package monitor import ( "fmt" "math/rand" "sync" "time" "github.com/pkg/errors" "github.com/tendermint/tendermint/libs/log" tmtypes "github.com/tendermint/tendermint/types" ) // waiting more than this many seconds for a block means we're unhealthy const nodeLivenessTimeout = 5 * time.Second // Monitor keeps track of the nodes and updates common statistics upon // receiving new events from nodes. // // Common statistics is stored in Network struct. type Monitor struct { mtx sync.Mutex Nodes []*Node Network *Network monitorQuit chan struct{} // monitor exitting nodeQuit map[string]chan struct{} // node is being stopped and removed from under the monitor recalculateNetworkUptimeEvery time.Duration numValidatorsUpdateInterval time.Duration logger log.Logger } // NewMonitor creates new instance of a Monitor. You can provide options to // change some default values. // // Example: // NewMonitor(monitor.SetNumValidatorsUpdateInterval(1 * time.Second)) func NewMonitor(options ...func(*Monitor)) *Monitor { m := &Monitor{ Nodes: make([]*Node, 0), Network: NewNetwork(), monitorQuit: make(chan struct{}), nodeQuit: make(map[string]chan struct{}), recalculateNetworkUptimeEvery: 10 * time.Second, numValidatorsUpdateInterval: 5 * time.Second, logger: log.NewNopLogger(), } for _, option := range options { option(m) } return m } // RecalculateNetworkUptimeEvery lets you change network uptime update interval. func RecalculateNetworkUptimeEvery(d time.Duration) func(m *Monitor) { return func(m *Monitor) { m.recalculateNetworkUptimeEvery = d } } // SetNumValidatorsUpdateInterval lets you change num validators update interval. func SetNumValidatorsUpdateInterval(d time.Duration) func(m *Monitor) { return func(m *Monitor) { m.numValidatorsUpdateInterval = d } } // SetLogger lets you set your own logger func (m *Monitor) SetLogger(l log.Logger) { m.logger = l } // Monitor begins to monitor the node `n`. The node will be started and added // to the monitor. func (m *Monitor) Monitor(n *Node) error { m.mtx.Lock() m.Nodes = append(m.Nodes, n) m.mtx.Unlock() blockCh := make(chan *tmtypes.Block, 10) n.SendBlocksTo(blockCh) blockLatencyCh := make(chan float64, 10) n.SendBlockLatenciesTo(blockLatencyCh) disconnectCh := make(chan bool, 10) n.NotifyAboutDisconnects(disconnectCh) if err := n.Start(); err != nil { return err } m.Network.NewNode(n.Name) m.nodeQuit[n.Name] = make(chan struct{}) go m.listen(n.Name, blockCh, blockLatencyCh, disconnectCh, m.nodeQuit[n.Name]) return nil } // Unmonitor stops monitoring node `n`. The node will be stopped and removed // from the monitor. func (m *Monitor) Unmonitor(n *Node) { m.Network.NodeDeleted(n.Name) n.Stop() close(m.nodeQuit[n.Name]) delete(m.nodeQuit, n.Name) i, _ := m.NodeByName(n.Name) m.mtx.Lock() m.Nodes[i] = m.Nodes[len(m.Nodes)-1] m.Nodes = m.Nodes[:len(m.Nodes)-1] m.mtx.Unlock() } // NodeByName returns the node and its index if such node exists within the // monitor. Otherwise, -1 and nil are returned. func (m *Monitor) NodeByName(name string) (index int, node *Node) { m.mtx.Lock() defer m.mtx.Unlock() for i, n := range m.Nodes { if name == n.Name { return i, n } } return -1, nil } // NodeIsOnline is called when connection to the node is restored. // Must be safe to call multiple times. func (m *Monitor) NodeIsOnline(name string) { _, node := m.NodeByName(name) if nil != node { if online, ok := m.Network.nodeStatusMap[name]; ok && online { m.mtx.Lock() node.Online = online m.mtx.Unlock() } } } // Start starts the monitor's routines: recalculating network uptime and // updating number of validators. func (m *Monitor) Start() error { go m.recalculateNetworkUptimeLoop() go m.updateNumValidatorLoop() return nil } // Stop stops the monitor's routines. func (m *Monitor) Stop() { close(m.monitorQuit) for _, n := range m.Nodes { m.Unmonitor(n) } } // main loop where we listen for events from the node func (m *Monitor) listen( nodeName string, blockCh <-chan *tmtypes.Block, blockLatencyCh <-chan float64, disconnectCh <-chan bool, quit <-chan struct{}) { logger := m.logger.With("node", nodeName) for { select { case <-quit: return case b := <-blockCh: m.Network.NewBlock(b) m.Network.NodeIsOnline(nodeName) m.NodeIsOnline(nodeName) case l := <-blockLatencyCh: m.Network.NewBlockLatency(l) m.Network.NodeIsOnline(nodeName) m.NodeIsOnline(nodeName) case disconnected := <-disconnectCh: if disconnected { m.Network.NodeIsDown(nodeName) } else { m.Network.NodeIsOnline(nodeName) m.NodeIsOnline(nodeName) } case <-time.After(nodeLivenessTimeout): logger.Info("event", fmt.Sprintf("node was not responding for %v", nodeLivenessTimeout)) m.Network.NodeIsDown(nodeName) } } } // recalculateNetworkUptimeLoop every N seconds. func (m *Monitor) recalculateNetworkUptimeLoop() { for { select { case <-m.monitorQuit: return case <-time.After(m.recalculateNetworkUptimeEvery): m.Network.RecalculateUptime() } } } // updateNumValidatorLoop sends a request to a random node once every N seconds, // which in turn makes an RPC call to get the latest validators. func (m *Monitor) updateNumValidatorLoop() { rand.Seed(time.Now().Unix()) var height int64 var num int var err error for { m.mtx.Lock() nodesCount := len(m.Nodes) m.mtx.Unlock() if nodesCount == 0 { time.Sleep(m.numValidatorsUpdateInterval) continue } randomNodeIndex := rand.Intn(nodesCount) select { case <-m.monitorQuit: return case <-time.After(m.numValidatorsUpdateInterval): i := 0 m.mtx.Lock() for _, n := range m.Nodes { if i == randomNodeIndex { height, num, err = n.NumValidators() if err != nil { m.logger.Info("err", errors.Wrap(err, "update num validators failed")) } break } i++ } m.mtx.Unlock() m.Network.UpdateNumValidatorsForHeight(num, height) } } }