X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/7314917d65573b0e9d55f7b6522463c470356fba..47e59a35d5ed9b2cdb052894d741972324058505:/services/crunch-dispatch-slurm/crunch-dispatch-slurm_test.go

diff --git a/services/crunch-dispatch-slurm/crunch-dispatch-slurm_test.go b/services/crunch-dispatch-slurm/crunch-dispatch-slurm_test.go
index e7ccf25fc9..a823755379 100644
--- a/services/crunch-dispatch-slurm/crunch-dispatch-slurm_test.go
+++ b/services/crunch-dispatch-slurm/crunch-dispatch-slurm_test.go
@@ -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()
 }
 
@@ -58,58 +64,11 @@ func (s *MockArvadosServerSuite) TearDownTest(c *C) {
 	arvadostest.ResetEnv()
 }
 
-func (s *TestSuite) TestIntegrationNormal(c *C) {
-	container := s.integrationTest(c, func() *exec.Cmd { return exec.Command("echo", "zzzzz-dz642-queuedcontainer") },
-		[]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)
-		})
-	c.Check(container.State, Equals, arvados.ContainerStateComplete)
-}
-
-func (s *TestSuite) TestIntegrationCancel(c *C) {
-
-	// Override sbatchCmd
-	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") },
-		[]string(nil),
-		func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
-			dispatcher.UpdateState(container.UUID, dispatch.Running)
-			time.Sleep(1 * time.Second)
-			dispatcher.Arv.Update("containers", container.UUID,
-				arvadosclient.Dict{
-					"container": arvadosclient.Dict{"priority": 0}},
-				nil)
-		})
-	c.Check(container.State, Equals, arvados.ContainerStateCancelled)
-	c.Check(scancelCmdLine, DeepEquals, []string{"scancel", "--name=zzzzz-dz642-queuedcontainer"})
-}
-
-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)},
-		func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
-			dispatcher.UpdateState(container.UUID, dispatch.Running)
-			time.Sleep(3 * time.Second)
-			dispatcher.UpdateState(container.UUID, dispatch.Complete)
-		})
-	c.Check(container.State, Equals, arvados.ContainerStateCancelled)
-}
-
 func (s *TestSuite) integrationTest(c *C,
 	newSqueueCmd func() *exec.Cmd,
+	newScancelCmd func(arvados.Container) *exec.Cmd,
+	newSbatchCmd func(arvados.Container) *exec.Cmd,
+	newScontrolCmd func(arvados.Container) *exec.Cmd,
 	sbatchCmdComps []string,
 	runContainer func(*dispatch.Dispatcher, arvados.Container)) arvados.Container {
 	arvadostest.ResetEnv()
@@ -123,9 +82,14 @@ 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 +98,19 @@ 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
+
+	// Override scontrol
+	defer func(orig func(arvados.Container) *exec.Cmd) {
+		scontrolCmd = orig
+	}(scontrolCmd)
+	scontrolCmd = newScontrolCmd
+
+	// There should be one queued container
 	params := arvadosclient.Dict{
 		"filters": [][]string{{"state", "=", "Queued"}},
 	}
@@ -145,25 +121,29 @@ 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)
 
@@ -179,6 +159,112 @@ func (s *TestSuite) integrationTest(c *C,
 	return container
 }
 
+func (s *TestSuite) TestIntegrationNormal(c *C) {
+	done := false
+	container := s.integrationTest(c,
+		func() *exec.Cmd {
+			if done {
+				return exec.Command("true")
+			} else {
+				return exec.Command("echo", "zzzzz-dz642-queuedcontainer 9990 100")
+			}
+		},
+		nil,
+		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) {
+	var cmd *exec.Cmd
+	var scancelCmdLine []string
+	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 9990 100")
+			}
+		},
+		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,
+		nil,
+		[]string(nil),
+		func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
+			dispatcher.UpdateState(container.UUID, dispatch.Running)
+			time.Sleep(1 * time.Second)
+			dispatcher.Arv.Update("containers", container.UUID,
+				arvadosclient.Dict{
+					"container": arvadosclient.Dict{"priority": 0}},
+				nil)
+		})
+	c.Check(container.State, Equals, arvados.ContainerStateCancelled)
+	c.Check(scancelCmdLine, DeepEquals, []string{"scancel", "--name=zzzzz-dz642-queuedcontainer"})
+}
+
+func (s *TestSuite) TestIntegrationMissingFromSqueue(c *C) {
+	container := s.integrationTest(c,
+		func() *exec.Cmd { return exec.Command("echo") },
+		nil,
+		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),
+			fmt.Sprintf("--nice=%d", 9990)},
+		func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
+			dispatcher.UpdateState(container.UUID, dispatch.Running)
+			time.Sleep(3 * time.Second)
+			dispatcher.UpdateState(container.UUID, dispatch.Complete)
+		})
+	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")
+		},
+		nil,
+		[]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{
+		{"object_uuid", "=", container.UUID},
+		{"event_type", "=", "dispatch"},
+	}}, &ll)
+	c.Assert(len(ll.Items), Equals, 1)
+}
+
 func (s *MockArvadosServerSuite) TestAPIErrorGettingContainers(c *C) {
 	apiStubResponses := make(map[string]arvadostest.StubResponse)
 	apiStubResponses["/arvados/v1/api_client_authorizations/current"] = arvadostest.StubResponse{200, `{"uuid":"` + arvadostest.Dispatch1AuthUUID + `"}`}
@@ -207,32 +293,30 @@ 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+`.*`)
 }
@@ -305,13 +389,69 @@ func (s *MockArvadosServerSuite) TestSbatchFuncWithConfigArgs(c *C) {
 func testSbatchFuncWithArgs(c *C, args []string) {
 	theConfig.SbatchArguments = append(theConfig.SbatchArguments, args...)
 
-	container := arvados.Container{UUID: "123", RuntimeConstraints: arvados.RuntimeConstraints{RAM: 1000000, VCPUs: 2}}
+	container := arvados.Container{
+		UUID:               "123",
+		RuntimeConstraints: arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 2},
+		Priority:           1}
 	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=1", "--cpus-per-task=2")
+	expected = append(expected, "--job-name=123", "--mem=239", "--cpus-per-task=2", "--tmp=0", "--nice=9990")
+
+	c.Check(sbatchCmd.Args, DeepEquals, expected)
+}
+
+func (s *MockArvadosServerSuite) TestSbatchPartition(c *C) {
+	theConfig.SbatchArguments = nil
+	container := arvados.Container{
+		UUID:                 "123",
+		RuntimeConstraints:   arvados.RuntimeConstraints{RAM: 250000000, VCPUs: 1},
+		SchedulingParameters: arvados.SchedulingParameters{Partitions: []string{"blurb", "b2"}},
+		Priority:             1}
+	sbatchCmd := sbatchFunc(container)
+
+	var expected []string
+	expected = append(expected, "sbatch")
+	expected = append(expected, "--job-name=123", "--mem=239", "--cpus-per-task=1", "--tmp=0", "--nice=9990", "--partition=blurb,b2")
 
 	c.Check(sbatchCmd.Args, DeepEquals, expected)
 }
+
+func (s *TestSuite) TestIntegrationChangePriority(c *C) {
+	var scontrolCmdLine []string
+	step := 0
+
+	container := s.integrationTest(c,
+		func() *exec.Cmd {
+			if step == 0 {
+				return exec.Command("echo", "zzzzz-dz642-queuedcontainer 9990 100")
+			} else if step == 1 {
+				return exec.Command("echo", "zzzzz-dz642-queuedcontainer 4000 100")
+			} else {
+				return exec.Command("echo")
+			}
+		},
+		func(arvados.Container) *exec.Cmd { return exec.Command("true") },
+		nil,
+		func(container arvados.Container) *exec.Cmd {
+			scontrolCmdLine = scontrolFunc(container).Args
+			step = 1
+			return exec.Command("true")
+		},
+		[]string(nil),
+		func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
+			dispatcher.UpdateState(container.UUID, dispatch.Running)
+			time.Sleep(1 * time.Second)
+			dispatcher.Arv.Update("containers", container.UUID,
+				arvadosclient.Dict{
+					"container": arvadosclient.Dict{"priority": 600}},
+				nil)
+			time.Sleep(1 * time.Second)
+			step = 2
+			dispatcher.UpdateState(container.UUID, dispatch.Complete)
+		})
+	c.Check(container.State, Equals, arvados.ContainerStateComplete)
+	c.Check(scontrolCmdLine, DeepEquals, []string{"scontrol", "update", "JobName=zzzzz-dz642-queuedcontainer", "Nice=4000"})
+}