9262: Stop setting sbatch --priority in crunch-dispatch-slurm.
[arvados.git] / services / crunch-dispatch-slurm / crunch-dispatch-slurm_test.go
1 package main
2
3 import (
4         "bytes"
5         "fmt"
6         "git.curoverse.com/arvados.git/sdk/go/arvados"
7         "git.curoverse.com/arvados.git/sdk/go/arvadosclient"
8         "git.curoverse.com/arvados.git/sdk/go/arvadostest"
9         "git.curoverse.com/arvados.git/sdk/go/dispatch"
10         "io"
11         "log"
12         "net/http"
13         "net/http/httptest"
14         "os"
15         "os/exec"
16         "strings"
17         "testing"
18         "time"
19
20         . "gopkg.in/check.v1"
21 )
22
23 // Gocheck boilerplate
24 func Test(t *testing.T) {
25         TestingT(t)
26 }
27
28 var _ = Suite(&TestSuite{})
29 var _ = Suite(&MockArvadosServerSuite{})
30
31 type TestSuite struct{}
32 type MockArvadosServerSuite struct{}
33
34 var initialArgs []string
35
36 func (s *TestSuite) SetUpSuite(c *C) {
37         initialArgs = os.Args
38 }
39
40 func (s *TestSuite) TearDownSuite(c *C) {
41 }
42
43 func (s *TestSuite) SetUpTest(c *C) {
44         args := []string{"crunch-dispatch-slurm"}
45         os.Args = args
46
47         arvadostest.StartAPI()
48         os.Setenv("ARVADOS_API_TOKEN", arvadostest.Dispatch1Token)
49 }
50
51 func (s *TestSuite) TearDownTest(c *C) {
52         os.Args = initialArgs
53         arvadostest.StopAPI()
54 }
55
56 func (s *MockArvadosServerSuite) TearDownTest(c *C) {
57         arvadostest.ResetEnv()
58 }
59
60 func (s *TestSuite) TestIntegrationNormal(c *C) {
61         container := s.integrationTest(c, func() *exec.Cmd { return exec.Command("echo", "zzzzz-dz642-queuedcontainer") },
62                 []string(nil),
63                 func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
64                         dispatcher.UpdateState(container.UUID, dispatch.Running)
65                         time.Sleep(3 * time.Second)
66                         dispatcher.UpdateState(container.UUID, dispatch.Complete)
67                 })
68         c.Check(container.State, Equals, arvados.ContainerStateComplete)
69 }
70
71 func (s *TestSuite) TestIntegrationCancel(c *C) {
72
73         // Override sbatchCmd
74         var scancelCmdLine []string
75         defer func(orig func(arvados.Container) *exec.Cmd) {
76                 scancelCmd = orig
77         }(scancelCmd)
78         scancelCmd = func(container arvados.Container) *exec.Cmd {
79                 scancelCmdLine = scancelFunc(container).Args
80                 return exec.Command("echo")
81         }
82
83         container := s.integrationTest(c, func() *exec.Cmd { return exec.Command("echo", "zzzzz-dz642-queuedcontainer") },
84                 []string(nil),
85                 func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
86                         dispatcher.UpdateState(container.UUID, dispatch.Running)
87                         time.Sleep(1 * time.Second)
88                         dispatcher.Arv.Update("containers", container.UUID,
89                                 arvadosclient.Dict{
90                                         "container": arvadosclient.Dict{"priority": 0}},
91                                 nil)
92                 })
93         c.Check(container.State, Equals, arvados.ContainerStateCancelled)
94         c.Check(scancelCmdLine, DeepEquals, []string{"scancel", "--name=zzzzz-dz642-queuedcontainer"})
95 }
96
97 func (s *TestSuite) TestIntegrationMissingFromSqueue(c *C) {
98         container := s.integrationTest(c, func() *exec.Cmd { return exec.Command("echo") }, []string{"sbatch", "--share",
99                 fmt.Sprintf("--job-name=%s", "zzzzz-dz642-queuedcontainer"),
100                 fmt.Sprintf("--mem-per-cpu=%d", 2862),
101                 fmt.Sprintf("--cpus-per-task=%d", 4)},
102                 func(dispatcher *dispatch.Dispatcher, container arvados.Container) {
103                         dispatcher.UpdateState(container.UUID, dispatch.Running)
104                         time.Sleep(3 * time.Second)
105                         dispatcher.UpdateState(container.UUID, dispatch.Complete)
106                 })
107         c.Check(container.State, Equals, arvados.ContainerStateCancelled)
108 }
109
110 func (s *TestSuite) integrationTest(c *C,
111         newSqueueCmd func() *exec.Cmd,
112         sbatchCmdComps []string,
113         runContainer func(*dispatch.Dispatcher, arvados.Container)) arvados.Container {
114         arvadostest.ResetEnv()
115
116         arv, err := arvadosclient.MakeArvadosClient()
117         c.Assert(err, IsNil)
118
119         var sbatchCmdLine []string
120
121         // Override sbatchCmd
122         defer func(orig func(arvados.Container) *exec.Cmd) {
123                 sbatchCmd = orig
124         }(sbatchCmd)
125         sbatchCmd = func(container arvados.Container) *exec.Cmd {
126                 sbatchCmdLine = sbatchFunc(container).Args
127                 return exec.Command("sh")
128         }
129
130         // Override squeueCmd
131         defer func(orig func() *exec.Cmd) {
132                 squeueCmd = orig
133         }(squeueCmd)
134         squeueCmd = newSqueueCmd
135
136         // There should be no queued containers now
137         params := arvadosclient.Dict{
138                 "filters": [][]string{{"state", "=", "Queued"}},
139         }
140         var containers arvados.ContainerList
141         err = arv.List("containers", params, &containers)
142         c.Check(err, IsNil)
143         c.Check(len(containers.Items), Equals, 1)
144
145         echo := "echo"
146         crunchRunCommand = &echo
147
148         doneProcessing := make(chan struct{})
149         dispatcher := dispatch.Dispatcher{
150                 Arv:          arv,
151                 PollInterval: time.Duration(1) * time.Second,
152                 RunContainer: func(dispatcher *dispatch.Dispatcher,
153                         container arvados.Container,
154                         status chan arvados.Container) {
155                         go runContainer(dispatcher, container)
156                         run(dispatcher, container, status)
157                         doneProcessing <- struct{}{}
158                 },
159                 DoneProcessing: doneProcessing}
160
161         squeueUpdater.StartMonitor(time.Duration(500) * time.Millisecond)
162
163         err = dispatcher.RunDispatcher()
164         c.Assert(err, IsNil)
165
166         squeueUpdater.Done()
167
168         c.Check(sbatchCmdLine, DeepEquals, sbatchCmdComps)
169
170         // There should be no queued containers now
171         err = arv.List("containers", params, &containers)
172         c.Check(err, IsNil)
173         c.Check(len(containers.Items), Equals, 0)
174
175         // Previously "Queued" container should now be in "Complete" state
176         var container arvados.Container
177         err = arv.Get("containers", "zzzzz-dz642-queuedcontainer", nil, &container)
178         c.Check(err, IsNil)
179         return container
180 }
181
182 func (s *MockArvadosServerSuite) Test_APIErrorGettingContainers(c *C) {
183         apiStubResponses := make(map[string]arvadostest.StubResponse)
184         apiStubResponses["/arvados/v1/api_client_authorizations/current"] = arvadostest.StubResponse{200, `{"uuid":"` + arvadostest.Dispatch1AuthUUID + `"}`}
185         apiStubResponses["/arvados/v1/containers"] = arvadostest.StubResponse{500, string(`{}`)}
186
187         testWithServerStub(c, apiStubResponses, "echo", "Error getting list of containers")
188 }
189
190 func testWithServerStub(c *C, apiStubResponses map[string]arvadostest.StubResponse, crunchCmd string, expected string) {
191         apiStub := arvadostest.ServerStub{apiStubResponses}
192
193         api := httptest.NewServer(&apiStub)
194         defer api.Close()
195
196         arv := arvadosclient.ArvadosClient{
197                 Scheme:    "http",
198                 ApiServer: api.URL[7:],
199                 ApiToken:  "abc123",
200                 Client:    &http.Client{Transport: &http.Transport{}},
201                 Retries:   0,
202         }
203
204         buf := bytes.NewBuffer(nil)
205         log.SetOutput(io.MultiWriter(buf, os.Stderr))
206         defer log.SetOutput(os.Stderr)
207
208         crunchRunCommand = &crunchCmd
209
210         doneProcessing := make(chan struct{})
211         dispatcher := dispatch.Dispatcher{
212                 Arv:          arv,
213                 PollInterval: time.Duration(1) * time.Second,
214                 RunContainer: func(dispatcher *dispatch.Dispatcher,
215                         container arvados.Container,
216                         status chan arvados.Container) {
217                         go func() {
218                                 time.Sleep(1 * time.Second)
219                                 dispatcher.UpdateState(container.UUID, dispatch.Running)
220                                 dispatcher.UpdateState(container.UUID, dispatch.Complete)
221                         }()
222                         run(dispatcher, container, status)
223                         doneProcessing <- struct{}{}
224                 },
225                 DoneProcessing: doneProcessing}
226
227         go func() {
228                 for i := 0; i < 80 && !strings.Contains(buf.String(), expected); i++ {
229                         time.Sleep(100 * time.Millisecond)
230                 }
231                 dispatcher.DoneProcessing <- struct{}{}
232         }()
233
234         err := dispatcher.RunDispatcher()
235         c.Assert(err, IsNil)
236
237         c.Check(buf.String(), Matches, `(?ms).*`+expected+`.*`)
238 }