// New creates an instance of a gossip node.
// The higher level manages the NodeIDContainer instance (which can be shared by
// various server components). The ambient context is expected to already
// contain the node ID.
func New(
ambient log.AmbientContext,
nodeID *base.NodeIDContainer,
rpcContext *rpc.Context,
grpcServer *grpc.Server,
resolvers []resolver.Resolver,
stopper *stop.Stopper,
registry *metric.Registry,
) *Gossip {
ambient.SetEventLog("gossip", "gossip")
g := &Gossip{
server: newServer(ambient, nodeID, stopper, registry),
Connected: make(chan struct{}),
rpcContext: rpcContext,
outgoing: makeNodeSet(minPeers, metric.NewGauge(MetaConnectionsOutgoingGauge)),
bootstrapping: map[string]struct{}{},
disconnected: make(chan *client, 10),
stalledCh: make(chan struct{}, 1),
stallInterval: defaultStallInterval,
bootstrapInterval: defaultBootstrapInterval,
cullInterval: defaultCullInterval,
nodeDescs: map[roachpb.NodeID]*roachpb.NodeDescriptor{},
resolverAddrs: map[util.UnresolvedAddr]resolver.Resolver{},
bootstrapAddrs: map[util.UnresolvedAddr]roachpb.NodeID{},
}
stopper.AddCloser(stop.CloserFn(g.server.AmbientContext.FinishEventLog))
registry.AddMetric(g.outgoing.gauge)
g.clientsMu.breakers = map[string]*circuit.Breaker{}
resolverAddrs := make([]string, len(resolvers))
for i, resolver := range resolvers {
resolverAddrs[i] = resolver.Addr()
}
ctx := g.AnnotateCtx(context.Background())
if log.V(1) {
log.Infof(ctx, "initial resolvers: %v", resolverAddrs)
}
g.SetResolvers(resolvers)
g.mu.Lock()
// Add ourselves as a SystemConfig watcher.
g.mu.is.registerCallback(KeySystemConfig, g.updateSystemConfig)
// Add ourselves as a node descriptor watcher.
g.mu.is.registerCallback(MakePrefixPattern(KeyNodeIDPrefix), g.updateNodeAddress)
g.mu.Unlock()
RegisterGossipServer(grpcServer, g.server)
return g
}
func TestGossipOrphanedStallDetection(t *testing.T) {
defer leaktest.AfterTest(t)()
stopper := stop.NewStopper()
defer stopper.Stop()
local := startGossip(1, stopper, t, metric.NewRegistry())
local.SetStallInterval(5 * time.Millisecond)
// Make sure we have the sentinel to ensure that its absence is not the
// cause of stall detection.
if err := local.AddInfo(KeySentinel, nil, time.Hour); err != nil {
t.Fatal(err)
}
peerStopper := stop.NewStopper()
peer := startGossip(2, peerStopper, t, metric.NewRegistry())
peerNodeID := peer.NodeID.Get()
peerAddr := peer.GetNodeAddr()
peerAddrStr := peerAddr.String()
local.startClient(peerAddr)
util.SucceedsSoon(t, func() error {
for _, peerID := range local.Outgoing() {
if peerID == peerNodeID {
return nil
}
}
return errors.Errorf("node %d not yet connected", peerNodeID)
})
util.SucceedsSoon(t, func() error {
for _, resolver := range local.GetResolvers() {
if resolver.Addr() == peerAddrStr {
return nil
}
}
return errors.Errorf("node %d descriptor not yet available", peerNodeID)
})
local.bootstrap()
local.manage()
peerStopper.Stop()
util.SucceedsSoon(t, func() error {
for _, peerID := range local.Outgoing() {
if peerID == peerNodeID {
return errors.Errorf("node %d still connected", peerNodeID)
}
}
return nil
})
peerStopper = stop.NewStopper()
defer peerStopper.Stop()
startGossipAtAddr(peerNodeID, peerAddr, peerStopper, t, metric.NewRegistry())
util.SucceedsSoon(t, func() error {
for _, peerID := range local.Outgoing() {
if peerID == peerNodeID {
return nil
}
}
return errors.Errorf("node %d not yet connected", peerNodeID)
})
}