rebase drand setup onto master

lotus-soup/baseline.go

@@ -46,6 +46,7 @@ var baselineRoles = map[string]func(*TestEnvironment) error{
 	"bootstrapper": runBootstrapper,
 	"miner":        runMiner,
 	"client":       runBaselineClient,
+	"drand":        runDrandNode,
 }
 
 func runBaselineClient(t *TestEnvironment) error {
@@ -236,3 +237,16 @@ func extractCarData(ctx context.Context, rdata []byte, rpath string) []byte {
 	}
 	return rdata
 }
+
+func runDrandNode(t *TestEnvironment) error {
+	t.RecordMessage("running drand node")
+	_, err := prepareDrandNode(t)
+	if err != nil {
+		return err
+	}
+
+	// TODO add ability to halt / recover on demand
+	ctx := context.Background()
+	t.SyncClient.MustSignalAndWait(ctx, stateDone, t.TestInstanceCount)
+	return nil
+}

lotus-soup/compositions/composition.toml

@@ -5,7 +5,7 @@
 [global]
   plan = "lotus-soup"
   case = "lotus-baseline"
-  total_instances = 3
+  total_instances = 6
   builder = "docker:go"
   runner = "local:docker"
 
@@ -56,3 +56,15 @@
       miners = "1"
       balance = "2000000000"
       sectors = "10"
+
+[[groups]]
+  id = "drand"
+  [groups.resources]
+    memory = "120Mi"
+    cpu = "10m"
+  [groups.instances]
+    count = 3
+    percentage = 0.0
+  [groups.run]
+    [groups.run.test_params]
+      role = "drand"

lotus-soup/drand.go

@@ -0,0 +1,148 @@
+package main
+
+import (
+	"bytes"
+	"context"
+	"fmt"
+	"time"
+
+	"github.com/drand/drand/chain"
+	hclient "github.com/drand/drand/client/http"
+	"github.com/filecoin-project/lotus/node/modules/dtypes"
+	"github.com/testground/sdk-go/sync"
+
+	"github.com/drand/drand/demo/node"
+)
+
+var (
+	PrepareDrandTimeout = time.Minute
+	drandConfigTopic = sync.NewTopic("drand-config", &dtypes.DrandConfig{})
+)
+
+// waitForDrandConfig should be called by filecoin instances before constructing the lotus Node
+// you can use the returned dtypes.DrandConfig to override the default production config.
+func waitForDrandConfig(ctx context.Context, client sync.Client) (*dtypes.DrandConfig, error) {
+	ch := make(chan *dtypes.DrandConfig, 1)
+	sub := client.MustSubscribe(ctx, drandConfigTopic, ch)
+	select {
+	case cfg := <-ch:
+		return cfg, nil
+	case err := <-sub.Done():
+		return nil, err
+	}
+}
+
+// prepareDrandNode starts a drand instance and runs a DKG with the other members of the composition group.
+// Once the chain is running, the leader publishes the chain info needed by lotus nodes on
+// drandConfigTopic
+func prepareDrandNode(t *TestEnvironment) (node.Node, error) {
+	ctx, cancel := context.WithTimeout(context.Background(), PrepareDrandTimeout)
+	defer cancel()
+
+	startTime := time.Now()
+
+	seq := t.GroupSeq
+	isLeader := seq == 1
+	nNodes := t.TestGroupInstanceCount
+
+	myAddr := t.NetClient.MustGetDataNetworkIP()
+	// TODO: add test params for drand
+	period := "10s"
+	beaconOffset := 12
+	threshold := 2
+
+	// TODO(maybe): use TLS?
+	n := node.NewLocalNode(int(seq), period, "~/", false, myAddr.String())
+
+	// share the node addresses with other nodes
+	// TODO: if we implement TLS, this is where we'd share public TLS keys
+	type NodeAddr struct {
+		PrivateAddr string
+		PublicAddr  string
+		IsLeader    bool
+	}
+	addrTopic := sync.NewTopic("drand-addrs", &NodeAddr{})
+	var publicAddrs []string
+	var leaderAddr string
+	ch := make(chan *NodeAddr)
+	t.SyncClient.MustPublishSubscribe(ctx, addrTopic, &NodeAddr{
+		PrivateAddr: n.PrivateAddr(),
+		PublicAddr: n.PublicAddr(),
+		IsLeader: isLeader,
+	}, ch)
+	for i := 0; i < nNodes; i++ {
+		msg, ok := <-ch
+		if !ok {
+			return nil, fmt.Errorf("failed to read drand node addr from sync service")
+		}
+		publicAddrs = append(publicAddrs, fmt.Sprintf("http://%s", msg.PublicAddr))
+		if msg.IsLeader {
+			leaderAddr = msg.PrivateAddr
+		}
+	}
+	if leaderAddr == "" {
+		return nil, fmt.Errorf("got %d drand addrs, but no leader", len(publicAddrs))
+	}
+
+	t.SyncClient.MustSignalAndWait(ctx, "drand-start", nNodes)
+	t.RecordMessage("Starting drand sharing ceremony")
+	if err := n.Start("~/"); err != nil {
+		return nil, err
+	}
+
+	alive := false
+	waitSecs := 10
+	for i := 0; i < waitSecs; i++ {
+		if !n.Ping() {
+			time.Sleep(time.Second)
+			continue
+		}
+		t.R().RecordPoint("drand_first_ping", time.Now().Sub(startTime).Seconds())
+		alive = true
+		break
+	}
+	if !alive {
+		return nil, fmt.Errorf("drand node %d failed to start after %d seconds", t.GroupSeq, waitSecs)
+	}
+
+	// run DKG
+	t.SyncClient.MustSignalAndWait(ctx, "drand-dkg-start", nNodes)
+	grp := n.RunDKG(nNodes, threshold, period, isLeader, leaderAddr, beaconOffset)
+	if grp == nil {
+		return nil, fmt.Errorf("drand dkg failed")
+	}
+	t.R().RecordPoint("drand_dkg_complete", time.Now().Sub(startTime).Seconds())
+
+	// wait for chain to begin
+	to := time.Until(time.Unix(grp.GenesisTime, 0).Add(3 * time.Second).Add(grp.Period))
+	time.Sleep(to)
+
+
+
+	// verify that we can get a round of randomness from the chain using an http client
+	info := chain.NewChainInfo(grp)
+	client, err := hclient.NewWithInfo(publicAddrs[0], info, nil)
+	if err != nil {
+		return nil, fmt.Errorf("unable to create drand http client: %w", err)
+	}
+
+	_, err = client.Get(ctx, 1)
+	if err != nil {
+		return nil, fmt.Errorf("unable to get initial drand round: %w", err)
+	}
+
+	// if we're the leader, publish the config to the sync service
+	if isLeader {
+		buf := bytes.Buffer{}
+		if err := info.ToJSON(&buf); err != nil {
+			return nil, fmt.Errorf("error marshaling chain info: %w", err)
+		}
+		msg := dtypes.DrandConfig{
+			Servers:       publicAddrs,
+			ChainInfoJSON: buf.String(),
+		}
+		t.SyncClient.MustPublish(ctx, drandConfigTopic, &msg)
+	}
+
+	return n, nil
+}

lotus-soup/go.mod

@@ -4,10 +4,11 @@ go 1.14
 
 require (
 	github.com/davecgh/go-spew v1.1.1
+	github.com/drand/drand v0.9.2-0.20200616080806-a94e9c1636a4
 	github.com/filecoin-project/go-address v0.0.2-0.20200504173055-8b6f2fb2b3ef
 	github.com/filecoin-project/go-fil-markets v0.3.0
 	github.com/filecoin-project/go-storedcounter v0.0.0-20200421200003-1c99c62e8a5b
-	github.com/filecoin-project/lotus v0.4.1-0.20200623104442-68d38eff33e4
+	github.com/filecoin-project/lotus v0.4.1-0.20200623211458-e8642442267b
 	github.com/filecoin-project/specs-actors v0.6.2-0.20200617175406-de392ca14121
 	github.com/ipfs/go-cid v0.0.6
 	github.com/ipfs/go-datastore v0.4.4

lotus-soup/go.sum

@@ -246,8 +246,8 @@ github.com/filecoin-project/go-statestore v0.1.0 h1:t56reH59843TwXHkMcwyuayStBIi
 github.com/filecoin-project/go-statestore v0.1.0/go.mod h1:LFc9hD+fRxPqiHiaqUEZOinUJB4WARkRfNl10O7kTnI=
 github.com/filecoin-project/go-storedcounter v0.0.0-20200421200003-1c99c62e8a5b h1:fkRZSPrYpk42PV3/lIXiL0LHetxde7vyYYvSsttQtfg=
 github.com/filecoin-project/go-storedcounter v0.0.0-20200421200003-1c99c62e8a5b/go.mod h1:Q0GQOBtKf1oE10eSXSlhN45kDBdGvEcVOqMiffqX+N8=
-github.com/filecoin-project/lotus v0.4.1-0.20200623104442-68d38eff33e4 h1:h0iwqUzi4+E+xs1vLtmM2uxTUoTSNk4P/hFG7Dk9jnw=
+github.com/filecoin-project/lotus v0.4.1-0.20200623211458-e8642442267b h1:Oq1ABSZVYNFtvWqnj3bHINcw34T0if/0/zlsRq8rGgc=
-github.com/filecoin-project/lotus v0.4.1-0.20200623104442-68d38eff33e4/go.mod h1:uxvEKQiyuXisy/7MB6pFwb4WxxcA3UY3hyJnhL+k7M4=
+github.com/filecoin-project/lotus v0.4.1-0.20200623211458-e8642442267b/go.mod h1:uxvEKQiyuXisy/7MB6pFwb4WxxcA3UY3hyJnhL+k7M4=
 github.com/filecoin-project/sector-storage v0.0.0-20200615154852-728a47ab99d6/go.mod h1:M59QnAeA/oV+Z8oHFLoNpGMv0LZ8Rll+vHVXX7GirPM=
 github.com/filecoin-project/sector-storage v0.0.0-20200618073200-d9de9b7cb4b4 h1:lQC8Fbyn31/H4QxYAYwVV3PYZ9vS61EmjktZc5CaiYs=
 github.com/filecoin-project/sector-storage v0.0.0-20200618073200-d9de9b7cb4b4/go.mod h1:M59QnAeA/oV+Z8oHFLoNpGMv0LZ8Rll+vHVXX7GirPM=

lotus-soup/manifest.toml

@@ -27,4 +27,5 @@ instances = { min = 1, max = 100, default = 5 }
   miners = { type = "int", default = 1 }
   balance = { type = "int", default = 1 }
   sectors = { type = "int", default = 1 }
+  real_drand = { type = "bool", default = "false", desc = "set to true to use the real drand network config baked into lotus" }
   role = { type = "string" }

lotus-soup/node.go

@@ -116,6 +116,11 @@ func prepareBootstrapper(t *TestEnvironment) (*Node, error) {
 	miners := t.IntParam("miners")
 	nodes := clients + miners
 
+	drandOpt, err := getDrandConfig(ctx, t)
+	if err != nil {
+		return nil, err
+	}
+
 	// the first duty of the boostrapper is to construct the genesis block
 	// first collect all client and miner balances to assign initial funds
 	balances, err := waitForBalances(t, ctx, nodes)
@@ -177,6 +182,7 @@ func prepareBootstrapper(t *TestEnvironment) (*Node, error) {
 		withListenAddress(bootstrapperIP),
 		withBootstrapper(nil),
 		withPubsubConfig(true),
+		drandOpt,
 	)
 	if err != nil {
 		return nil, err
@@ -229,6 +235,11 @@ func prepareMiner(t *TestEnvironment) (*Node, error) {
 	ctx, cancel := context.WithTimeout(context.Background(), PrepareNodeTimeout)
 	defer cancel()
 
+	drandOpt, err := getDrandConfig(ctx, t)
+	if err != nil {
+		return nil, err
+	}
+
 	// first create a wallet
 	walletKey, err := wallet.GenerateKey(crypto.SigTypeBLS)
 	if err != nil {
@@ -342,6 +353,7 @@ func prepareMiner(t *TestEnvironment) (*Node, error) {
 		withListenAddress(minerIP),
 		withBootstrapper(genesisMsg.Bootstrapper),
 		withPubsubConfig(false),
+		drandOpt,
 	)
 	if err != nil {
 		return nil, err
@@ -446,6 +458,11 @@ func prepareClient(t *TestEnvironment) (*Node, error) {
 	ctx, cancel := context.WithTimeout(context.Background(), PrepareNodeTimeout)
 	defer cancel()
 
+	drandOpt, err := getDrandConfig(ctx, t)
+	if err != nil {
+		return nil, err
+	}
+
 	// first create a wallet
 	walletKey, err := wallet.GenerateKey(crypto.SigTypeBLS)
 	if err != nil {
@@ -475,6 +492,7 @@ func prepareClient(t *TestEnvironment) (*Node, error) {
 		withListenAddress(clientIP),
 		withBootstrapper(genesisMsg.Bootstrapper),
 		withPubsubConfig(false),
+		drandOpt,
 	)
 	if err != nil {
 		return nil, err
@@ -636,3 +654,25 @@ func collectClientAddrs(t *TestEnvironment, ctx context.Context, clients int) ([
 
 	return addrs, nil
 }
+
+func getDrandConfig(ctx context.Context, t *TestEnvironment) (node.Option, error) {
+	if t.BooleanParam("real_drand") {
+		noop := func(settings *node.Settings) error {
+			return nil
+		}
+		return noop, nil
+	}
+	cfg, err := waitForDrandConfig(ctx, t.SyncClient)
+	if err != nil {
+		t.RecordMessage("error getting drand config: %w", err)
+		return nil, err
+	}
+	t.RecordMessage("setting drand config: %v", cfg)
+
+	return node.Options(
+		node.Override(new(dtypes.DrandConfig), *cfg),
+
+		// FIXME: re-enable drand bootstrap peers once drand gossip relays are running in testground
+		node.Override(new(dtypes.DrandBootstrap), dtypes.DrandBootstrap{}),
+	), nil
+}