ipfs-cluster/peer_manager_test.go

package ipfscluster

import (
	"context"
	"fmt"
	"sync"
	"testing"
	"time"

	"github.com/ipfs/ipfs-cluster/api"
	"github.com/ipfs/ipfs-cluster/config"
	"github.com/ipfs/ipfs-cluster/test"

	cid "github.com/ipfs/go-cid"
	host "github.com/libp2p/go-libp2p-host"
	peer "github.com/libp2p/go-libp2p-peer"
	peerstore "github.com/libp2p/go-libp2p-peerstore"
	ma "github.com/multiformats/go-multiaddr"
)

func peerManagerClusters(t *testing.T) ([]*Cluster, []*test.IpfsMock, host.Host) {
	cls := make([]*Cluster, nClusters, nClusters)
	mocks := make([]*test.IpfsMock, nClusters, nClusters)
	var wg sync.WaitGroup
	for i := 0; i < nClusters; i++ {
		wg.Add(1)
		go func(i int) {
			defer wg.Done()
			cl, m := createOnePeerCluster(t, i, testingClusterSecret)
			cls[i] = cl
			mocks[i] = m
		}(i)
	}
	wg.Wait()

	// Creat an identity
	ident, err := config.NewIdentity()
	if err != nil {
		t.Fatal(err)
	}
	// Create a config
	cfg := &Config{}
	cfg.Default()
	listen, _ := ma.NewMultiaddr("/ip4/127.0.0.1/tcp/0")
	cfg.ListenAddr = listen
	cfg.Secret = testingClusterSecret

	// Create a bootstrapping libp2p host
	h, _, dht, err := NewClusterHost(context.Background(), ident, cfg)
	if err != nil {
		t.Fatal(err)
	}

	// Connect all peers to that host. This will allow that they
	// can discover each others via DHT.
	for i := 0; i < nClusters; i++ {
		err := cls[i].host.Connect(
			context.Background(),
			peerstore.PeerInfo{
				ID:    h.ID(),
				Addrs: h.Addrs(),
			},
		)
		if err != nil {
			t.Fatal(err)
		}
	}
	dht.Bootstrap(context.Background())

	return cls, mocks, h
}

func clusterAddr(c *Cluster) ma.Multiaddr {
	for _, a := range c.host.Addrs() {
		if _, err := a.ValueForProtocol(ma.P_IP4); err == nil {
			p := peer.IDB58Encode(c.id)
			cAddr, _ := ma.NewMultiaddr(fmt.Sprintf("%s/ipfs/%s", a, p))
			return cAddr
		}
	}
	return nil
}

func TestClustersPeerAdd(t *testing.T) {
	ctx := context.Background()
	clusters, mocks, boot := peerManagerClusters(t)
	defer shutdownClusters(t, clusters, mocks)
	defer boot.Close()

	if len(clusters) < 2 {
		t.Skip("need at least 2 nodes for this test")
	}

	for i := 1; i < len(clusters); i++ {
		id, err := clusters[0].PeerAdd(ctx, clusters[i].id)
		if err != nil {
			t.Fatal(err)
		}
		if !containsPeer(id.ClusterPeers, clusters[0].id) {
			// ClusterPeers is originally empty and contains nodes as we add them
			t.Log(i, id.ClusterPeers)
			t.Fatal("cluster peers should be up to date with the cluster")
		}

		for j := 0; j < i; j++ {
			if err := clusters[j].consensus.Trust(ctx, clusters[i].id); err != nil {
				t.Fatal(err)
			}
			if err := clusters[i].consensus.Trust(ctx, clusters[j].id); err != nil {
				t.Fatal(err)
			}
		}
	}

	h := test.Cid1
	err := clusters[1].Pin(ctx, api.PinCid(h))
	if err != nil {
		t.Fatal(err)
	}
	pinDelay()

	f := func(t *testing.T, c *Cluster) {
		ids := c.Peers(ctx)

		// check they are tracked by the peer manager
		if len(ids) != nClusters {
			//t.Log(ids)
			t.Error("added clusters are not part of clusters")
		}

		// Check that they are part of the consensus
		pins, err := c.Pins(ctx)
		if err != nil {
			t.Fatal(err)
		}
		if len(pins) != 1 {
			t.Log(pins)
			t.Error("expected 1 pin everywhere")
		}

		if len(c.ID(ctx).ClusterPeers) != nClusters {
			t.Log(c.ID(ctx).ClusterPeers)
			t.Error("By now cluster peers should reflect all peers")
		}
	}
	runF(t, clusters, f)

	for _, c := range clusters {
		c.Shutdown(ctx)
	}

	f2 := func(t *testing.T, c *Cluster) {
		// check that all peers are part of the peerstore
		// (except ourselves)
		addrs := c.peerManager.LoadPeerstore()
		peerMap := make(map[peer.ID]struct{})
		for _, a := range addrs {
			pid, _, err := api.Libp2pMultiaddrSplit(a)
			if err != nil {
				t.Fatal(err)
			}
			peerMap[pid] = struct{}{}
		}

		if len(peerMap) == 0 {
			t.Errorf("%s: peerstore to store at least 1 peer", c.id)
		}

	}
	runF(t, clusters, f2)
}

func TestClustersJoinBadPeer(t *testing.T) {
	ctx := context.Background()
	clusters, mocks, boot := peerManagerClusters(t)
	defer shutdownClusters(t, clusters, mocks)
	defer boot.Close()

	if len(clusters) < 2 {
		t.Skip("need at least 2 nodes for this test")
	}

	addr := clusterAddr(clusters[1])

	// We add a cluster that has been shutdown
	// (closed transports)
	clusters[1].Shutdown(ctx)

	// Let the OS actually close the ports.
	// Sometimes we hang otherwise.
	delay()

	err := clusters[0].Join(ctx, addr)
	if err == nil {
		t.Error("expected an error")
	}
	ids := clusters[0].Peers(ctx)
	if len(ids) != 1 {
		t.Error("cluster should have only one member")
	}
}

func TestClustersPeerAddInUnhealthyCluster(t *testing.T) {
	ctx := context.Background()
	clusters, mocks, boot := peerManagerClusters(t)
	defer shutdownClusters(t, clusters, mocks)
	defer boot.Close()

	if len(clusters) < 3 {
		t.Skip("need at least 3 nodes for this test")
	}

	_, err := clusters[0].PeerAdd(ctx, clusters[1].id)
	ttlDelay()
	ids := clusters[1].Peers(ctx)
	if len(ids) != 2 {
		t.Error("expected 2 peers")
	}

	// Now we shutdown the one member of the running cluster
	// and try to add someone else.
	err = clusters[1].Shutdown(ctx)
	if err != nil {
		t.Error("Shutdown should be clean: ", err)
	}
	switch consensus {
	case "raft":
		delay() // This makes sure the leader realizes that it's not
		// leader anymore. Otherwise it commits fine.

		_, err = clusters[0].PeerAdd(ctx, clusters[2].id)

		if err == nil {
			t.Error("expected an error")
		}

		ids = clusters[0].Peers(ctx)
		if len(ids) != 2 {
			t.Error("cluster should still have 2 peers")
		}
	case "crdt":
		// crdt does not really care whether we add or remove

		delay() // let metrics expire
		_, err = clusters[0].PeerAdd(ctx, clusters[2].id)

		if err != nil {
			t.Error(err)
		}

		ttlDelay()
		ids = clusters[0].Peers(ctx)
		if len(ids) != 2 {
			t.Error("cluster should have 2 peers after removing and adding 1")
		}
	default:
		t.Fatal("bad consensus")
	}
}

func TestClustersPeerRemove(t *testing.T) {
	ctx := context.Background()
	clusters, mocks := createClusters(t)
	defer shutdownClusters(t, clusters, mocks)

	if len(clusters) < 2 {
		t.Skip("test needs at least 2 clusters")
	}

	switch consensus {
	case "crdt":
		// Peer Rm is a no op.
		return
	case "raft":
		p := clusters[1].ID(ctx).ID
		err := clusters[0].PeerRemove(ctx, p)
		if err != nil {
			t.Error(err)
		}

		delay()

		f := func(t *testing.T, c *Cluster) {
			if c.ID(ctx).ID == p { //This is the removed cluster
				_, ok := <-c.Done()
				if ok {
					t.Error("removed peer should have exited")
				}
			} else {
				ids := c.Peers(ctx)
				if len(ids) != nClusters-1 {
					t.Error("should have removed 1 peer")
				}
			}
		}

		runF(t, clusters, f)
	default:
		t.Fatal("bad consensus")
	}
}

func TestClustersPeerRemoveSelf(t *testing.T) {
	ctx := context.Background()
	// this test hangs sometimes if there are problems
	clusters, mocks := createClusters(t)
	defer shutdownClusters(t, clusters, mocks)

	switch consensus {
	case "crdt":
		// remove is a no op in CRDTs
		return

	case "raft":
		for i := 0; i < len(clusters); i++ {
			waitForLeaderAndMetrics(t, clusters)
			peers := clusters[i].Peers(ctx)
			t.Logf("Current cluster size: %d", len(peers))
			if len(peers) != (len(clusters) - i) {
				t.Fatal("Previous peers not removed correctly")
			}
			err := clusters[i].PeerRemove(ctx, clusters[i].ID(ctx).ID)
			// Last peer member won't be able to remove itself
			// In this case, we shut it down.
			if err != nil {
				if i != len(clusters)-1 { //not last
					t.Error(err)
				} else {
					err := clusters[i].Shutdown(ctx)
					if err != nil {
						t.Fatal(err)
					}
				}
			}
			_, more := <-clusters[i].Done()
			if more {
				t.Error("should be done")
			}
		}
	default:
		t.Fatal("bad consensus")
	}
}

func TestClustersPeerRemoveLeader(t *testing.T) {
	ctx := context.Background()
	// this test is like the one above, except it always
	// removes the current leader.
	// this test hangs sometimes if there are problems
	clusters, mocks := createClusters(t)
	defer shutdownClusters(t, clusters, mocks)

	switch consensus {
	case "crdt":
		return
	case "raft":

		findLeader := func() *Cluster {
			var l peer.ID
			for _, c := range clusters {
				if !c.shutdownB {
					waitForLeaderAndMetrics(t, clusters)
					l, _ = c.consensus.Leader(ctx)
				}
			}
			for _, c := range clusters {
				if c.id == l {
					return c
				}
			}
			return nil
		}

		for i := 0; i < len(clusters); i++ {
			leader := findLeader()
			peers := leader.Peers(ctx)
			t.Logf("Current cluster size: %d", len(peers))
			if len(peers) != (len(clusters) - i) {
				t.Fatal("Previous peers not removed correctly")
			}
			err := leader.PeerRemove(ctx, leader.id)
			// Last peer member won't be able to remove itself
			// In this case, we shut it down.
			if err != nil {
				if i != len(clusters)-1 { //not last
					t.Error(err)
				} else {
					err := leader.Shutdown(ctx)
					if err != nil {
						t.Fatal(err)
					}
				}
			}
			_, more := <-leader.Done()
			if more {
				t.Error("should be done")
			}
			time.Sleep(time.Second / 2)
		}
	default:
		t.Fatal("bad consensus")
	}
}

func TestClustersPeerRemoveReallocsPins(t *testing.T) {
	ctx := context.Background()
	clusters, mocks := createClusters(t)
	defer shutdownClusters(t, clusters, mocks)

	if consensus == "crdt" {
		t.Log("FIXME when re-alloc changes come through")
		return
	}

	if len(clusters) < 3 {
		t.Skip("test needs at least 3 clusters")
	}

	// Adjust the replication factor for re-allocation
	for _, c := range clusters {
		c.config.ReplicationFactorMin = nClusters - 1
		c.config.ReplicationFactorMax = nClusters - 1
	}

	// We choose to remove the leader, to make things even more interesting
	leaderID, err := clusters[0].consensus.Leader(ctx)
	if err != nil {
		t.Fatal(err)
	}

	var leader *Cluster
	var leaderi int
	for i, cl := range clusters {
		if id := cl.ID(ctx).ID; id == leaderID {
			leader = cl
			leaderi = i
			break
		}
	}
	if leader == nil {
		t.Fatal("did not find a leader?")
	}

	leaderMock := mocks[leaderi]

	// Remove leader from set
	clusters = append(clusters[:leaderi], clusters[leaderi+1:]...)
	mocks = append(mocks[:leaderi], mocks[leaderi+1:]...)
	defer leader.Shutdown(ctx)
	defer leaderMock.Close()

	prefix := test.Cid1.Prefix()

	// Pin nCluster random pins. This ensures each peer will
	// pin the same number of Cids.
	for i := 0; i < nClusters; i++ {
		h, err := prefix.Sum(randomBytes())
		checkErr(t, err)
		err = leader.Pin(ctx, api.PinCid(h))
		checkErr(t, err)
		ttlDelay()
	}

	pinDelay()

	// At this point, all peers must have nClusters -1  pins
	// associated to them.
	// Find out which pins are associated to the leader.
	interestingCids := []cid.Cid{}

	pins, err := leader.Pins(ctx)
	if err != nil {
		t.Fatal(err)
	}
	if len(pins) != nClusters {
		t.Fatal("expected number of tracked pins to be nClusters")
	}
	for _, p := range pins {
		if containsPeer(p.Allocations, leaderID) {
			//t.Logf("%s pins %s", leaderID, p.Cid)
			interestingCids = append(interestingCids, p.Cid)
		}
	}

	if len(interestingCids) != nClusters-1 {
		//t.Fatal("The number of allocated Cids is not expected")
		t.Fatalf("Expected %d allocated CIDs but got %d", nClusters-1,
			len(interestingCids))
	}

	// Now the leader removes itself
	err = leader.PeerRemove(ctx, leaderID)
	if err != nil {
		t.Fatal("error removing peer:", err)
	}

	delay()
	waitForLeaderAndMetrics(t, clusters)
	delay() // this seems to fail when not waiting enough...

	for _, icid := range interestingCids {
		// Now check that the allocations are new.
		newPin, err := clusters[1].PinGet(ctx, icid)
		if err != nil {
			t.Fatal("error getting the new allocations for", icid)
		}
		if containsPeer(newPin.Allocations, leaderID) {
			t.Fatal("pin should not be allocated to the removed peer")
		}
	}
}

func TestClustersPeerJoin(t *testing.T) {
	ctx := context.Background()
	clusters, mocks, boot := peerManagerClusters(t)
	defer shutdownClusters(t, clusters, mocks)
	defer boot.Close()

	if len(clusters) < 3 {
		t.Skip("test needs at least 3 clusters")
	}

	for i := 1; i < len(clusters); i++ {
		for j := 0; j < i; j++ {
			if err := clusters[j].consensus.Trust(ctx, clusters[i].id); err != nil {
				t.Fatal(err)
			}
			if err := clusters[i].consensus.Trust(ctx, clusters[j].id); err != nil {
				t.Fatal(err)
			}
		}

		err := clusters[i].Join(ctx, clusterAddr(clusters[0]))
		if err != nil {
			t.Fatal(err)
		}
	}

	hash := test.Cid1
	clusters[0].Pin(ctx, api.PinCid(hash))
	pinDelay()

	for _, p := range clusters {
		t.Log(p.id.String())
	}

	f := func(t *testing.T, c *Cluster) {
		peers := c.Peers(ctx)
		str := c.id.String() + "\n"
		for _, p := range peers {
			str += "  - " + p.ID.String() + "\n"
		}
		t.Log(str)
		if len(peers) != nClusters {
			t.Error("all peers should be connected")
		}
		pins, err := c.Pins(ctx)
		if err != nil {
			t.Fatal(err)
		}
		if len(pins) != 1 || !pins[0].Cid.Equals(hash) {
			t.Error("all peers should have pinned the cid")
		}
	}
	runF(t, clusters, f)
}

func TestClustersPeerJoinAllAtOnce(t *testing.T) {
	ctx := context.Background()
	clusters, mocks, boot := peerManagerClusters(t)
	defer shutdownClusters(t, clusters, mocks)
	defer boot.Close()

	if len(clusters) < 2 {
		t.Skip("test needs at least 2 clusters")
	}

	f := func(t *testing.T, c *Cluster) {
		if err := c.consensus.Trust(ctx, clusters[0].id); err != nil {
			t.Fatal(err)
		}

		err := c.Join(ctx, clusterAddr(clusters[0]))
		if err != nil {
			t.Fatal(err)
		}
	}
	runF(t, clusters[1:], f)

	hash := test.Cid1
	clusters[0].Pin(ctx, api.PinCid(hash))
	pinDelay()

	f2 := func(t *testing.T, c *Cluster) {
		peers := c.Peers(ctx)
		if len(peers) != nClusters {
			t.Error("all peers should be connected")
		}
		pins, err := c.Pins(ctx)
		if err != nil {
			t.Fatal(err)
		}
		if len(pins) != 1 || !pins[0].Cid.Equals(hash) {
			t.Error("all peers should have pinned the cid")
		}
	}
	runF(t, clusters, f2)
}

// This test fails a lot when re-use port is not available (MacOS, Windows)
// func TestClustersPeerJoinAllAtOnceWithRandomBootstrap(t *testing.T) {
// 	clusters, mocks,boot := peerManagerClusters(t)
// 	defer shutdownClusters(t, clusters, mocks)
//      defer boot.Close()
// 	if len(clusters) < 3 {
// 		t.Skip("test needs at least 3 clusters")
// 	}

// 	delay()

// 	// We have a 2 node cluster and the rest of nodes join
// 	// one of the two seeds randomly

// 	err := clusters[1].Join(clusterAddr(clusters[0]))
// 	if err != nil {
// 		t.Fatal(err)
// 	}

// 	f := func(t *testing.T, c *Cluster) {
// 		j := rand.Intn(2)
// 		err := c.Join(clusterAddr(clusters[j]))
// 		if err != nil {
// 			t.Fatal(err)
// 		}
// 	}
// 	runF(t, clusters[2:], f)

// 	hash := test.Cid1
// 	clusters[0].Pin(api.PinCid(hash))
// 	delay()

// 	f2 := func(t *testing.T, c *Cluster) {
// 		peers := c.Peers()
// 		if len(peers) != nClusters {
// 			peersIds := []peer.ID{}
// 			for _, p := range peers {
// 				peersIds = append(peersIds, p.ID)
// 			}
// 			t.Errorf("%s sees %d peers: %s", c.id, len(peers), peersIds)
// 		}
// 		pins := c.Pins()
// 		if len(pins) != 1 || !pins[0].Cid.Equals(hash) {
// 			t.Error("all peers should have pinned the cid")
// 		}
// 	}
// 	runF(t, clusters, f2)
// }

// Tests that a peer catches up on the state correctly after rejoining
func TestClustersPeerRejoin(t *testing.T) {
	ctx := context.Background()
	clusters, mocks, boot := peerManagerClusters(t)
	defer shutdownClusters(t, clusters, mocks)
	defer boot.Close()

	// pin something in c0
	pin1 := test.Cid1
	err := clusters[0].Pin(ctx, api.PinCid(pin1))
	if err != nil {
		t.Fatal(err)
	}

	// add all clusters
	for i := 1; i < len(clusters); i++ {
		for j := 0; j < i; j++ {
			if err := clusters[j].consensus.Trust(ctx, clusters[i].id); err != nil {
				t.Fatal(err)
			}
			if err := clusters[i].consensus.Trust(ctx, clusters[j].id); err != nil {
				t.Fatal(err)
			}
		}

		err := clusters[i].Join(ctx, clusterAddr(clusters[0]))
		if err != nil {
			t.Fatal(err)
		}
	}

	delay()

	// all added peers should have the content
	for i := 1; i < len(clusters); i++ {
		pinfo := clusters[i].tracker.Status(ctx, pin1)
		if pinfo.Status != api.TrackerStatusPinned {
			t.Error("Added peers should pin the content")
		}
	}

	clusters[0].config.LeaveOnShutdown = true
	err = clusters[0].Shutdown(ctx)
	if err != nil {
		t.Fatal(err)
	}
	mocks[0].Close()

	delay()

	// Forget peer so we can re-add one in same address/port
	f := func(t *testing.T, c *Cluster) {
		c.peerManager.RmPeer(clusters[0].id) // errors ignore for crdts
	}
	runF(t, clusters[1:], f)

	// Pin something on the rest
	pin2 := test.Cid2
	err = clusters[1].Pin(ctx, api.PinCid(pin2))
	if err != nil {
		t.Fatal(err)
	}

	pinDelay()

	// Rejoin c0
	c0, m0 := createOnePeerCluster(t, 0, testingClusterSecret)
	clusters[0] = c0
	mocks[0] = m0
	c0.consensus.Trust(ctx, clusters[1].id)
	err = c0.Join(ctx, clusterAddr(clusters[1]))
	if err != nil {
		t.Fatal(err)
	}

	delay()

	pinfo := clusters[0].tracker.Status(ctx, pin2)
	if pinfo.Status != api.TrackerStatusPinned {
		t.Error("re-joined cluster should have caught up")
	}

	pinfo = clusters[0].tracker.Status(ctx, pin1)
	if pinfo.Status != api.TrackerStatusPinned {
		t.Error("re-joined cluster should have original pin")
	}
}