|
| 1 | +package orchestrator |
| 2 | + |
| 3 | +import ( |
| 4 | + "context" |
| 5 | + "math/big" |
| 6 | + "sync" |
| 7 | + "time" |
| 8 | + |
| 9 | + "github.com/rs/zerolog/log" |
| 10 | + "github.com/thirdweb-dev/indexer/internal/metrics" |
| 11 | + "github.com/thirdweb-dev/indexer/internal/rpc" |
| 12 | + "github.com/thirdweb-dev/indexer/internal/storage" |
| 13 | +) |
| 14 | + |
| 15 | +type WorkMode string |
| 16 | + |
| 17 | +const ( |
| 18 | + WORK_MODE_CHECK_INTERVAL = 10 * time.Minute |
| 19 | + WORK_MODE_BACKFILL_THRESHOLD = 500 |
| 20 | + WorkModeLive WorkMode = "live" |
| 21 | + WorkModeBackfill WorkMode = "backfill" |
| 22 | +) |
| 23 | + |
| 24 | +type WorkModeMonitor struct { |
| 25 | + rpc rpc.IRPCClient |
| 26 | + storage storage.IStorage |
| 27 | + workModeChannels map[chan WorkMode]struct{} |
| 28 | + channelsMutex sync.RWMutex |
| 29 | + currentMode WorkMode |
| 30 | +} |
| 31 | + |
| 32 | +func NewWorkModeMonitor(rpc rpc.IRPCClient, storage storage.IStorage) *WorkModeMonitor { |
| 33 | + return &WorkModeMonitor{ |
| 34 | + rpc: rpc, |
| 35 | + storage: storage, |
| 36 | + workModeChannels: make(map[chan WorkMode]struct{}), |
| 37 | + currentMode: "", |
| 38 | + } |
| 39 | +} |
| 40 | + |
| 41 | +// RegisterChannel adds a new channel to receive work mode updates |
| 42 | +func (m *WorkModeMonitor) RegisterChannel(ch chan WorkMode) { |
| 43 | + m.channelsMutex.Lock() |
| 44 | + defer m.channelsMutex.Unlock() |
| 45 | + |
| 46 | + m.workModeChannels[ch] = struct{}{} |
| 47 | + // Send current mode to the new channel only if it's not empty |
| 48 | + if m.currentMode != "" { |
| 49 | + select { |
| 50 | + case ch <- m.currentMode: |
| 51 | + log.Debug().Msg("Initial work mode sent to new channel") |
| 52 | + default: |
| 53 | + log.Warn().Msg("Failed to send initial work mode to new channel - channel full") |
| 54 | + } |
| 55 | + } |
| 56 | +} |
| 57 | + |
| 58 | +// UnregisterChannel removes a channel from receiving work mode updates |
| 59 | +func (m *WorkModeMonitor) UnregisterChannel(ch chan WorkMode) { |
| 60 | + m.channelsMutex.Lock() |
| 61 | + defer m.channelsMutex.Unlock() |
| 62 | + |
| 63 | + delete(m.workModeChannels, ch) |
| 64 | +} |
| 65 | + |
| 66 | +func (m *WorkModeMonitor) updateWorkModeMetric(mode WorkMode) { |
| 67 | + var value float64 |
| 68 | + if mode == WorkModeLive { |
| 69 | + value = 1 |
| 70 | + } |
| 71 | + metrics.CurrentWorkMode.Set(value) |
| 72 | +} |
| 73 | + |
| 74 | +func (m *WorkModeMonitor) Start(ctx context.Context) { |
| 75 | + // Perform immediate check |
| 76 | + newMode, err := m.determineWorkMode(ctx) |
| 77 | + if err != nil { |
| 78 | + log.Error().Err(err).Msg("Error checking work mode during startup") |
| 79 | + } else if newMode != m.currentMode { |
| 80 | + log.Info().Msgf("Work mode changing from %s to %s during startup", m.currentMode, newMode) |
| 81 | + m.currentMode = newMode |
| 82 | + m.updateWorkModeMetric(newMode) |
| 83 | + m.broadcastWorkMode(newMode) |
| 84 | + } |
| 85 | + |
| 86 | + ticker := time.NewTicker(WORK_MODE_CHECK_INTERVAL) |
| 87 | + defer ticker.Stop() |
| 88 | + |
| 89 | + log.Info().Msgf("Work mode monitor started with initial mode: %s", m.currentMode) |
| 90 | + |
| 91 | + for { |
| 92 | + select { |
| 93 | + case <-ctx.Done(): |
| 94 | + log.Info().Msg("Work mode monitor shutting down") |
| 95 | + return |
| 96 | + case <-ticker.C: |
| 97 | + newMode, err := m.determineWorkMode(ctx) |
| 98 | + if err != nil { |
| 99 | + log.Error().Err(err).Msg("Error checking work mode") |
| 100 | + continue |
| 101 | + } |
| 102 | + |
| 103 | + if newMode != m.currentMode { |
| 104 | + log.Info().Msgf("Work mode changing from %s to %s", m.currentMode, newMode) |
| 105 | + m.currentMode = newMode |
| 106 | + m.updateWorkModeMetric(newMode) |
| 107 | + m.broadcastWorkMode(newMode) |
| 108 | + } |
| 109 | + } |
| 110 | + } |
| 111 | +} |
| 112 | + |
| 113 | +func (m *WorkModeMonitor) broadcastWorkMode(mode WorkMode) { |
| 114 | + m.channelsMutex.RLock() |
| 115 | + defer m.channelsMutex.RUnlock() |
| 116 | + |
| 117 | + for ch := range m.workModeChannels { |
| 118 | + select { |
| 119 | + case ch <- mode: |
| 120 | + log.Debug().Msg("Work mode change notification sent") |
| 121 | + default: |
| 122 | + if r := recover(); r != nil { |
| 123 | + log.Warn().Msg("Work mode notification dropped - channel closed") |
| 124 | + delete(m.workModeChannels, ch) |
| 125 | + } |
| 126 | + } |
| 127 | + } |
| 128 | +} |
| 129 | + |
| 130 | +func (m *WorkModeMonitor) determineWorkMode(ctx context.Context) (WorkMode, error) { |
| 131 | + lastCommittedBlock, err := m.storage.MainStorage.GetMaxBlockNumber(m.rpc.GetChainID()) |
| 132 | + if err != nil { |
| 133 | + return "", err |
| 134 | + } |
| 135 | + |
| 136 | + if lastCommittedBlock.Sign() == 0 { |
| 137 | + log.Debug().Msg("No blocks committed yet, using backfill mode") |
| 138 | + return WorkModeBackfill, nil |
| 139 | + } |
| 140 | + |
| 141 | + latestBlock, err := m.rpc.GetLatestBlockNumber(ctx) |
| 142 | + if err != nil { |
| 143 | + return "", err |
| 144 | + } |
| 145 | + |
| 146 | + blockDiff := new(big.Int).Sub(latestBlock, lastCommittedBlock) |
| 147 | + log.Debug().Msgf("Committer is %d blocks behind the chain", blockDiff.Int64()) |
| 148 | + if blockDiff.Cmp(big.NewInt(WORK_MODE_BACKFILL_THRESHOLD)) < 0 { |
| 149 | + return WorkModeLive, nil |
| 150 | + } |
| 151 | + |
| 152 | + return WorkModeBackfill, nil |
| 153 | +} |
0 commit comments