|
| 1 | +package main |
| 2 | + |
| 3 | +import ( |
| 4 | + "bufio" |
| 5 | + "fmt" |
| 6 | + "io" |
| 7 | + "os" |
| 8 | + "os/exec" |
| 9 | + "os/signal" |
| 10 | + "sort" |
| 11 | + "strconv" |
| 12 | + "strings" |
| 13 | + "sync" |
| 14 | + "syscall" |
| 15 | + "time" |
| 16 | +) |
| 17 | + |
| 18 | +type ErrorTracker struct { |
| 19 | + Substring string |
| 20 | + Limit int |
| 21 | + Window time.Duration |
| 22 | + Errors []time.Time |
| 23 | +} |
| 24 | + |
| 25 | +func (et *ErrorTracker) AddError(t time.Time) { |
| 26 | + et.Errors = append(et.Errors, t) |
| 27 | + if len(et.Errors) > et.Limit { |
| 28 | + et.Errors = et.Errors[1:] |
| 29 | + } |
| 30 | +} |
| 31 | + |
| 32 | +func (et *ErrorTracker) ShouldRestartProcess() bool { |
| 33 | + if len(et.Errors) < et.Limit { |
| 34 | + return false |
| 35 | + } |
| 36 | + |
| 37 | + firstError := et.Errors[0] |
| 38 | + lastError := et.Errors[len(et.Errors)-1] |
| 39 | + return lastError.Sub(firstError) <= et.Window |
| 40 | +} |
| 41 | + |
| 42 | +func parseErrorTrackers() ([]ErrorTracker, []string) { |
| 43 | + trackers := make([]ErrorTracker, 0) |
| 44 | + args := os.Args[1:] |
| 45 | + maxArg := -1 |
| 46 | + |
| 47 | + for i, arg := range args { |
| 48 | + if arg == "--errortrack" { |
| 49 | + if i+3 >= len(args) { |
| 50 | + log("Invalid --errortrack parameters") |
| 51 | + os.Exit(1) |
| 52 | + } |
| 53 | + |
| 54 | + substring := args[i+1] |
| 55 | + limit, err := strconv.Atoi(args[i+2]) |
| 56 | + if err != nil { |
| 57 | + log("Error parsing limit: %v", err) |
| 58 | + os.Exit(1) |
| 59 | + } |
| 60 | + |
| 61 | + window, err := time.ParseDuration(args[i+3]) |
| 62 | + if err != nil { |
| 63 | + log("Error parsing time window: %v", err) |
| 64 | + os.Exit(1) |
| 65 | + } |
| 66 | + |
| 67 | + trackers = append(trackers, ErrorTracker{Substring: substring, Limit: limit, Window: window}) |
| 68 | + if i+3 > maxArg { |
| 69 | + maxArg = i + 3 |
| 70 | + } |
| 71 | + } |
| 72 | + } |
| 73 | + |
| 74 | + if maxArg >= len(args)-1 { |
| 75 | + log("No command provided") |
| 76 | + os.Exit(1) |
| 77 | + } |
| 78 | + |
| 79 | + cmdArgs := args[maxArg+1:] |
| 80 | + return trackers, cmdArgs |
| 81 | +} |
| 82 | + |
| 83 | +func log(format string, a ...interface{}) { |
| 84 | + fmt.Printf(fmt.Sprintf("[starter] %s ", time.Now().Format("[01-02|15:04:05.000]"))+format+"\n", a...) |
| 85 | +} |
| 86 | + |
| 87 | +func main() { |
| 88 | + trackers, cmdArgs := parseErrorTrackers() |
| 89 | + if len(cmdArgs) == 0 { |
| 90 | + log("No command provided") |
| 91 | + os.Exit(1) |
| 92 | + } |
| 93 | + |
| 94 | + cmdName := cmdArgs[0] |
| 95 | + cmdArgs = cmdArgs[1:] |
| 96 | + |
| 97 | + // Set up signal handling |
| 98 | + sigChan := make(chan os.Signal, 1) |
| 99 | + signal.Notify(sigChan, syscall.SIGINT, syscall.SIGTERM) |
| 100 | + restartChan := make(chan struct{}, 2) |
| 101 | + exitChan := make(chan struct{}, 2) |
| 102 | + var cmd *exec.Cmd |
| 103 | + var stdoutPipe, stderrPipe io.ReadCloser |
| 104 | + var wg sync.WaitGroup |
| 105 | + restartInit := false |
| 106 | + |
| 107 | + scanAndTrackErrors := func(scanner *bufio.Scanner, trackers []ErrorTracker, lastwriter io.Writer, restartChan, exitChan chan struct{}) { |
| 108 | + |
| 109 | + for scanner.Scan() { |
| 110 | + |
| 111 | + line := scanner.Text() |
| 112 | + |
| 113 | + fmt.Fprintln(lastwriter, line) |
| 114 | + |
| 115 | + if restartInit { |
| 116 | + continue |
| 117 | + } |
| 118 | + |
| 119 | + for i := range trackers { |
| 120 | + if strings.Contains(line, trackers[i].Substring) { |
| 121 | + trackers[i].AddError(time.Now()) |
| 122 | + |
| 123 | + if len(trackers[i].Errors) == 1 { |
| 124 | + log("Find error substring %q", trackers[i].Substring) |
| 125 | + } else if len(trackers[i].Errors) < trackers[i].Limit { |
| 126 | + log("Find error substring %q, %d errors in %s", trackers[i].Substring, len(trackers[i].Errors), trackers[i].Errors[len(trackers[i].Errors)-1].Sub(trackers[i].Errors[0]).Round(time.Second)) |
| 127 | + } else { |
| 128 | + // find the number of errors in the last window (last is the most recent) |
| 129 | + n := sort.Search(len(trackers[i].Errors), func(j int) bool { |
| 130 | + return trackers[i].Errors[len(trackers[i].Errors)-1].Sub(trackers[i].Errors[j]) <= trackers[i].Window |
| 131 | + }) |
| 132 | + log("Find error substring %q, %d errors in %s (%d errors in last %s)", trackers[i].Substring, len(trackers[i].Errors), |
| 133 | + trackers[i].Errors[len(trackers[i].Errors)-1].Sub(trackers[i].Errors[0]).Round(time.Second), |
| 134 | + len(trackers[i].Errors)-n, trackers[i].Window.Round(time.Second), |
| 135 | + ) |
| 136 | + } |
| 137 | + |
| 138 | + if trackers[i].ShouldRestartProcess() && !restartInit { |
| 139 | + log("Restarting process due to error limit for substring %q", trackers[i].Substring) |
| 140 | + restartInit = true |
| 141 | + restartChan <- struct{}{} |
| 142 | + } |
| 143 | + } |
| 144 | + } |
| 145 | + } |
| 146 | + if !restartInit { |
| 147 | + exitChan <- struct{}{} |
| 148 | + } |
| 149 | + } |
| 150 | + |
| 151 | + runCmd := func() { |
| 152 | + cmd = exec.Command(cmdName, cmdArgs...) |
| 153 | + stdoutPipe, _ = cmd.StdoutPipe() |
| 154 | + stderrPipe, _ = cmd.StderrPipe() |
| 155 | + restartInit = false |
| 156 | + |
| 157 | + wg.Add(2) |
| 158 | + go func() { |
| 159 | + defer wg.Done() |
| 160 | + scanner := bufio.NewScanner(stdoutPipe) |
| 161 | + scanAndTrackErrors(scanner, trackers, os.Stdout, restartChan, exitChan) |
| 162 | + }() |
| 163 | + |
| 164 | + go func() { |
| 165 | + defer wg.Done() |
| 166 | + scanner := bufio.NewScanner(stderrPipe) |
| 167 | + scanAndTrackErrors(scanner, trackers, os.Stderr, restartChan, exitChan) |
| 168 | + }() |
| 169 | + |
| 170 | + log("Starting process %q", cmdName) |
| 171 | + cmd.Start() |
| 172 | + |
| 173 | + } |
| 174 | + |
| 175 | + runCmd() |
| 176 | + |
| 177 | + for { |
| 178 | + select { |
| 179 | + case sig := <-sigChan: |
| 180 | + log("Received signal %v", sig) |
| 181 | + switch sig { |
| 182 | + case syscall.SIGINT, syscall.SIGTERM: |
| 183 | + log("Stopping %q process", cmdName) |
| 184 | + cmd.Process.Signal(sig) |
| 185 | + cmd.Wait() |
| 186 | + wg.Wait() |
| 187 | + time.Sleep(100 * time.Millisecond) |
| 188 | + os.Exit(0) |
| 189 | + } |
| 190 | + case <-restartChan: |
| 191 | + log("Too many errors, restarting process") |
| 192 | + if err := cmd.Process.Signal(syscall.SIGINT); err != nil { |
| 193 | + log("Error sending signal to process: %v", err) |
| 194 | + } |
| 195 | + cmd.Wait() |
| 196 | + wg.Wait() |
| 197 | + log("%q exited, restarting", cmdName) |
| 198 | + for i := range trackers { |
| 199 | + trackers[i].Errors = []time.Time{} |
| 200 | + } |
| 201 | + runCmd() |
| 202 | + case <-exitChan: |
| 203 | + log("%q exited, exiting", cmdName) |
| 204 | + cmd.Wait() |
| 205 | + wg.Wait() |
| 206 | + time.Sleep(100 * time.Millisecond) |
| 207 | + os.Exit(0) |
| 208 | + } |
| 209 | + } |
| 210 | +} |
0 commit comments