|
| 1 | +package daemon |
| 2 | + |
| 3 | +import ( |
| 4 | + "bufio" |
| 5 | + "context" |
| 6 | + "encoding/json" |
| 7 | + "fmt" |
| 8 | + "log/slog" |
| 9 | + "os" |
| 10 | + "sync" |
| 11 | + "time" |
| 12 | + |
| 13 | + "github.com/malamtime/cli/model" |
| 14 | +) |
| 15 | + |
| 16 | +const ( |
| 17 | + // HeartbeatResyncInterval is the interval for retrying failed heartbeats |
| 18 | + HeartbeatResyncInterval = 30 * time.Minute |
| 19 | +) |
| 20 | + |
| 21 | +// HeartbeatResyncService handles periodic resync of failed heartbeats |
| 22 | +type HeartbeatResyncService struct { |
| 23 | + config model.ShellTimeConfig |
| 24 | + ticker *time.Ticker |
| 25 | + stopChan chan struct{} |
| 26 | + wg sync.WaitGroup |
| 27 | +} |
| 28 | + |
| 29 | +// NewHeartbeatResyncService creates a new heartbeat resync service |
| 30 | +func NewHeartbeatResyncService(config model.ShellTimeConfig) *HeartbeatResyncService { |
| 31 | + return &HeartbeatResyncService{ |
| 32 | + config: config, |
| 33 | + stopChan: make(chan struct{}), |
| 34 | + } |
| 35 | +} |
| 36 | + |
| 37 | +// Start begins the periodic resync job |
| 38 | +func (s *HeartbeatResyncService) Start(ctx context.Context) error { |
| 39 | + s.ticker = time.NewTicker(HeartbeatResyncInterval) |
| 40 | + s.wg.Add(1) |
| 41 | + |
| 42 | + go func() { |
| 43 | + defer s.wg.Done() |
| 44 | + |
| 45 | + // Run once at startup |
| 46 | + s.resync(ctx) |
| 47 | + |
| 48 | + for { |
| 49 | + select { |
| 50 | + case <-s.ticker.C: |
| 51 | + s.resync(ctx) |
| 52 | + case <-s.stopChan: |
| 53 | + return |
| 54 | + case <-ctx.Done(): |
| 55 | + return |
| 56 | + } |
| 57 | + } |
| 58 | + }() |
| 59 | + |
| 60 | + slog.Info("Heartbeat resync service started", slog.Duration("interval", HeartbeatResyncInterval)) |
| 61 | + return nil |
| 62 | +} |
| 63 | + |
| 64 | +// Stop stops the resync service |
| 65 | +func (s *HeartbeatResyncService) Stop() { |
| 66 | + if s.ticker != nil { |
| 67 | + s.ticker.Stop() |
| 68 | + } |
| 69 | + close(s.stopChan) |
| 70 | + s.wg.Wait() |
| 71 | + slog.Info("Heartbeat resync service stopped") |
| 72 | +} |
| 73 | + |
| 74 | +// resync reads failed heartbeats from the log file and attempts to send them |
| 75 | +func (s *HeartbeatResyncService) resync(ctx context.Context) { |
| 76 | + logFilePath := os.ExpandEnv(fmt.Sprintf("%s/%s", "$HOME", model.HEARTBEAT_LOG_FILE)) |
| 77 | + |
| 78 | + // Check if file exists |
| 79 | + if _, err := os.Stat(logFilePath); os.IsNotExist(err) { |
| 80 | + slog.Debug("No heartbeat log file found, nothing to resync") |
| 81 | + return |
| 82 | + } |
| 83 | + |
| 84 | + // Read the file |
| 85 | + file, err := os.Open(logFilePath) |
| 86 | + if err != nil { |
| 87 | + slog.Error("Failed to open heartbeat log file for resync", slog.Any("err", err)) |
| 88 | + return |
| 89 | + } |
| 90 | + |
| 91 | + var lines []string |
| 92 | + scanner := bufio.NewScanner(file) |
| 93 | + for scanner.Scan() { |
| 94 | + line := scanner.Text() |
| 95 | + if line != "" { |
| 96 | + lines = append(lines, line) |
| 97 | + } |
| 98 | + } |
| 99 | + file.Close() |
| 100 | + |
| 101 | + if err := scanner.Err(); err != nil { |
| 102 | + slog.Error("Error reading heartbeat log file", slog.Any("err", err)) |
| 103 | + return |
| 104 | + } |
| 105 | + |
| 106 | + if len(lines) == 0 { |
| 107 | + slog.Debug("No failed heartbeats to resync") |
| 108 | + return |
| 109 | + } |
| 110 | + |
| 111 | + slog.Info("Starting heartbeat resync", slog.Int("pendingCount", len(lines))) |
| 112 | + |
| 113 | + // Process each line |
| 114 | + var failedLines []string |
| 115 | + successCount := 0 |
| 116 | + |
| 117 | + for _, line := range lines { |
| 118 | + var payload model.HeartbeatPayload |
| 119 | + if err := json.Unmarshal([]byte(line), &payload); err != nil { |
| 120 | + slog.Error("Failed to parse heartbeat line, discarding", slog.Any("err", err), slog.String("line", line)) |
| 121 | + continue |
| 122 | + } |
| 123 | + |
| 124 | + // Try to send to server |
| 125 | + if err := model.SendHeartbeatsToServer(ctx, s.config, payload); err != nil { |
| 126 | + slog.Warn("Failed to resync heartbeat, keeping for next retry", slog.Any("err", err)) |
| 127 | + failedLines = append(failedLines, line) |
| 128 | + } else { |
| 129 | + successCount++ |
| 130 | + } |
| 131 | + } |
| 132 | + |
| 133 | + // Rewrite the file with only failed lines |
| 134 | + if err := s.rewriteLogFile(logFilePath, failedLines); err != nil { |
| 135 | + slog.Error("Failed to update heartbeat log file", slog.Any("err", err)) |
| 136 | + return |
| 137 | + } |
| 138 | + |
| 139 | + slog.Info("Heartbeat resync completed", |
| 140 | + slog.Int("success", successCount), |
| 141 | + slog.Int("remaining", len(failedLines))) |
| 142 | +} |
| 143 | + |
| 144 | +// rewriteLogFile atomically rewrites the log file with the given lines |
| 145 | +func (s *HeartbeatResyncService) rewriteLogFile(logFilePath string, lines []string) error { |
| 146 | + // If no lines remaining, remove the file |
| 147 | + if len(lines) == 0 { |
| 148 | + if err := os.Remove(logFilePath); err != nil && !os.IsNotExist(err) { |
| 149 | + return fmt.Errorf("failed to remove empty log file: %w", err) |
| 150 | + } |
| 151 | + return nil |
| 152 | + } |
| 153 | + |
| 154 | + // Write to temp file first |
| 155 | + tempFile := logFilePath + ".tmp" |
| 156 | + file, err := os.OpenFile(tempFile, os.O_CREATE|os.O_WRONLY|os.O_TRUNC, 0644) |
| 157 | + if err != nil { |
| 158 | + return fmt.Errorf("failed to create temp file: %w", err) |
| 159 | + } |
| 160 | + |
| 161 | + for _, line := range lines { |
| 162 | + if _, err := file.WriteString(line + "\n"); err != nil { |
| 163 | + file.Close() |
| 164 | + os.Remove(tempFile) |
| 165 | + return fmt.Errorf("failed to write to temp file: %w", err) |
| 166 | + } |
| 167 | + } |
| 168 | + |
| 169 | + if err := file.Close(); err != nil { |
| 170 | + os.Remove(tempFile) |
| 171 | + return fmt.Errorf("failed to close temp file: %w", err) |
| 172 | + } |
| 173 | + |
| 174 | + // Atomic rename |
| 175 | + if err := os.Rename(tempFile, logFilePath); err != nil { |
| 176 | + os.Remove(tempFile) |
| 177 | + return fmt.Errorf("failed to rename temp file: %w", err) |
| 178 | + } |
| 179 | + |
| 180 | + return nil |
| 181 | +} |
0 commit comments