lib/lsf/dispatch_test.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 package lsf
   6
   7 import (
   8         "context"
   9         "fmt"
  10         "math/rand"
  11         "os/exec"
  12         "strconv"
  13         "sync"
  14         "testing"
  15         "time"
  16
  17         "git.arvados.org/arvados.git/lib/config"
  18         "git.arvados.org/arvados.git/sdk/go/arvados"
  19         "git.arvados.org/arvados.git/sdk/go/arvadostest"
  20         "git.arvados.org/arvados.git/sdk/go/ctxlog"
  21         "github.com/prometheus/client_golang/prometheus"
  22         "gopkg.in/check.v1"
  23 )
  24
  25 func Test(t *testing.T) {
  26         check.TestingT(t)
  27 }
  28
  29 var _ = check.Suite(&suite{})
  30
  31 type suite struct {
  32         disp *dispatcher
  33 }
  34
  35 func (s *suite) TearDownTest(c *check.C) {
  36         arvados.NewClientFromEnv().RequestAndDecode(nil, "POST", "database/reset", nil, nil)
  37 }
  38
  39 func (s *suite) SetUpTest(c *check.C) {
  40         cfg, err := config.NewLoader(nil, ctxlog.TestLogger(c)).Load()
  41         c.Assert(err, check.IsNil)
  42         cluster, err := cfg.GetCluster("")
  43         c.Assert(err, check.IsNil)
  44         cluster.Containers.CloudVMs.PollInterval = arvados.Duration(time.Second)
  45         s.disp = newHandler(context.Background(), cluster, arvadostest.Dispatch1Token, prometheus.NewRegistry()).(*dispatcher)
  46         s.disp.lsfcli.stubCommand = func(string, ...string) *exec.Cmd {
  47                 return exec.Command("bash", "-c", "echo >&2 unimplemented stub; false")
  48         }
  49 }
  50
  51 type lsfstub struct {
  52         errorRate float64
  53 }
  54
  55 func (stub lsfstub) stubCommand(c *check.C) func(prog string, args ...string) *exec.Cmd {
  56         mtx := sync.Mutex{}
  57         nextjobid := 100
  58         fakejobq := map[int]string{}
  59         return func(prog string, args ...string) *exec.Cmd {
  60                 c.Logf("stubCommand: %q %q", prog, args)
  61                 if rand.Float64() < stub.errorRate {
  62                         return exec.Command("bash", "-c", "echo >&2 'stub random failure' && false")
  63                 }
  64                 switch prog {
  65                 case "bsub":
  66                         c.Assert(args, check.HasLen, 4)
  67                         c.Check(args[0], check.Equals, "-J")
  68                         switch args[1] {
  69                         case arvadostest.LockedContainerUUID:
  70                                 c.Check(args, check.DeepEquals, []string{"-J", arvadostest.LockedContainerUUID, "-R", "rusage[mem=11701MB:tmp=0MB] affinity[core(4)]"})
  71                                 mtx.Lock()
  72                                 fakejobq[nextjobid] = args[1]
  73                                 nextjobid++
  74                                 mtx.Unlock()
  75                         case arvadostest.QueuedContainerUUID:
  76                                 c.Check(args, check.DeepEquals, []string{"-J", arvadostest.QueuedContainerUUID, "-R", "rusage[mem=11701MB:tmp=45777MB] affinity[core(4)]"})
  77                                 mtx.Lock()
  78                                 fakejobq[nextjobid] = args[1]
  79                                 nextjobid++
  80                                 mtx.Unlock()
  81                         default:
  82                                 c.Errorf("unexpected uuid passed to bsub: args %q", args)
  83                                 return exec.Command("false")
  84                         }
  85                         return exec.Command("echo", "submitted job")
  86                 case "bjobs":
  87                         c.Check(args, check.DeepEquals, []string{"-noheader", "-o", "jobid stat job_name:30"})
  88                         out := ""
  89                         for jobid, uuid := range fakejobq {
  90                                 out += fmt.Sprintf(`%d %s %s\n`, jobid, "RUN", uuid)
  91                         }
  92                         c.Logf("bjobs out: %q", out)
  93                         return exec.Command("printf", out)
  94                 case "bkill":
  95                         killid, _ := strconv.Atoi(args[0])
  96                         if uuid, ok := fakejobq[killid]; !ok {
  97                                 return exec.Command("bash", "-c", fmt.Sprintf("printf >&2 'Job <%d>: No matching job found\n'", killid))
  98                         } else if uuid == "" {
  99                                 return exec.Command("bash", "-c", fmt.Sprintf("printf >&2 'Job <%d>: Job has already finished\n'", killid))
 100                         } else {
 101                                 go func() {
 102                                         time.Sleep(time.Millisecond)
 103                                         mtx.Lock()
 104                                         delete(fakejobq, killid)
 105                                         mtx.Unlock()
 106                                 }()
 107                                 return exec.Command("bash", "-c", fmt.Sprintf("printf 'Job <%d> is being terminated\n'", killid))
 108                         }
 109                 default:
 110                         return exec.Command("bash", "-c", fmt.Sprintf("echo >&2 'stub: command not found: %+q'", prog))
 111                 }
 112         }
 113 }
 114
 115 func (s *suite) TestSubmit(c *check.C) {
 116         s.disp.lsfcli.stubCommand = lsfstub{errorRate: 0.1}.stubCommand(c)
 117         s.disp.Start()
 118         deadline := time.Now().Add(20 * time.Second)
 119         for range time.NewTicker(time.Second).C {
 120                 if time.Now().After(deadline) {
 121                         c.Error("timed out")
 122                         break
 123                 }
 124                 // "queuedcontainer" should be running
 125                 if _, ok := s.disp.lsfqueue.JobID(arvadostest.QueuedContainerUUID); !ok {
 126                         continue
 127                 }
 128                 // "lockedcontainer" should be cancelled because it
 129                 // has priority 0 (no matching container requests)
 130                 if _, ok := s.disp.lsfqueue.JobID(arvadostest.LockedContainerUUID); ok {
 131                         continue
 132                 }
 133                 var ctr arvados.Container
 134                 if err := s.disp.arvDispatcher.Arv.Get("containers", arvadostest.LockedContainerUUID, nil, &ctr); err != nil {
 135                         c.Logf("error getting container state for %s: %s", arvadostest.LockedContainerUUID, err)
 136                         continue
 137                 }
 138                 if ctr.State != arvados.ContainerStateQueued {
 139                         c.Logf("LockedContainer is not in the LSF queue but its arvados record has not been updated to state==Queued (state is %q)", ctr.State)
 140                         continue
 141                 }
 142                 c.Log("reached desired state")
 143                 break
 144         }
 145 }