.RandClient()
		applyf := func(stm concurrency.STM) error {
			vs := []string{}
			for i := range keys {
				vs = append(vs, stm.Get(keys[i]))
			}
			for i := range vs {
				if vs[0] != vs[i] {
					return fmt.Errorf("got vs[%d] = %v, want %v", i, vs[i], vs[0])
				}
			}
			return nil
		}
		go func() {
			_, err := concurrency.NewSTMSerializable(context.TODO(), curEtcdc, applyf)
			errc <- err
		}()
	}

	for i := 0; i < 5; i++ {
		if err := <-errc; err != nil {
			t.Error(err)
		}
	}
}

// GuaranteedUpdate implements storage.Interface.GuaranteedUpdate.
func (s *store) GuaranteedUpdate(ctx context.Context, key string, out runtime.Object, ignoreNotFound bool, precondtions *storage.Preconditions, tryUpdate storage.UpdateFunc) error {
	v, err := conversion.EnforcePtr(out)
	if err != nil {
		panic("unable to convert output object to pointer")
	}
	key = keyWithPrefix(s.pathPrefix, key)
	getResp, err := s.client.KV.Get(ctx, key)
	if err != nil {
		return err
	}
	for {
		origState, err := s.getState(getResp, key, v, ignoreNotFound)
		if err != nil {
			return err
		}

		if err := checkPreconditions(key, precondtions, origState.obj); err != nil {
			return err
		}

		ret, ttl, err := s.updateState(origState, tryUpdate)
		if err != nil {
			return err
		}

		data, err := runtime.Encode(s.codec, ret)
		if err != nil {
			return err
		}
		if bytes.Equal(data, origState.data) {
			return decode(s.codec, s.versioner, origState.data, out, origState.rev)
		}

		opts, err := s.ttlOpts(ctx, int64(ttl))
		if err != nil {
			return err
		}

		txnResp, err := s.client.KV.Txn(ctx).If(
			clientv3.Compare(clientv3.ModRevision(key), "=", origState.rev),
		).Then(
			clientv3.OpPut(key, string(data), opts...),
		).Else(
			clientv3.OpGet(key),
		).Commit()
		if err != nil {
			return err
		}
		if !txnResp.Succeeded {
			getResp = (*clientv3.GetResponse)(txnResp.Responses[0].GetResponseRange())
			glog.V(4).Infof("GuaranteedUpdate of %s failed because of a conflict, going to retry", key)
			continue
		}
		putResp := txnResp.Responses[0].GetResponsePut()
		return decode(s.codec, s.versioner, data, out, putResp.Header.Revision)
	}
}

func initEtcdMetadata(cfg torus.Config, gmd torus.GlobalMetadata, ringType torus.RingType) error {
	gmdbytes, err := json.Marshal(gmd)
	if err != nil {
		return err
	}
	emptyRing, err := ring.CreateRing(&models.Ring{
		Type:              uint32(ringType),
		Version:           1,
		ReplicationFactor: 2,
	})
	if err != nil {
		return err
	}
	ringb, err := emptyRing.Marshal()
	if err != nil {
		return err
	}

	client, err := etcdv3.New(etcdv3.Config{Endpoints: []string{cfg.MetadataAddress}, TLS: cfg.TLS})
	if err != nil {
		return err
	}
	defer client.Close()

	txn := client.Txn(context.Background())
	resp, err := txn.If(
		etcdv3.Compare(etcdv3.Version(MkKey("meta", "globalmetadata")), "=", 0),
	).Then(
		etcdv3.OpPut(MkKey("meta", "volumeminter"), string(Uint64ToBytes(1))),
		etcdv3.OpPut(MkKey("meta", "globalmetadata"), string(gmdbytes)),
	).Commit()
	if err != nil {
		return err
	}
	if !resp.Succeeded {
		return torus.ErrExists
	}
	_, err = client.Put(context.Background(), MkKey("meta", "the-one-ring"), string(ringb))
	if err != nil {
		return err
	}
	return nil
}

func putFunc(cmd *cobra.Command, args []string) {
	if keySpaceSize <= 0 {
		fmt.Fprintf(os.Stderr, "expected positive --key-space-size, got (%v)", keySpaceSize)
		os.Exit(1)
	}

	requests := make(chan v3.Op, totalClients)
	clients := mustCreateClients(totalClients, totalConns)
	k, v := make([]byte, keySize), string(mustRandBytes(valSize))

	bar = pb.New(putTotal)
	bar.Format("Bom !")
	bar.Start()

	r := newReport()
	for i := range clients {
		wg.Add(1)
		go func(c *v3.Client) {
			defer wg.Done()
			for op := range requests {
				st := time.Now()
				_, err := c.Do(context.Background(), op)
				r.Results() <- report.Result{Err: err, Start: st, End: time.Now()}
				bar.Increment()
			}
		}(clients[i])
	}

	go func() {
		for i := 0; i < putTotal; i++ {
			if seqKeys {
				binary.PutVarint(k, int64(i%keySpaceSize))
			} else {
				binary.PutVarint(k, int64(rand.Intn(keySpaceSize)))
			}
			requests <- v3.OpPut(string(k), v)
		}
		close(requests)
	}()

	if compactInterval > 0 {
		go func() {
			for {
				time.Sleep(compactInterval)
				compactKV(clients)
			}
		}()
	}

	rc := r.Run()
	wg.Wait()
	close(r.Results())
	bar.Finish()
	fmt.Println(<-rc)
}

// putNewKV attempts to create the given key, only succeeding if the key did
// not yet exist.
func putNewKV(kv v3.KV, key, val string, leaseID v3.LeaseID) (int64, error) {
	cmp := v3.Compare(v3.Version(key), "=", 0)
	req := v3.OpPut(key, val, v3.WithLease(leaseID))
	txnresp, err := kv.Txn(context.TODO()).If(cmp).Then(req).Commit()
	if err != nil {
		return 0, err
	}
	if !txnresp.Succeeded {
		return 0, ErrKeyExists
	}
	return txnresp.Header.Revision, nil
}

func ExampleKV_do() {
	cli, err := clientv3.New(clientv3.Config{
		Endpoints:   endpoints,
		DialTimeout: dialTimeout,
	})
	if err != nil {
		log.Fatal(err)
	}
	defer cli.Close()

	ops := []clientv3.Op{
		clientv3.OpPut("put-key", "123"),
		clientv3.OpGet("put-key"),
		clientv3.OpPut("put-key", "456")}

	for _, op := range ops {
		if _, err := cli.Do(context.TODO(), op); err != nil {
			log.Fatal(err)
		}
	}
}

// newSequentialKV allocates a new sequential key <prefix>/nnnnn with a given
// value and lease.  Note: a bookkeeping node __<prefix> is also allocated.
func newSequentialKV(kv v3.KV, prefix, val string, leaseID v3.LeaseID) (*RemoteKV, error) {
	resp, err := kv.Get(context.TODO(), prefix, v3.WithLastKey()...)
	if err != nil {
		return nil, err
	}

	// add 1 to last key, if any
	newSeqNum := 0
	if len(resp.Kvs) != 0 {
		fields := strings.Split(string(resp.Kvs[0].Key), "/")
		_, serr := fmt.Sscanf(fields[len(fields)-1], "%d", &newSeqNum)
		if serr != nil {
			return nil, serr
		}
		newSeqNum++
	}
	newKey := fmt.Sprintf("%s/%016d", prefix, newSeqNum)

	// base prefix key must be current (i.e., <=) with the server update;
	// the base key is important to avoid the following:
	// N1: LastKey() == 1, start txn.
	// N2: New Key 2, New Key 3, Delete Key 2
	// N1: txn succeeds allocating key 2 when it shouldn't
	baseKey := "__" + prefix

	// current revision might contain modification so +1
	cmp := v3.Compare(v3.ModRevision(baseKey), "<", resp.Header.Revision+1)
	reqPrefix := v3.OpPut(baseKey, "", v3.WithLease(leaseID))
	reqNewKey := v3.OpPut(newKey, val, v3.WithLease(leaseID))

	txn := kv.Txn(context.TODO())
	txnresp, err := txn.If(cmp).Then(reqPrefix, reqNewKey).Commit()
	if err != nil {
		return nil, err
	}
	if !txnresp.Succeeded {
		return newSequentialKV(kv, prefix, val, leaseID)
	}
	return &RemoteKV{kv, newKey, txnresp.Header.Revision, val}, nil
}

func putFunc(cmd *cobra.Command, args []string) {
	if keySpaceSize <= 0 {
		fmt.Fprintf(os.Stderr, "expected positive --key-space-size, got (%v)", keySpaceSize)
		os.Exit(1)
	}

	results = make(chan result)
	requests := make(chan v3.Op, totalClients)
	bar = pb.New(putTotal)

	k, v := make([]byte, keySize), string(mustRandBytes(valSize))

	clients := mustCreateClients(totalClients, totalConns)

	bar.Format("Bom !")
	bar.Start()

	for i := range clients {
		wg.Add(1)
		go doPut(context.Background(), clients[i], requests)
	}

	pdoneC := printReport(results)

	go func() {
		for i := 0; i < putTotal; i++ {
			if seqKeys {
				binary.PutVarint(k, int64(i%keySpaceSize))
			} else {
				binary.PutVarint(k, int64(rand.Intn(keySpaceSize)))
			}
			requests <- v3.OpPut(string(k), v)
		}
		close(requests)
	}()

	if compactInterval > 0 {
		go func() {
			for {
				time.Sleep(compactInterval)
				compactKV(clients)
			}
		}()
	}

	wg.Wait()

	bar.Finish()

	close(results)
	<-pdoneC
}

func ExampleKV_txn() {
	cli, err := clientv3.New(clientv3.Config{
		Endpoints:   endpoints,
		DialTimeout: dialTimeout,
	})
	if err != nil {
		log.Fatal(err)
	}
	defer cli.Close()

	kvc := clientv3.NewKV(cli)

	_, err = kvc.Put(context.TODO(), "key", "xyz")
	if err != nil {
		log.Fatal(err)
	}

	ctx, cancel := context.WithTimeout(context.Background(), requestTimeout)
	_, err = kvc.Txn(ctx).
		If(clientv3.Compare(clientv3.Value("key"), ">", "abc")). // txn value comparisons are lexical
		Then(clientv3.OpPut("key", "XYZ")).                      // this runs, since 'xyz' > 'abc'
		Else(clientv3.OpPut("key", "ABC")).
		Commit()
	cancel()
	if err != nil {
		log.Fatal(err)
	}

	gresp, err := kvc.Get(context.TODO(), "key")
	cancel()
	if err != nil {
		log.Fatal(err)
	}
	for _, ev := range gresp.Kvs {
		fmt.Printf("%s : %s\n", ev.Key, ev.Value)
	}
	// key : XYZ
}

func TestTxnError(t *testing.T) {
	defer testutil.AfterTest(t)

	clus := integration.NewClusterV3(t, &integration.ClusterConfig{Size: 1})
	defer clus.Terminate(t)

	kv := clientv3.NewKV(clus.RandClient())
	ctx := context.TODO()

	_, err := kv.Txn(ctx).Then(clientv3.OpPut("foo", "bar1"), clientv3.OpPut("foo", "bar2")).Commit()
	if err != rpctypes.ErrDuplicateKey {
		t.Fatalf("expected %v, got %v", rpctypes.ErrDuplicateKey, err)
	}

	ops := make([]clientv3.Op, v3rpc.MaxOpsPerTxn+10)
	for i := range ops {
		ops[i] = clientv3.OpPut(fmt.Sprintf("foo%d", i), "")
	}
	_, err = kv.Txn(ctx).Then(ops...).Commit()
	if err != rpctypes.ErrTooManyOps {
		t.Fatalf("expected %v, got %v", rpctypes.ErrTooManyOps, err)
	}
}

// Lock locks resources on sync
// This call blocks until you can get lock
func (s *Sync) Lock(path string, block bool) error {
	for {
		if s.HasLock(path) {
			return nil
		}
		var err error
		lease, err := s.etcdClient.Grant(s.withTimeout(), masterTTL)
		var resp *etcd.TxnResponse
		if err == nil {
			cmp := etcd.Compare(etcd.CreateRevision(path), "=", 0)
			put := etcd.OpPut(path, s.processID, etcd.WithLease(lease.ID))
			resp, err = s.etcdClient.Txn(s.withTimeout()).If(cmp).Then(put).Commit()
		}
		if err != nil || !resp.Succeeded {
			msg := fmt.Sprintf("failed to lock path %s", path)
			if err != nil {
				msg = fmt.Sprintf("failed to lock path %s: %s", path, err)
			}
			log.Notice(msg)

			s.locks.Remove(path)
			if !block {
				return errors.New(msg)
			}
			time.Sleep(masterTTL * time.Second)
			continue
		}
		log.Info("Locked %s", path)
		s.locks.Set(path, lease.ID)
		//Refresh master token
		go func() {
			defer func() {
				log.Notice("releasing keepalive lock for %s", path)
				s.locks.Remove(path)
			}()
			for s.HasLock(path) {
				ch, err := s.etcdClient.KeepAlive(s.withTimeout(), lease.ID)
				if err != nil {
					log.Notice("failed to keepalive lock for %s %s", path, err)
					return
				}
				for range ch {
				}
			}
		}()

		return nil
	}
}

func NewUniqueKV(ctx context.Context, kv v3.KV, pfx, val string, opts ...v3.OpOption) (string, int64, error) {
	for {
		newKey := fmt.Sprintf("%s/%v", pfx, time.Now().UnixNano())
		put := v3.OpPut(newKey, val, opts...)
		cmp := v3.Compare(v3.ModifiedRevision(newKey), "=", 0)
		resp, err := kv.Txn(ctx).If(cmp).Then(put).Commit()
		if err != nil {
			return "", 0, err
		}
		if !resp.Succeeded {
			continue
		}
		return newKey, resp.Header.Revision, nil
	}
}