X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/b35dfa1f2b6c2fe57b7bc8a6e107425ed4e44f2a..6b3a880d607ee3e3dd273f019981fd6cae62373c:/lib/dispatchcloud/test/stub_driver.go diff --git a/lib/dispatchcloud/test/stub_driver.go b/lib/dispatchcloud/test/stub_driver.go index 132bd4d695..f57db0f09f 100644 --- a/lib/dispatchcloud/test/stub_driver.go +++ b/lib/dispatchcloud/test/stub_driver.go @@ -18,6 +18,7 @@ import ( "time" "git.arvados.org/arvados.git/lib/cloud" + "git.arvados.org/arvados.git/lib/crunchrun" "git.arvados.org/arvados.git/sdk/go/arvados" "github.com/sirupsen/logrus" "golang.org/x/crypto/ssh" @@ -131,7 +132,7 @@ func (sis *StubInstanceSet) Create(it arvados.InstanceType, image cloud.ImageID, tags: copyTags(tags), providerType: it.ProviderType, initCommand: cmd, - running: map[string]int64{}, + running: map[string]stubProcess{}, killing: map[string]bool{}, } svm.SSHService = SSHService{ @@ -189,8 +190,11 @@ type StubVM struct { CrunchRunMissing bool CrunchRunCrashRate float64 CrunchRunDetachDelay time.Duration + ArvMountMaxExitLag time.Duration + ArvMountDeadlockRate float64 ExecuteContainer func(arvados.Container) int CrashRunningContainer func(arvados.Container) + ExtraCrunchRunArgs string // extra args expected after "crunch-run --detach --stdin-config " sis *StubInstanceSet id cloud.InstanceID @@ -198,12 +202,21 @@ type StubVM struct { initCommand cloud.InitCommand providerType string SSHService SSHService - running map[string]int64 + running map[string]stubProcess killing map[string]bool lastPID int64 + deadlocked string sync.Mutex } +type stubProcess struct { + pid int64 + + // crunch-run has exited, but arv-mount process (or something) + // still holds lock in /var/run/ + exited bool +} + func (svm *StubVM) Instance() stubInstance { svm.Lock() defer svm.Unlock() @@ -240,15 +253,15 @@ func (svm *StubVM) Exec(env map[string]string, command string, stdin io.Reader, fmt.Fprint(stderr, "crunch-run: command not found\n") return 1 } - if strings.HasPrefix(command, "crunch-run --detach --stdin-env ") { - var stdinKV map[string]string - err := json.Unmarshal(stdinData, &stdinKV) + if strings.HasPrefix(command, "crunch-run --detach --stdin-config "+svm.ExtraCrunchRunArgs) { + var configData crunchrun.ConfigData + err := json.Unmarshal(stdinData, &configData) if err != nil { fmt.Fprintf(stderr, "unmarshal stdin: %s (stdin was: %q)\n", err, stdinData) return 1 } for _, name := range []string{"ARVADOS_API_HOST", "ARVADOS_API_TOKEN"} { - if stdinKV[name] == "" { + if configData.Env[name] == "" { fmt.Fprintf(stderr, "%s env var missing from stdin %q\n", name, stdinData) return 1 } @@ -256,7 +269,7 @@ func (svm *StubVM) Exec(env map[string]string, command string, stdin io.Reader, svm.Lock() svm.lastPID++ pid := svm.lastPID - svm.running[uuid] = pid + svm.running[uuid] = stubProcess{pid: pid} svm.Unlock() time.Sleep(svm.CrunchRunDetachDelay) fmt.Fprintf(stderr, "starting %s\n", uuid) @@ -273,14 +286,13 @@ func (svm *StubVM) Exec(env map[string]string, command string, stdin io.Reader, logger.Print("[test] exiting crunch-run stub") svm.Lock() defer svm.Unlock() - if svm.running[uuid] != pid { + if svm.running[uuid].pid != pid { bugf := svm.sis.driver.Bugf if bugf == nil { bugf = logger.Warnf } - bugf("[test] StubDriver bug or caller bug: pid %d exiting, running[%s]==%d", pid, uuid, svm.running[uuid]) - } else { - delete(svm.running, uuid) + bugf("[test] StubDriver bug or caller bug: pid %d exiting, running[%s].pid==%d", pid, uuid, svm.running[uuid].pid) + return } if !completed { logger.WithField("State", ctr.State).Print("[test] crashing crunch-run stub") @@ -288,6 +300,15 @@ func (svm *StubVM) Exec(env map[string]string, command string, stdin io.Reader, svm.CrashRunningContainer(ctr) } } + sproc := svm.running[uuid] + sproc.exited = true + svm.running[uuid] = sproc + svm.Unlock() + time.Sleep(svm.ArvMountMaxExitLag * time.Duration(math_rand.Float64())) + svm.Lock() + if math_rand.Float64() >= svm.ArvMountDeadlockRate { + delete(svm.running, uuid) + } }() crashluck := math_rand.Float64() @@ -333,26 +354,31 @@ func (svm *StubVM) Exec(env map[string]string, command string, stdin io.Reader, if command == "crunch-run --list" { svm.Lock() defer svm.Unlock() - for uuid := range svm.running { - fmt.Fprintf(stdout, "%s\n", uuid) + for uuid, sproc := range svm.running { + if sproc.exited { + fmt.Fprintf(stdout, "%s stale\n", uuid) + } else { + fmt.Fprintf(stdout, "%s\n", uuid) + } } if !svm.ReportBroken.IsZero() && svm.ReportBroken.Before(time.Now()) { fmt.Fprintln(stdout, "broken") } + fmt.Fprintln(stdout, svm.deadlocked) return 0 } if strings.HasPrefix(command, "crunch-run --kill ") { svm.Lock() - _, running := svm.running[uuid] - if running { + sproc, running := svm.running[uuid] + if running && !sproc.exited { svm.killing[uuid] = true svm.Unlock() time.Sleep(time.Duration(math_rand.Float64()*2) * time.Millisecond) svm.Lock() - _, running = svm.running[uuid] + sproc, running = svm.running[uuid] } svm.Unlock() - if running { + if running && !sproc.exited { fmt.Fprintf(stderr, "%s: container is running\n", uuid) return 1 }