X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/fc390927833d14b6c439db8ea72d3d52b60a5e6d..28f250ee0a43e873760d43b39119e5710de75fe8:/services/crunchstat/crunchstat.go diff --git a/services/crunchstat/crunchstat.go b/services/crunchstat/crunchstat.go index 510df3bac6..7e2dc01271 100644 --- a/services/crunchstat/crunchstat.go +++ b/services/crunchstat/crunchstat.go @@ -1,8 +1,13 @@ +// Copyright (C) The Arvados Authors. All rights reserved. +// +// SPDX-License-Identifier: AGPL-3.0 + package main import ( "bufio" "flag" + "fmt" "io" "log" "os" @@ -16,8 +21,11 @@ import ( const MaxLogLine = 1 << 14 // Child stderr lines >16KiB will be split -var signalOnDeadPPID int -var ppidCheckInterval = time.Second +var ( + signalOnDeadPPID int = 15 + ppidCheckInterval = time.Second + version = "dev" +) func main() { reporter := crunchstat.Reporter{ @@ -27,14 +35,25 @@ func main() { flag.StringVar(&reporter.CgroupRoot, "cgroup-root", "", "Root of cgroup tree") flag.StringVar(&reporter.CgroupParent, "cgroup-parent", "", "Name of container parent under cgroup") flag.StringVar(&reporter.CIDFile, "cgroup-cid", "", "Path to container id file") - flag.IntVar(&signalOnDeadPPID, "signal-on-dead-ppid", 15, "Signal to send child if crunchstat's parent process disappears") + flag.IntVar(&signalOnDeadPPID, "signal-on-dead-ppid", signalOnDeadPPID, "Signal to send child if crunchstat's parent process disappears (0 to disable)") flag.DurationVar(&ppidCheckInterval, "ppid-check-interval", ppidCheckInterval, "Time between checks for parent process disappearance") pollMsec := flag.Int64("poll", 1000, "Reporting interval, in milliseconds") + getVersion := flag.Bool("version", false, "Print version information and exit.") flag.Parse() + // Print version information if requested + if *getVersion { + fmt.Printf("crunchstat %s\n", version) + return + } + + reporter.Logger.Printf("crunchstat %s started", version) + if reporter.CgroupRoot == "" { reporter.Logger.Fatal("error: must provide -cgroup-root") + } else if signalOnDeadPPID < 0 { + reporter.Logger.Fatalf("-signal-on-dead-ppid=%d is invalid (use a positive signal number, or 0 to disable)", signalOnDeadPPID) } reporter.PollPeriod = time.Duration(*pollMsec) * time.Millisecond @@ -84,11 +103,11 @@ func runCommand(argv []string, logger *log.Logger) error { // Kill our child proc if our parent process disappears if signalOnDeadPPID != 0 { - go sendSignalOnDeadPPID(signalOnDeadPPID, os.Getppid(), cmd, logger) + go sendSignalOnDeadPPID(ppidCheckInterval, signalOnDeadPPID, os.Getppid(), cmd, logger) } // Funnel stderr through our channel - stderr_pipe, err := cmd.StderrPipe() + stderrPipe, err := cmd.StderrPipe() if err != nil { logger.Fatalln("error in StderrPipe:", err) } @@ -102,13 +121,14 @@ func runCommand(argv []string, logger *log.Logger) error { os.Stdin.Close() os.Stdout.Close() - copyPipeToChildLog(stderr_pipe, log.New(os.Stderr, "", 0)) + copyPipeToChildLog(stderrPipe, log.New(os.Stderr, "", 0)) return cmd.Wait() } -func sendSignalOnDeadPPID(signum, ppidOrig int, cmd *exec.Cmd, logger *log.Logger) { - for _ = range time.NewTicker(ppidCheckInterval).C { +func sendSignalOnDeadPPID(intvl time.Duration, signum, ppidOrig int, cmd *exec.Cmd, logger *log.Logger) { + ticker := time.NewTicker(intvl) + for range ticker.C { ppid := os.Getppid() if ppid == ppidOrig { continue @@ -120,9 +140,10 @@ func sendSignalOnDeadPPID(signum, ppidOrig int, cmd *exec.Cmd, logger *log.Logge logger.Printf("notice: crunchstat ppid changed from %d to %d -- killing child pid %d with signal %d", ppidOrig, ppid, cmd.Process.Pid, signum) err := cmd.Process.Signal(syscall.Signal(signum)) if err != nil { - logger.Printf("error: sending signal: %d", err) + logger.Printf("error: sending signal: %s", err) continue } + ticker.Stop() break } }