X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/674630a6f461527f5b26e917814736b444cb4f51..cb690390d4f253c3bbb9c543e243cf988f39fbb3:/lib/crunchrun/crunchrun_test.go diff --git a/lib/crunchrun/crunchrun_test.go b/lib/crunchrun/crunchrun_test.go index aaba1c4204..c533821351 100644 --- a/lib/crunchrun/crunchrun_test.go +++ b/lib/crunchrun/crunchrun_test.go @@ -6,6 +6,7 @@ package crunchrun import ( "bytes" + "context" "crypto/md5" "encoding/json" "errors" @@ -13,12 +14,19 @@ import ( "io" "io/ioutil" "log" + "math/rand" + "net/http" + "net/http/httptest" + "net/http/httputil" + "net/url" "os" "os/exec" + "path" "regexp" "runtime/pprof" "strings" "sync" + "sync/atomic" "syscall" "testing" "time" @@ -29,9 +37,10 @@ import ( "git.arvados.org/arvados.git/sdk/go/arvadosclient" "git.arvados.org/arvados.git/sdk/go/arvadostest" "git.arvados.org/arvados.git/sdk/go/manifest" - "golang.org/x/net/context" . "gopkg.in/check.v1" + git_client "gopkg.in/src-d/go-git.v4/plumbing/transport/client" + git_http "gopkg.in/src-d/go-git.v4/plumbing/transport/http" ) // Gocheck boilerplate @@ -39,6 +48,8 @@ func TestCrunchExec(t *testing.T) { TestingT(t) } +const logLineStart = `(?m)(.*\n)*\d{4}-\d\d-\d\dT\d\d:\d\d:\d\d\.\d+Z ` + var _ = Suite(&TestSuite{}) type TestSuite struct { @@ -50,6 +61,20 @@ type TestSuite struct { keepmountTmp []string testDispatcherKeepClient KeepTestClient testContainerKeepClient KeepTestClient + debian12MemoryCurrent int64 + debian12SwapCurrent int64 +} + +func (s *TestSuite) SetUpSuite(c *C) { + buf, err := os.ReadFile("../crunchstat/testdata/debian12/sys/fs/cgroup/user.slice/user-1000.slice/session-4.scope/memory.current") + c.Assert(err, IsNil) + _, err = fmt.Sscanf(string(buf), "%d", &s.debian12MemoryCurrent) + c.Assert(err, IsNil) + + buf, err = os.ReadFile("../crunchstat/testdata/debian12/sys/fs/cgroup/user.slice/user-1000.slice/session-4.scope/memory.swap.current") + c.Assert(err, IsNil) + _, err = fmt.Sscanf(string(buf), "%d", &s.debian12SwapCurrent) + c.Assert(err, IsNil) } func (s *TestSuite) SetUpTest(c *C) { @@ -121,7 +146,7 @@ type stubExecutor struct { stopErr error stopped bool closed bool - runFunc func() + runFunc func() int exit chan int } @@ -133,10 +158,14 @@ func (e *stubExecutor) LoadImage(imageId string, tarball string, container arvad func (e *stubExecutor) Runtime() string { return "stub" } func (e *stubExecutor) Version() string { return "stub " + cmd.Version.String() } func (e *stubExecutor) Create(spec containerSpec) error { e.created = spec; return e.createErr } -func (e *stubExecutor) Start() error { e.exit = make(chan int, 1); go e.runFunc(); return e.startErr } -func (e *stubExecutor) CgroupID() string { return "cgroupid" } -func (e *stubExecutor) Stop() error { e.stopped = true; go func() { e.exit <- -1 }(); return e.stopErr } -func (e *stubExecutor) Close() { e.closed = true } +func (e *stubExecutor) Start() error { + e.exit = make(chan int, 1) + go func() { e.exit <- e.runFunc() }() + return e.startErr +} +func (e *stubExecutor) Pid() int { return 1115883 } // matches pid in ../crunchstat/testdata/debian12/proc/ +func (e *stubExecutor) Stop() error { e.stopped = true; go func() { e.exit <- -1 }(); return e.stopErr } +func (e *stubExecutor) Close() { e.closed = true } func (e *stubExecutor) Wait(context.Context) (int, error) { return <-e.exit, e.waitErr } @@ -404,6 +433,67 @@ func (client *KeepTestClient) ManifestFileReader(m manifest.Manifest, filename s return nil, nil } +type apiStubServer struct { + server *httptest.Server + proxy *httputil.ReverseProxy + intercept func(http.ResponseWriter, *http.Request) bool + + container arvados.Container + logs map[string]string +} + +func apiStub() (*arvados.Client, *apiStubServer) { + client := arvados.NewClientFromEnv() + apistub := &apiStubServer{} + apistub.server = httptest.NewTLSServer(apistub) + apistub.proxy = httputil.NewSingleHostReverseProxy(&url.URL{Scheme: "https", Host: client.APIHost}) + if client.Insecure { + apistub.proxy.Transport = arvados.InsecureHTTPClient.Transport + } + client.APIHost = apistub.server.Listener.Addr().String() + return client, apistub +} + +func (apistub *apiStubServer) ServeHTTP(w http.ResponseWriter, r *http.Request) { + if apistub.intercept != nil && apistub.intercept(w, r) { + return + } + if r.Method == "POST" && r.URL.Path == "/arvados/v1/logs" { + var body struct { + Log struct { + EventType string `json:"event_type"` + Properties struct { + Text string + } + } + } + json.NewDecoder(r.Body).Decode(&body) + apistub.logs[body.Log.EventType] += body.Log.Properties.Text + return + } + if r.Method == "GET" && r.URL.Path == "/arvados/v1/collections/"+hwPDH { + json.NewEncoder(w).Encode(arvados.Collection{ManifestText: hwManifest}) + return + } + if r.Method == "GET" && r.URL.Path == "/arvados/v1/collections/"+otherPDH { + json.NewEncoder(w).Encode(arvados.Collection{ManifestText: otherManifest}) + return + } + if r.Method == "GET" && r.URL.Path == "/arvados/v1/collections/"+normalizedWithSubdirsPDH { + json.NewEncoder(w).Encode(arvados.Collection{ManifestText: normalizedManifestWithSubdirs}) + return + } + if r.Method == "GET" && r.URL.Path == "/arvados/v1/collections/"+denormalizedWithSubdirsPDH { + json.NewEncoder(w).Encode(arvados.Collection{ManifestText: denormalizedManifestWithSubdirs}) + return + } + if r.Method == "GET" && r.URL.Path == "/arvados/v1/containers/"+apistub.container.UUID { + json.NewEncoder(w).Encode(apistub.container) + return + } + apistub.proxy.ServeHTTP(w, r) +} + func (s *TestSuite) TestLoadImage(c *C) { s.runner.Container.ContainerImage = arvadostest.DockerImage112PDH s.runner.Container.Mounts = map[string]arvados.Mount{ @@ -541,9 +631,9 @@ func dockerLog(fd byte, msg string) []byte { } func (s *TestSuite) TestRunContainer(c *C) { - s.executor.runFunc = func() { + s.executor.runFunc = func() int { fmt.Fprintf(s.executor.created.Stdout, "Hello world\n") - s.executor.exit <- 0 + return 0 } var logs TestLogs @@ -644,7 +734,7 @@ func (s *TestSuite) TestUpdateContainerCancelled(c *C) { // Used by the TestFullRun*() test below to DRY up boilerplate setup to do full // dress rehearsal of the Run() function, starting from a JSON container record. -func (s *TestSuite) fullRunHelper(c *C, record string, extraMounts []string, exitCode int, fn func()) (*ArvTestClient, *ContainerRunner, string) { +func (s *TestSuite) fullRunHelper(c *C, record string, extraMounts []string, fn func() int) (*ArvTestClient, *ContainerRunner, string) { err := json.Unmarshal([]byte(record), &s.api.Container) c.Assert(err, IsNil) initialState := s.api.Container.State @@ -658,10 +748,7 @@ func (s *TestSuite) fullRunHelper(c *C, record string, extraMounts []string, exi c.Assert(err, IsNil) c.Logf("SecretMounts decoded %v json %q", sm, secretMounts) - s.executor.runFunc = func() { - fn() - s.executor.exit <- exitCode - } + s.executor.runFunc = fn s.runner.statInterval = 100 * time.Millisecond s.runner.containerWatchdogInterval = time.Second @@ -678,8 +765,9 @@ func (s *TestSuite) fullRunHelper(c *C, record string, extraMounts []string, exi } return d, err } + client, _ := apiStub() s.runner.MkArvClient = func(token string) (IArvadosClient, IKeepClient, *arvados.Client, error) { - return &ArvTestClient{secretMounts: secretMounts}, &s.testContainerKeepClient, nil, nil + return &ArvTestClient{secretMounts: secretMounts}, &s.testContainerKeepClient, client, nil } if extraMounts != nil && len(extraMounts) > 0 { @@ -732,7 +820,7 @@ func (s *TestSuite) TestFullRunHello(c *C) { "runtime_constraints": {"vcpus":1,"ram":1000000}, "state": "Locked", "output_storage_classes": ["default"] -}`, nil, 0, func() { +}`, nil, func() int { c.Check(s.executor.created.Command, DeepEquals, []string{"echo", "hello world"}) c.Check(s.executor.created.Image, Equals, "sha256:d8309758b8fe2c81034ffc8a10c36460b77db7bc5e7b448c4e5b684f9d95a678") c.Check(s.executor.created.Env, DeepEquals, map[string]string{"foo": "bar", "baz": "waz"}) @@ -742,6 +830,7 @@ func (s *TestSuite) TestFullRunHello(c *C) { c.Check(s.executor.created.EnableNetwork, Equals, false) c.Check(s.executor.created.CUDADeviceCount, Equals, 0) fmt.Fprintln(s.executor.created.Stdout, "hello world") + return 0 }) c.Check(s.api.CalledWith("container.exit_code", 0), NotNil) @@ -764,14 +853,92 @@ func (s *TestSuite) TestRunAlreadyRunning(c *C) { "runtime_constraints": {}, "scheduling_parameters":{"max_run_time": 1}, "state": "Running" -}`, nil, 2, func() { +}`, nil, func() int { ran = true + return 2 }) c.Check(s.api.CalledWith("container.state", "Cancelled"), IsNil) c.Check(s.api.CalledWith("container.state", "Complete"), IsNil) c.Check(ran, Equals, false) } +func ec2MetadataServerStub(c *C, token *string, failureRate float64, stoptime *atomic.Value) *httptest.Server { + failedOnce := false + return httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { + if !failedOnce || rand.Float64() < failureRate { + w.WriteHeader(http.StatusServiceUnavailable) + failedOnce = true + return + } + switch r.URL.Path { + case "/latest/api/token": + fmt.Fprintln(w, *token) + case "/latest/meta-data/spot/instance-action": + if r.Header.Get("X-aws-ec2-metadata-token") != *token { + w.WriteHeader(http.StatusUnauthorized) + } else if t, _ := stoptime.Load().(time.Time); t.IsZero() { + w.WriteHeader(http.StatusNotFound) + } else { + fmt.Fprintf(w, `{"action":"stop","time":"%s"}`, t.Format(time.RFC3339)) + } + default: + w.WriteHeader(http.StatusNotFound) + } + })) +} + +func (s *TestSuite) TestSpotInterruptionNotice(c *C) { + s.testSpotInterruptionNotice(c, 0.1) +} + +func (s *TestSuite) TestSpotInterruptionNoticeNotAvailable(c *C) { + s.testSpotInterruptionNotice(c, 1) +} + +func (s *TestSuite) testSpotInterruptionNotice(c *C, failureRate float64) { + var stoptime atomic.Value + token := "fake-ec2-metadata-token" + stub := ec2MetadataServerStub(c, &token, failureRate, &stoptime) + defer stub.Close() + + defer func(i time.Duration, u string) { + spotInterruptionCheckInterval = i + ec2MetadataBaseURL = u + }(spotInterruptionCheckInterval, ec2MetadataBaseURL) + spotInterruptionCheckInterval = time.Second / 8 + ec2MetadataBaseURL = stub.URL + + go s.runner.checkSpotInterruptionNotices() + s.fullRunHelper(c, `{ + "command": ["sleep", "3"], + "container_image": "`+arvadostest.DockerImage112PDH+`", + "cwd": ".", + "environment": {}, + "mounts": {"/tmp": {"kind": "tmp"} }, + "output_path": "/tmp", + "priority": 1, + "runtime_constraints": {}, + "state": "Locked" +}`, nil, func() int { + time.Sleep(time.Second) + stoptime.Store(time.Now().Add(time.Minute).UTC()) + token = "different-fake-ec2-metadata-token" + time.Sleep(time.Second) + return 0 + }) + c.Check(s.api.Logs["crunch-run"].String(), Matches, `(?ms).*Checking for spot interruptions every 125ms using instance metadata at http://.*`) + c.Check(s.api.Logs["crunch-run"].String(), Matches, `(?ms).*Error checking spot interruptions: 503 Service Unavailable.*`) + if failureRate == 1 { + c.Check(s.api.Logs["crunch-run"].String(), Matches, `(?ms).*Giving up on checking spot interruptions after too many consecutive failures.*`) + } else { + text := `Cloud provider scheduled instance stop at ` + stoptime.Load().(time.Time).Format(time.RFC3339) + c.Check(s.api.Logs["crunch-run"].String(), Matches, `(?ms).*`+text+`.*`) + c.Check(s.api.CalledWith("container.runtime_status.warning", "preemption notice"), NotNil) + c.Check(s.api.CalledWith("container.runtime_status.warningDetail", text), NotNil) + c.Check(s.api.CalledWith("container.runtime_status.preemptionNotice", text), NotNil) + } +} + func (s *TestSuite) TestRunTimeExceeded(c *C) { s.fullRunHelper(c, `{ "command": ["sleep", "3"], @@ -784,8 +951,9 @@ func (s *TestSuite) TestRunTimeExceeded(c *C) { "runtime_constraints": {}, "scheduling_parameters":{"max_run_time": 1}, "state": "Locked" -}`, nil, 0, func() { +}`, nil, func() int { time.Sleep(3 * time.Second) + return 0 }) c.Check(s.api.CalledWith("container.state", "Cancelled"), NotNil) @@ -801,8 +969,9 @@ func (s *TestSuite) TestContainerWaitFails(c *C) { "output_path": "/tmp", "priority": 1, "state": "Locked" -}`, nil, 0, func() { +}`, nil, func() int { s.executor.waitErr = errors.New("Container is not running") + return 0 }) c.Check(s.api.CalledWith("container.state", "Cancelled"), NotNil) @@ -810,6 +979,7 @@ func (s *TestSuite) TestContainerWaitFails(c *C) { } func (s *TestSuite) TestCrunchstat(c *C) { + s.runner.crunchstatFakeFS = os.DirFS("../crunchstat/testdata/debian12") s.fullRunHelper(c, `{ "command": ["sleep", "1"], "container_image": "`+arvadostest.DockerImage112PDH+`", @@ -820,25 +990,19 @@ func (s *TestSuite) TestCrunchstat(c *C) { "priority": 1, "runtime_constraints": {}, "state": "Locked" - }`, nil, 0, func() { + }`, nil, func() int { time.Sleep(time.Second) + return 0 }) c.Check(s.api.CalledWith("container.exit_code", 0), NotNil) c.Check(s.api.CalledWith("container.state", "Complete"), NotNil) - // We didn't actually start a container, so crunchstat didn't - // find accounting files and therefore didn't log any stats. - // It should have logged a "can't find accounting files" - // message after one poll interval, though, so we can confirm - // it's alive: c.Assert(s.api.Logs["crunchstat"], NotNil) - c.Check(s.api.Logs["crunchstat"].String(), Matches, `(?ms).*cgroup stats files have not appeared after 100ms.*`) + c.Check(s.api.Logs["crunchstat"].String(), Matches, `(?ms).*mem \d+ swap \d+ pgmajfault \d+ rss.*`) - // The "files never appeared" log assures us that we called - // (*crunchstat.Reporter)Stop(), and that we set it up with - // the correct container ID "abcde": - c.Check(s.api.Logs["crunchstat"].String(), Matches, `(?ms).*cgroup stats files never appeared for cgroupid\n`) + // Check that we called (*crunchstat.Reporter)Stop(). + c.Check(s.api.Logs["crunch-run"].String(), Matches, `(?ms).*Maximum crunch-run memory rss usage was \d+ bytes\n.*`) } func (s *TestSuite) TestNodeInfoLog(c *C) { @@ -853,10 +1017,10 @@ func (s *TestSuite) TestNodeInfoLog(c *C) { "priority": 1, "runtime_constraints": {}, "state": "Locked" - }`, nil, 0, - func() { - time.Sleep(time.Second) - }) + }`, nil, func() int { + time.Sleep(time.Second) + return 0 + }) c.Check(s.api.CalledWith("container.exit_code", 0), NotNil) c.Check(s.api.CalledWith("container.state", "Complete"), NotNil) @@ -887,9 +1051,9 @@ func (s *TestSuite) TestLogVersionAndRuntime(c *C) { "priority": 1, "runtime_constraints": {}, "state": "Locked" - }`, nil, 0, - func() { - }) + }`, nil, func() int { + return 0 + }) c.Assert(s.api.Logs["crunch-run"], NotNil) c.Check(s.api.Logs["crunch-run"].String(), Matches, `(?ms).*crunch-run \S+ \(go\S+\) start.*`) @@ -898,6 +1062,70 @@ func (s *TestSuite) TestLogVersionAndRuntime(c *C) { c.Check(s.api.Logs["crunch-run"].String(), Matches, `(?ms).*Using container runtime: stub.*`) } +func (s *TestSuite) testLogRSSThresholds(c *C, ram int64, expected []int, notExpected int) { + s.runner.crunchstatFakeFS = os.DirFS("../crunchstat/testdata/debian12") + s.fullRunHelper(c, `{ + "command": ["true"], + "container_image": "`+arvadostest.DockerImage112PDH+`", + "cwd": ".", + "environment": {}, + "mounts": {"/tmp": {"kind": "tmp"} }, + "output_path": "/tmp", + "priority": 1, + "runtime_constraints": {"ram": `+fmt.Sprintf("%d", ram)+`}, + "state": "Locked" + }`, nil, func() int { return 0 }) + c.Logf("=== crunchstat logs\n%s\n", s.api.Logs["crunchstat"].String()) + logs := s.api.Logs["crunch-run"].String() + pattern := logLineStart + `Container using over %d%% of memory \(rss %d/%d bytes\)` + var threshold int + for _, threshold = range expected { + c.Check(logs, Matches, fmt.Sprintf(pattern, threshold, s.debian12MemoryCurrent, ram)) + } + if notExpected > threshold { + c.Check(logs, Not(Matches), fmt.Sprintf(pattern, notExpected, s.debian12MemoryCurrent, ram)) + } +} + +func (s *TestSuite) TestLogNoRSSThresholds(c *C) { + s.testLogRSSThresholds(c, s.debian12MemoryCurrent*10, []int{}, 90) +} + +func (s *TestSuite) TestLogSomeRSSThresholds(c *C) { + onePercentRSS := s.debian12MemoryCurrent / 100 + s.testLogRSSThresholds(c, 102*onePercentRSS, []int{90, 95}, 99) +} + +func (s *TestSuite) TestLogAllRSSThresholds(c *C) { + s.testLogRSSThresholds(c, s.debian12MemoryCurrent, []int{90, 95, 99}, 0) +} + +func (s *TestSuite) TestLogMaximaAfterRun(c *C) { + s.runner.crunchstatFakeFS = os.DirFS("../crunchstat/testdata/debian12") + s.runner.parentTemp = c.MkDir() + s.fullRunHelper(c, `{ + "command": ["true"], + "container_image": "`+arvadostest.DockerImage112PDH+`", + "cwd": ".", + "environment": {}, + "mounts": {"/tmp": {"kind": "tmp"} }, + "output_path": "/tmp", + "priority": 1, + "runtime_constraints": {"ram": `+fmt.Sprintf("%d", s.debian12MemoryCurrent*10)+`}, + "state": "Locked" + }`, nil, func() int { return 0 }) + logs := s.api.Logs["crunch-run"].String() + for _, expected := range []string{ + `Maximum disk usage was \d+%, \d+/\d+ bytes`, + fmt.Sprintf(`Maximum container memory swap usage was %d bytes`, s.debian12SwapCurrent), + `Maximum container memory pgmajfault usage was \d+ faults`, + fmt.Sprintf(`Maximum container memory rss usage was 10%%, %d/%d bytes`, s.debian12MemoryCurrent, s.debian12MemoryCurrent*10), + `Maximum crunch-run memory rss usage was \d+ bytes`, + } { + c.Check(logs, Matches, logLineStart+expected) + } +} + func (s *TestSuite) TestCommitNodeInfoBeforeStart(c *C) { var collection_create, container_update arvadosclient.Dict s.fullRunHelper(c, `{ @@ -911,11 +1139,11 @@ func (s *TestSuite) TestCommitNodeInfoBeforeStart(c *C) { "runtime_constraints": {}, "state": "Locked", "uuid": "zzzzz-dz642-202301121543210" - }`, nil, 0, - func() { - collection_create = s.api.CalledWith("ensure_unique_name", true) - container_update = s.api.CalledWith("container.state", "Running") - }) + }`, nil, func() int { + collection_create = s.api.CalledWith("ensure_unique_name", true) + container_update = s.api.CalledWith("container.state", "Running") + return 0 + }) c.Assert(collection_create, NotNil) log_collection := collection_create["collection"].(arvadosclient.Dict) @@ -945,9 +1173,10 @@ func (s *TestSuite) TestContainerRecordLog(c *C) { "priority": 1, "runtime_constraints": {}, "state": "Locked" - }`, nil, 0, - func() { + }`, nil, + func() int { time.Sleep(time.Second) + return 0 }) c.Check(s.api.CalledWith("container.exit_code", 0), NotNil) @@ -968,9 +1197,10 @@ func (s *TestSuite) TestFullRunStderr(c *C) { "priority": 1, "runtime_constraints": {}, "state": "Locked" -}`, nil, 1, func() { +}`, nil, func() int { fmt.Fprintln(s.executor.created.Stdout, "hello") fmt.Fprintln(s.executor.created.Stderr, "world") + return 1 }) final := s.api.CalledWith("container.state", "Complete") @@ -993,8 +1223,9 @@ func (s *TestSuite) TestFullRunDefaultCwd(c *C) { "priority": 1, "runtime_constraints": {}, "state": "Locked" -}`, nil, 0, func() { +}`, nil, func() int { fmt.Fprintf(s.executor.created.Stdout, "workdir=%q", s.executor.created.WorkingDir) + return 0 }) c.Check(s.api.CalledWith("container.exit_code", 0), NotNil) @@ -1014,8 +1245,9 @@ func (s *TestSuite) TestFullRunSetCwd(c *C) { "priority": 1, "runtime_constraints": {}, "state": "Locked" -}`, nil, 0, func() { +}`, nil, func() int { fmt.Fprintln(s.executor.created.Stdout, s.executor.created.WorkingDir) + return 0 }) c.Check(s.api.CalledWith("container.exit_code", 0), NotNil) @@ -1035,8 +1267,9 @@ func (s *TestSuite) TestFullRunSetOutputStorageClasses(c *C) { "runtime_constraints": {}, "state": "Locked", "output_storage_classes": ["foo", "bar"] -}`, nil, 0, func() { +}`, nil, func() int { fmt.Fprintln(s.executor.created.Stdout, s.executor.created.WorkingDir) + return 0 }) c.Check(s.api.CalledWith("container.exit_code", 0), NotNil) @@ -1058,8 +1291,9 @@ func (s *TestSuite) TestEnableCUDADeviceCount(c *C) { "runtime_constraints": {"cuda": {"device_count": 2}}, "state": "Locked", "output_storage_classes": ["foo", "bar"] -}`, nil, 0, func() { +}`, nil, func() int { fmt.Fprintln(s.executor.created.Stdout, "ok") + return 0 }) c.Check(s.executor.created.CUDADeviceCount, Equals, 2) } @@ -1076,25 +1310,30 @@ func (s *TestSuite) TestEnableCUDAHardwareCapability(c *C) { "runtime_constraints": {"cuda": {"hardware_capability": "foo"}}, "state": "Locked", "output_storage_classes": ["foo", "bar"] -}`, nil, 0, func() { +}`, nil, func() int { fmt.Fprintln(s.executor.created.Stdout, "ok") + return 0 }) c.Check(s.executor.created.CUDADeviceCount, Equals, 0) } func (s *TestSuite) TestStopOnSignal(c *C) { - s.executor.runFunc = func() { + s.executor.runFunc = func() int { s.executor.created.Stdout.Write([]byte("foo\n")) s.runner.SigChan <- syscall.SIGINT + time.Sleep(10 * time.Second) + return 0 } s.testStopContainer(c) } func (s *TestSuite) TestStopOnArvMountDeath(c *C) { - s.executor.runFunc = func() { + s.executor.runFunc = func() int { s.executor.created.Stdout.Write([]byte("foo\n")) s.runner.ArvMountExit <- nil close(s.runner.ArvMountExit) + time.Sleep(10 * time.Second) + return 0 } s.runner.ArvMountExit = make(chan error) s.testStopContainer(c) @@ -1153,8 +1392,9 @@ func (s *TestSuite) TestFullRunSetEnv(c *C) { "priority": 1, "runtime_constraints": {}, "state": "Locked" -}`, nil, 0, func() { +}`, nil, func() int { fmt.Fprintf(s.executor.created.Stdout, "%v", s.executor.created.Env) + return 0 }) c.Check(s.api.CalledWith("container.exit_code", 0), NotNil) @@ -1173,11 +1413,11 @@ func (am *ArvMountCmdLine) ArvMountTest(c []string, token string) (*exec.Cmd, er return nil, nil } -func stubCert(temp string) string { +func stubCert(c *C, temp string) string { path := temp + "/ca-certificates.crt" - crt, _ := os.Create(path) - crt.Close() - arvadosclient.CertFiles = []string{path} + err := os.WriteFile(path, []byte{}, 0666) + c.Assert(err, IsNil) + os.Setenv("SSL_CERT_FILE", path) return path } @@ -1185,13 +1425,14 @@ func (s *TestSuite) TestSetupMounts(c *C) { cr := s.runner am := &ArvMountCmdLine{} cr.RunArvMount = am.ArvMountTest + cr.containerClient, _ = apiStub() cr.ContainerArvClient = &ArvTestClient{} cr.ContainerKeepClient = &KeepTestClient{} cr.Container.OutputStorageClasses = []string{"default"} realTemp := c.MkDir() certTemp := c.MkDir() - stubCertPath := stubCert(certTemp) + stubCertPath := stubCert(c, certTemp) cr.parentTemp = realTemp i := 0 @@ -1507,7 +1748,7 @@ func (s *TestSuite) TestSetupMounts(c *C) { { i = 0 cr.ArvMountPoint = "" - (*GitMountSuite)(nil).useTestGitServer(c) + git_client.InstallProtocol("https", git_http.NewClient(arvados.InsecureHTTPClient)) cr.token = arvadostest.ActiveToken cr.Container.Mounts = make(map[string]arvados.Mount) cr.Container.Mounts = map[string]arvados.Mount{ @@ -1565,8 +1806,9 @@ func (s *TestSuite) TestStdout(c *C) { "state": "Locked" }` - s.fullRunHelper(c, helperRecord, nil, 0, func() { + s.fullRunHelper(c, helperRecord, nil, func() int { fmt.Fprintln(s.executor.created.Stdout, s.executor.created.Env["FROBIZ"]) + return 0 }) c.Check(s.api.CalledWith("container.exit_code", 0), NotNil) @@ -1575,7 +1817,7 @@ func (s *TestSuite) TestStdout(c *C) { } // Used by the TestStdoutWithWrongPath*() -func (s *TestSuite) stdoutErrorRunHelper(c *C, record string, fn func()) (*ArvTestClient, *ContainerRunner, error) { +func (s *TestSuite) stdoutErrorRunHelper(c *C, record string, fn func() int) (*ArvTestClient, *ContainerRunner, error) { err := json.Unmarshal([]byte(record), &s.api.Container) c.Assert(err, IsNil) s.executor.runFunc = fn @@ -1591,7 +1833,7 @@ func (s *TestSuite) TestStdoutWithWrongPath(c *C) { "mounts": {"/tmp": {"kind": "tmp"}, "stdout": {"kind": "file", "path":"/tmpa.out"} }, "output_path": "/tmp", "state": "Locked" -}`, func() {}) +}`, func() int { return 0 }) c.Check(err, ErrorMatches, ".*Stdout path does not start with OutputPath.*") } @@ -1600,7 +1842,7 @@ func (s *TestSuite) TestStdoutWithWrongKindTmp(c *C) { "mounts": {"/tmp": {"kind": "tmp"}, "stdout": {"kind": "tmp", "path":"/tmp/a.out"} }, "output_path": "/tmp", "state": "Locked" -}`, func() {}) +}`, func() int { return 0 }) c.Check(err, ErrorMatches, ".*unsupported mount kind 'tmp' for stdout.*") } @@ -1609,7 +1851,7 @@ func (s *TestSuite) TestStdoutWithWrongKindCollection(c *C) { "mounts": {"/tmp": {"kind": "tmp"}, "stdout": {"kind": "collection", "path":"/tmp/a.out"} }, "output_path": "/tmp", "state": "Locked" -}`, func() {}) +}`, func() int { return 0 }) c.Check(err, ErrorMatches, ".*unsupported mount kind 'collection' for stdout.*") } @@ -1624,9 +1866,9 @@ func (s *TestSuite) TestFullRunWithAPI(c *C) { "priority": 1, "runtime_constraints": {"API": true}, "state": "Locked" -}`, nil, 0, func() { +}`, nil, func() int { c.Check(s.executor.created.Env["ARVADOS_API_HOST"], Equals, os.Getenv("ARVADOS_API_HOST")) - s.executor.exit <- 3 + return 3 }) c.Check(s.api.CalledWith("container.exit_code", 3), NotNil) c.Check(s.api.CalledWith("container.state", "Complete"), NotNil) @@ -1646,8 +1888,9 @@ func (s *TestSuite) TestFullRunSetOutput(c *C) { "priority": 1, "runtime_constraints": {"API": true}, "state": "Locked" -}`, nil, 0, func() { +}`, nil, func() int { s.api.Container.Output = arvadostest.DockerImage112PDH + return 0 }) c.Check(s.api.CalledWith("container.exit_code", 0), NotNil) @@ -1661,9 +1904,9 @@ func (s *TestSuite) TestArvMountRuntimeStatusWarning(c *C) { ioutil.WriteFile(s.runner.ArvMountPoint+"/by_id/README", nil, 0666) return s.runner.ArvMountCmd([]string{"bash", "-c", "echo >&2 Test: Keep write error: I am a teapot; sleep 3"}, "") } - s.executor.runFunc = func() { + s.executor.runFunc = func() int { time.Sleep(time.Second) - s.executor.exit <- 137 + return 137 } record := `{ "command": ["sleep", "1"], @@ -1709,8 +1952,9 @@ func (s *TestSuite) TestStdoutWithExcludeFromOutputMountPointUnderOutputDir(c *C extraMounts := []string{"a3e8f74c6f101eae01fa08bfb4e49b3a+54"} - s.fullRunHelper(c, helperRecord, extraMounts, 0, func() { + s.fullRunHelper(c, helperRecord, extraMounts, func() int { fmt.Fprintln(s.executor.created.Stdout, s.executor.created.Env["FROBIZ"]) + return 0 }) c.Check(s.api.CalledWith("container.exit_code", 0), NotNil) @@ -1745,8 +1989,9 @@ func (s *TestSuite) TestStdoutWithMultipleMountPointsUnderOutputDir(c *C) { "a0def87f80dd594d4675809e83bd4f15+367/subdir1/subdir2/file2_in_subdir2.txt", } - api, _, realtemp := s.fullRunHelper(c, helperRecord, extraMounts, 0, func() { + api, _, realtemp := s.fullRunHelper(c, helperRecord, extraMounts, func() int { fmt.Fprintln(s.executor.created.Stdout, s.executor.created.Env["FROBIZ"]) + return 0 }) c.Check(s.executor.created.BindMounts, DeepEquals, map[string]bindmount{ @@ -1802,8 +2047,9 @@ func (s *TestSuite) TestStdoutWithMountPointsUnderOutputDirDenormalizedManifest( "b0def87f80dd594d4675809e83bd4f15+367/subdir1/file2_in_subdir1.txt", } - s.fullRunHelper(c, helperRecord, extraMounts, 0, func() { + s.fullRunHelper(c, helperRecord, extraMounts, func() int { fmt.Fprintln(s.executor.created.Stdout, s.executor.created.Env["FROBIZ"]) + return 0 }) c.Check(s.api.CalledWith("container.exit_code", 0), NotNil) @@ -1839,8 +2085,9 @@ func (s *TestSuite) TestOutputError(c *C) { "runtime_constraints": {}, "state": "Locked" }` - s.fullRunHelper(c, helperRecord, nil, 0, func() { + s.fullRunHelper(c, helperRecord, nil, func() int { os.Symlink("/etc/hosts", s.runner.HostOutputDir+"/baz") + return 0 }) c.Check(s.api.CalledWith("container.state", "Cancelled"), NotNil) @@ -1867,8 +2114,9 @@ func (s *TestSuite) TestStdinCollectionMountPoint(c *C) { "b0def87f80dd594d4675809e83bd4f15+367/file1_in_main.txt", } - api, _, _ := s.fullRunHelper(c, helperRecord, extraMounts, 0, func() { + api, _, _ := s.fullRunHelper(c, helperRecord, extraMounts, func() int { fmt.Fprintln(s.executor.created.Stdout, s.executor.created.Env["FROBIZ"]) + return 0 }) c.Check(api.CalledWith("container.exit_code", 0), NotNil) @@ -1902,8 +2150,9 @@ func (s *TestSuite) TestStdinJsonMountPoint(c *C) { "state": "Locked" }` - api, _, _ := s.fullRunHelper(c, helperRecord, nil, 0, func() { + api, _, _ := s.fullRunHelper(c, helperRecord, nil, func() int { fmt.Fprintln(s.executor.created.Stdout, s.executor.created.Env["FROBIZ"]) + return 0 }) c.Check(api.CalledWith("container.exit_code", 0), NotNil) @@ -1933,9 +2182,10 @@ func (s *TestSuite) TestStderrMount(c *C) { "priority": 1, "runtime_constraints": {}, "state": "Locked" -}`, nil, 1, func() { +}`, nil, func() int { fmt.Fprintln(s.executor.created.Stdout, "hello") fmt.Fprintln(s.executor.created.Stderr, "oops") + return 1 }) final := api.CalledWith("container.state", "Complete") @@ -1985,7 +2235,7 @@ func (s *TestSuite) TestFullBrokenDocker(c *C) { "priority": 1, "runtime_constraints": {}, "state": "Locked" -}`, nil, 0, func() {}) +}`, nil, func() int { return 0 }) c.Check(s.api.CalledWith("container.state", nextState), NotNil) c.Check(s.api.Logs["crunch-run"].String(), Matches, "(?ms).*unable to run containers.*") if s.runner.brokenNodeHook != "" { @@ -2016,7 +2266,7 @@ func (s *TestSuite) TestBadCommand(c *C) { "priority": 1, "runtime_constraints": {}, "state": "Locked" -}`, nil, 0, func() {}) +}`, nil, func() int { return 0 }) c.Check(s.api.CalledWith("container.state", "Cancelled"), NotNil) c.Check(s.api.Logs["crunch-run"].String(), Matches, "(?ms).*Possible causes:.*is missing.*") } @@ -2039,10 +2289,11 @@ func (s *TestSuite) TestSecretTextMountPoint(c *C) { "state": "Locked" }` - s.fullRunHelper(c, helperRecord, nil, 0, func() { + s.fullRunHelper(c, helperRecord, nil, func() int { content, err := ioutil.ReadFile(s.runner.HostOutputDir + "/secret.conf") c.Check(err, IsNil) c.Check(string(content), Equals, "mypassword") + return 0 }) c.Check(s.api.CalledWith("container.exit_code", 0), NotNil) @@ -2068,10 +2319,11 @@ func (s *TestSuite) TestSecretTextMountPoint(c *C) { }` s.SetUpTest(c) - s.fullRunHelper(c, helperRecord, nil, 0, func() { + s.fullRunHelper(c, helperRecord, nil, func() int { content, err := ioutil.ReadFile(s.runner.HostOutputDir + "/secret.conf") c.Check(err, IsNil) c.Check(string(content), Equals, "mypassword") + return 0 }) c.Check(s.api.CalledWith("container.exit_code", 0), NotNil) @@ -2097,7 +2349,7 @@ func (s *TestSuite) TestSecretTextMountPoint(c *C) { }` s.SetUpTest(c) - _, _, realtemp := s.fullRunHelper(c, helperRecord, nil, 0, func() { + _, _, realtemp := s.fullRunHelper(c, helperRecord, nil, func() int { // secret.conf should be provisioned as a separate // bind mount, i.e., it should not appear in the // (fake) fuse filesystem as viewed from the host. @@ -2107,6 +2359,7 @@ func (s *TestSuite) TestSecretTextMountPoint(c *C) { } err = ioutil.WriteFile(s.runner.HostOutputDir+"/.arvados#collection", []byte(`{"manifest_text":". acbd18db4cc2f85cedef654fccc4a4d8+3 0:3:foo.txt\n"}`), 0700) c.Check(err, IsNil) + return 0 }) content, err := ioutil.ReadFile(realtemp + "/text1/mountdata.text") @@ -2172,6 +2425,80 @@ func (s *TestSuite) TestCalculateCost(c *C) { c.Check(logbuf.String(), Not(Matches), `(?ms).*changed to 2\.00 .* changed to 2\.00 .*`) } +func (s *TestSuite) TestSIGUSR2CostUpdate(c *C) { + pid := os.Getpid() + now := time.Now() + pricesJSON, err := json.Marshal([]cloud.InstancePrice{ + {StartTime: now.Add(-4 * time.Hour), Price: 2.4}, + {StartTime: now.Add(-2 * time.Hour), Price: 2.6}, + }) + c.Assert(err, IsNil) + + os.Setenv("InstanceType", `{"Price":2.2}`) + defer os.Unsetenv("InstanceType") + defer func(s string) { lockdir = s }(lockdir) + lockdir = c.MkDir() + + // We can't use s.api.CalledWith because timing differences will yield + // different cost values across runs. getCostUpdate iterates over API + // calls until it finds one that sets the cost, then writes that value + // to the next index of costUpdates. + deadline := now.Add(time.Second) + costUpdates := make([]float64, 2) + costIndex := 0 + apiIndex := 0 + getCostUpdate := func() { + for ; time.Now().Before(deadline); time.Sleep(time.Second / 10) { + for apiIndex < len(s.api.Content) { + update := s.api.Content[apiIndex] + apiIndex++ + var ok bool + var cost float64 + if update, ok = update["container"].(arvadosclient.Dict); !ok { + continue + } + if cost, ok = update["cost"].(float64); !ok { + continue + } + c.Logf("API call #%d updates cost to %v", apiIndex-1, cost) + costUpdates[costIndex] = cost + costIndex++ + return + } + } + } + + s.fullRunHelper(c, `{ + "command": ["true"], + "container_image": "`+arvadostest.DockerImage112PDH+`", + "cwd": ".", + "environment": {}, + "mounts": {"/tmp": {"kind": "tmp"} }, + "output_path": "/tmp", + "priority": 1, + "runtime_constraints": {}, + "state": "Locked", + "uuid": "zzzzz-dz642-20230320101530a" + }`, nil, func() int { + s.runner.costStartTime = now.Add(-3 * time.Hour) + err := syscall.Kill(pid, syscall.SIGUSR2) + c.Check(err, IsNil, Commentf("error sending first SIGUSR2 to runner")) + getCostUpdate() + + err = os.WriteFile(path.Join(lockdir, pricesfile), pricesJSON, 0o700) + c.Check(err, IsNil, Commentf("error writing JSON prices file")) + err = syscall.Kill(pid, syscall.SIGUSR2) + c.Check(err, IsNil, Commentf("error sending second SIGUSR2 to runner")) + getCostUpdate() + + return 0 + }) + // Comparing with format strings makes it easy to ignore minor variations + // in cost across runs while keeping diagnostics pretty. + c.Check(fmt.Sprintf("%.3f", costUpdates[0]), Equals, "6.600") + c.Check(fmt.Sprintf("%.3f", costUpdates[1]), Equals, "7.600") +} + type FakeProcess struct { cmdLine []string }