Merge branch '8784-dir-listings'
[arvados.git] / services / crunch-dispatch-slurm / crunch-dispatch-slurm_test.go
index fbea48e548a59f78718cb0afa419b5a84a1cd89b..5ab0e4fcb3092aa74743f58bdec05bc104359eb1 100644 (file)
@@ -1,12 +1,13 @@
+// Copyright (C) The Arvados Authors. All rights reserved.
+//
+// SPDX-License-Identifier: AGPL-3.0
+
 package main
 
 import (
        "bytes"
+       "context"
        "fmt"
-       "git.curoverse.com/arvados.git/sdk/go/arvados"
-       "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
-       "git.curoverse.com/arvados.git/sdk/go/arvadostest"
-       "git.curoverse.com/arvados.git/sdk/go/dispatch"
        "io"
        "io/ioutil"
        "log"
@@ -18,6 +19,10 @@ import (
        "testing"
        "time"
 
+       "git.curoverse.com/arvados.git/sdk/go/arvados"
+       "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
+       "git.curoverse.com/arvados.git/sdk/go/arvadostest"
+       "git.curoverse.com/arvados.git/sdk/go/dispatch"
        . "gopkg.in/check.v1"
 )
 
@@ -51,6 +56,7 @@ func (s *TestSuite) SetUpTest(c *C) {
 
 func (s *TestSuite) TearDownTest(c *C) {
        os.Args = initialArgs
+       arvadostest.ResetEnv()
        arvadostest.StopAPI()
 }
 
@@ -59,29 +65,50 @@ func (s *MockArvadosServerSuite) TearDownTest(c *C) {
 }
 
 func (s *TestSuite) TestIntegrationNormal(c *C) {
-       container := s.integrationTest(c, func() *exec.Cmd { return exec.Command("echo", "zzzzz-dz642-queuedcontainer") },
+       done := false
+       container := s.integrationTest(c,
+               func() *exec.Cmd {
+                       if done {
+                               return exec.Command("true")
+                       } else {
+                               return exec.Command("echo", "zzzzz-dz642-queuedcontainer")
+                       }
+               },
+               nil,
+               nil,
                []string(nil),
                func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
                        dispatcher.UpdateState(container.UUID, dispatch.Running)
                        time.Sleep(3 * time.Second)
                        dispatcher.UpdateState(container.UUID, dispatch.Complete)
+                       done = true
                })
        c.Check(container.State, Equals, arvados.ContainerStateComplete)
 }
 
 func (s *TestSuite) TestIntegrationCancel(c *C) {
-
-       // Override sbatchCmd
+       var cmd *exec.Cmd
        var scancelCmdLine []string
-       defer func(orig func(arvados.Container) *exec.Cmd) {
-               scancelCmd = orig
-       }(scancelCmd)
-       scancelCmd = func(container arvados.Container) *exec.Cmd {
-               scancelCmdLine = scancelFunc(container).Args
-               return exec.Command("echo")
-       }
-
-       container := s.integrationTest(c, func() *exec.Cmd { return exec.Command("echo", "zzzzz-dz642-queuedcontainer") },
+       attempt := 0
+
+       container := s.integrationTest(c,
+               func() *exec.Cmd {
+                       if cmd != nil && cmd.ProcessState != nil {
+                               return exec.Command("true")
+                       } else {
+                               return exec.Command("echo", "zzzzz-dz642-queuedcontainer")
+                       }
+               },
+               func(container arvados.Container) *exec.Cmd {
+                       if attempt++; attempt == 1 {
+                               return exec.Command("false")
+                       } else {
+                               scancelCmdLine = scancelFunc(container).Args
+                               cmd = exec.Command("echo")
+                               return cmd
+                       }
+               },
+               nil,
                []string(nil),
                func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
                        dispatcher.UpdateState(container.UUID, dispatch.Running)
@@ -96,10 +123,15 @@ func (s *TestSuite) TestIntegrationCancel(c *C) {
 }
 
 func (s *TestSuite) TestIntegrationMissingFromSqueue(c *C) {
-       container := s.integrationTest(c, func() *exec.Cmd { return exec.Command("echo") }, []string{"sbatch", "--share",
-               fmt.Sprintf("--job-name=%s", "zzzzz-dz642-queuedcontainer"),
-               fmt.Sprintf("--mem-per-cpu=%d", 2862),
-               fmt.Sprintf("--cpus-per-task=%d", 4)},
+       container := s.integrationTest(c,
+               func() *exec.Cmd { return exec.Command("echo") },
+               nil,
+               nil,
+               []string{"sbatch",
+                       fmt.Sprintf("--job-name=%s", "zzzzz-dz642-queuedcontainer"),
+                       fmt.Sprintf("--mem=%d", 11445),
+                       fmt.Sprintf("--cpus-per-task=%d", 4),
+                       fmt.Sprintf("--tmp=%d", 45777)},
                func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
                        dispatcher.UpdateState(container.UUID, dispatch.Running)
                        time.Sleep(3 * time.Second)
@@ -108,8 +140,35 @@ func (s *TestSuite) TestIntegrationMissingFromSqueue(c *C) {
        c.Check(container.State, Equals, arvados.ContainerStateCancelled)
 }
 
+func (s *TestSuite) TestSbatchFail(c *C) {
+       container := s.integrationTest(c,
+               func() *exec.Cmd { return exec.Command("echo") },
+               nil,
+               func(container arvados.Container) *exec.Cmd {
+                       return exec.Command("false")
+               },
+               []string(nil),
+               func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
+                       dispatcher.UpdateState(container.UUID, dispatch.Running)
+                       dispatcher.UpdateState(container.UUID, dispatch.Complete)
+               })
+       c.Check(container.State, Equals, arvados.ContainerStateComplete)
+
+       arv, err := arvadosclient.MakeArvadosClient()
+       c.Assert(err, IsNil)
+
+       var ll arvados.LogList
+       err = arv.List("logs", arvadosclient.Dict{"filters": [][]string{
+               []string{"object_uuid", "=", container.UUID},
+               []string{"event_type", "=", "dispatch"},
+       }}, &ll)
+       c.Assert(len(ll.Items), Equals, 1)
+}
+
 func (s *TestSuite) integrationTest(c *C,
        newSqueueCmd func() *exec.Cmd,
+       newScancelCmd func(arvados.Container) *exec.Cmd,
+       newSbatchCmd func(arvados.Container) *exec.Cmd,
        sbatchCmdComps []string,
        runContainer func(*dispatch.Dispatcher, arvados.Container)) arvados.Container {
        arvadostest.ResetEnv()
@@ -123,9 +182,14 @@ func (s *TestSuite) integrationTest(c *C,
        defer func(orig func(arvados.Container) *exec.Cmd) {
                sbatchCmd = orig
        }(sbatchCmd)
-       sbatchCmd = func(container arvados.Container) *exec.Cmd {
-               sbatchCmdLine = sbatchFunc(container).Args
-               return exec.Command("sh")
+
+       if newSbatchCmd != nil {
+               sbatchCmd = newSbatchCmd
+       } else {
+               sbatchCmd = func(container arvados.Container) *exec.Cmd {
+                       sbatchCmdLine = sbatchFunc(container).Args
+                       return exec.Command("sh")
+               }
        }
 
        // Override squeueCmd
@@ -134,7 +198,13 @@ func (s *TestSuite) integrationTest(c *C,
        }(squeueCmd)
        squeueCmd = newSqueueCmd
 
-       // There should be no queued containers now
+       // Override scancel
+       defer func(orig func(arvados.Container) *exec.Cmd) {
+               scancelCmd = orig
+       }(scancelCmd)
+       scancelCmd = newScancelCmd
+
+       // There should be one queued container
        params := arvadosclient.Dict{
                "filters": [][]string{{"state", "=", "Queued"}},
        }
@@ -145,25 +215,29 @@ func (s *TestSuite) integrationTest(c *C,
 
        theConfig.CrunchRunCommand = []string{"echo"}
 
-       doneProcessing := make(chan struct{})
+       ctx, cancel := context.WithCancel(context.Background())
+       doneRun := make(chan struct{})
+
        dispatcher := dispatch.Dispatcher{
-               Arv:          arv,
-               PollInterval: time.Duration(1) * time.Second,
-               RunContainer: func(dispatcher *dispatch.Dispatcher,
-                       container arvados.Container,
-                       status chan arvados.Container) {
-                       go runContainer(dispatcher, container)
-                       run(dispatcher, container, status)
-                       doneProcessing <- struct{}{}
+               Arv:        arv,
+               PollPeriod: time.Duration(1) * time.Second,
+               RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) {
+                       go func() {
+                               runContainer(disp, ctr)
+                               doneRun <- struct{}{}
+                       }()
+                       run(disp, ctr, status)
+                       cancel()
                },
-               DoneProcessing: doneProcessing}
+       }
 
-       squeueUpdater.StartMonitor(time.Duration(500) * time.Millisecond)
+       sqCheck = &SqueueChecker{Period: 500 * time.Millisecond}
 
-       err = dispatcher.RunDispatcher()
-       c.Assert(err, IsNil)
+       err = dispatcher.Run(ctx)
+       <-doneRun
+       c.Assert(err, Equals, context.Canceled)
 
-       squeueUpdater.Done()
+       sqCheck.Stop()
 
        c.Check(sbatchCmdLine, DeepEquals, sbatchCmdComps)
 
@@ -207,32 +281,30 @@ func testWithServerStub(c *C, apiStubResponses map[string]arvadostest.StubRespon
 
        theConfig.CrunchRunCommand = []string{crunchCmd}
 
-       doneProcessing := make(chan struct{})
+       ctx, cancel := context.WithCancel(context.Background())
        dispatcher := dispatch.Dispatcher{
-               Arv:          arv,
-               PollInterval: time.Duration(1) * time.Second,
-               RunContainer: func(dispatcher *dispatch.Dispatcher,
-                       container arvados.Container,
-                       status chan arvados.Container) {
+               Arv:        arv,
+               PollPeriod: time.Duration(1) * time.Second,
+               RunContainer: func(disp *dispatch.Dispatcher, ctr arvados.Container, status <-chan arvados.Container) {
                        go func() {
                                time.Sleep(1 * time.Second)
-                               dispatcher.UpdateState(container.UUID, dispatch.Running)
-                               dispatcher.UpdateState(container.UUID, dispatch.Complete)
+                               disp.UpdateState(ctr.UUID, dispatch.Running)
+                               disp.UpdateState(ctr.UUID, dispatch.Complete)
                        }()
-                       run(dispatcher, container, status)
-                       doneProcessing <- struct{}{}
+                       run(disp, ctr, status)
+                       cancel()
                },
-               DoneProcessing: doneProcessing}
+       }
 
        go func() {
                for i := 0; i < 80 && !strings.Contains(buf.String(), expected); i++ {
                        time.Sleep(100 * time.Millisecond)
                }
-               dispatcher.DoneProcessing <- struct{}{}
+               cancel()
        }()
 
-       err := dispatcher.RunDispatcher()
-       c.Assert(err, IsNil)
+       err := dispatcher.Run(ctx)
+       c.Assert(err, Equals, context.Canceled)
 
        c.Check(buf.String(), Matches, `(?ms).*`+expected+`.*`)
 }
@@ -309,9 +381,9 @@ func testSbatchFuncWithArgs(c *C, args []string) {
        sbatchCmd := sbatchFunc(container)
 
        var expected []string
-       expected = append(expected, "sbatch", "--share")
+       expected = append(expected, "sbatch")
        expected = append(expected, theConfig.SbatchArguments...)
-       expected = append(expected, "--job-name=123", "--mem-per-cpu=120", "--cpus-per-task=2")
+       expected = append(expected, "--job-name=123", "--mem=239", "--cpus-per-task=2", "--tmp=0")
 
        c.Check(sbatchCmd.Args, DeepEquals, expected)
 }
@@ -322,8 +394,8 @@ func (s *MockArvadosServerSuite) TestSbatchPartition(c *C) {
        sbatchCmd := sbatchFunc(container)
 
        var expected []string
-       expected = append(expected, "sbatch", "--share")
-       expected = append(expected, "--job-name=123", "--mem-per-cpu=239", "--cpus-per-task=1", "--partition=blurb,b2")
+       expected = append(expected, "sbatch")
+       expected = append(expected, "--job-name=123", "--mem=239", "--cpus-per-task=1", "--tmp=0", "--partition=blurb,b2")
 
        c.Check(sbatchCmd.Args, DeepEquals, expected)
 }