func executeNecessityCheck(wg *sync.WaitGroup, cmd buddha.Command, check buddha.Check, done chan bool, fail chan error) {
defer wg.Done()
for i := 1; true; i++ {
log.Println(log.LevelInfo, "Check %s: checking...", check.String())
err := check.Execute(cmd.Timeout.Duration())
if err != nil {
switch e := err.(type) {
case buddha.CheckFalse:
log.Println(log.LevelInfo, "Check %s: deemed job unnecessary: %s", check.String(), e)
done <- false
return
default:
// unexpected failure
log.Println(log.LevelInfo, "Check %d/%d: %s: returned error: %s", i, cmd.Failures, check.String(), e)
if i < cmd.Failures {
log.Println(log.LevelInfo, "Check %d/%d: %s: waiting %s...", i, cmd.Failures, check.String(), cmd.Interval)
time.Sleep(cmd.Interval.Duration())
} else {
fail <- err
return
}
}
} else {
log.Println(log.LevelInfo, "Check %s: deemed job necessary", check.String())
done <- true
return
}
}
}
func main() {
var jobs buddha.Jobs
var err error
if *ConfigFile != "" {
// load manual job configuration file
jobs, err = buddha.OpenFile(*ConfigFile)
if err != nil {
log.Println(log.LevelFail, "fatal: could not read config file %s", *ConfigFile)
}
} else if *ConfigStdin {
// load job configuration from stdin
jobs, err = buddha.Open(os.Stdin)
if err != nil {
log.Println(log.LevelFail, "fatal: could not read config from STDIN")
}
} else {
jobs, err = buddha.OpenDir(*ConfigDir)
if err != nil {
log.Println(log.LevelFail, "fatal: could not read config directory %s", *ConfigDir)
}
}
if err != nil {
log.Println(log.LevelFail, "fatal: %s", err)
os.Exit(2)
return
}
// exit with status code of run
os.Exit(run(jobs))
}
func (c CheckTCP) Execute(timeout time.Duration) error {
conn, err := net.DialTimeout("tcp", c.Addr, timeout)
if err != nil {
log.Println(log.LevelInfo, "TCP connection failed: %s", err)
return CheckFalse(fmt.Sprintf("TCP connection failed: %s", err))
}
defer conn.Close()
return nil
}
func run(jobs buddha.Jobs) int {
lock, err := flock.Lock(*LockPath)
if err != nil {
if err == flock.ErrLocked {
log.Println(log.LevelFail, "fatal: another instance of buddha is running")
return 2
}
log.Println(log.LevelFail, "fatal: could not obtain exclusive lock at %s", *LockPath)
log.Println(log.LevelFail, "fatal: %s", err)
return 1
}
defer lock.Close()
// sort jobs by name
sort.Sort(jobs)
jobsToRun := flag.Args()
if len(jobsToRun) == 0 {
log.Println(log.LevelFail, "please specify job names, or 'all' to run all")
return 2
}
// if not running all jobs, filter job list
if jobsToRun[0] != "all" {
var missing []string
jobs, missing = jobs.Select(jobsToRun)
if len(missing) > 0 {
log.Println(log.LevelInfo, "info: missing jobs %v", missing)
}
}
// perform sanity checks against jobs
for i := 0; i < len(jobs); i++ {
if jobs[i].Root && (os.Getuid() != 0) {
log.Println(log.LevelFail, "fatal: job %s requires root privileges", jobs[i].Name)
return 1
}
}
// execute jobs
for i := 0; i < len(jobs); i++ {
err := runJob(jobs[i])
if err != nil {
log.Println(log.LevelFail, "fatal: job %s failed with unexpected error: %s", jobs[i].Name, err)
return 1
}
}
return 0
}
// pipe exec stdout to log
func execStdout(line string) {
log.Println(log.LevelInfo, line)
}
func runJob(job *buddha.Job) error {
log.Println(log.LevelPrim, "Job: %s", job.Name)
for _, cmd := range job.Commands {
log.Println(log.LevelPrim, "Command: %s", cmd.Name)
log.Println(log.LevelScnd, "Executing necessity checks")
isNecessaryResults, err := executeChecks(cmd, cmd.Necessity, executeNecessityCheck)
if err != nil {
log.Println(log.LevelFail, "fatal: unexpected error from necessity check, ending run")
return err
}
if allFalse(isNecessaryResults) {
switch *OnUnnecessary {
case ContinueBehaviour:
log.Println(log.LevelFail, "warning: job unnecessary, continuing anyway")
default:
log.Println(log.LevelInfo, "Job deemed unnecessary, skipping")
continue
}
}
// execute before health checks
// these will execute once and depending on --on-before-fail skip this job
log.Println(log.LevelScnd, "Executing before checks")
checksResults, err := executeChecks(cmd, cmd.Before, executeHealthCheck)
if err != nil {
log.Println(log.LevelFail, "fatal: unexpected error from before check, ending run")
return err
}
if anyFalse(checksResults) {
switch *OnBeforeFail {
case StopBehaviour:
log.Println(log.LevelFail, "fatal: before returned false, ending run")
return nil
case ContinueBehaviour:
log.Println(log.LevelFail, "warning: before returned false, continuing anyway")
default:
log.Println(log.LevelFail, "warning: before returned false, skipping job")
continue
}
}
// execute command
log.Println(log.LevelScnd, "Executing Command: %s %s", cmd.Path, strings.Join(cmd.Args, " "))
cmd.Stdout = execStdout
err = cmd.Execute()
if err != nil {
log.Println(log.LevelFail, "fatal: %s", err)
return err
}
// grace period between executing command and executing health checks/next command
log.Println(log.LevelInfo, "Waiting %s grace...", cmd.Grace)
time.Sleep(cmd.Grace.Duration())
// execute after health checks
log.Println(log.LevelScnd, "Executing after checks")
checksResults, err = executeChecks(cmd, cmd.After, executeHealthCheck)
if err != nil {
log.Println(log.LevelFail, "fatal: unexpected error from after check, ending run. err: %s", err)
return err
}
if anyFalse(checksResults) {
if *OnAfterFail == ContinueBehaviour {
log.Println(log.LevelFail, "warning: after checks failed, continuing anyway")
continue
}
log.Println(log.LevelFail, "fatal: after checks failed, ending run")
return err
}
}
return nil
}