lib/diagnostics/cmd.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: AGPL-3.0
   4
   5 package diagnostics
   6
   7 import (
   8         "bytes"
   9         "context"
  10         "flag"
  11         "fmt"
  12         "io"
  13         "io/ioutil"
  14         "net"
  15         "net/http"
  16         "net/url"
  17         "strings"
  18         "time"
  19
  20         "git.arvados.org/arvados.git/lib/cmd"
  21         "git.arvados.org/arvados.git/sdk/go/arvados"
  22         "git.arvados.org/arvados.git/sdk/go/ctxlog"
  23         "github.com/sirupsen/logrus"
  24 )
  25
  26 type Command struct{}
  27
  28 func (Command) RunCommand(prog string, args []string, stdin io.Reader, stdout, stderr io.Writer) int {
  29         var diag diagnoser
  30         f := flag.NewFlagSet(prog, flag.ContinueOnError)
  31         f.StringVar(&diag.projectName, "project-name", "scratch area for diagnostics", "name of project to find/create in home project and use for temporary/test objects")
  32         f.StringVar(&diag.logLevel, "log-level", "info", "logging level (debug, info, warning, error)")
  33         f.StringVar(&diag.dockerImage, "docker-image", "alpine:latest", "image to use when running a test container")
  34         f.BoolVar(&diag.checkInternal, "internal-client", false, "check that this host is considered an \"internal\" client")
  35         f.BoolVar(&diag.checkExternal, "external-client", false, "check that this host is considered an \"external\" client")
  36         f.IntVar(&diag.priority, "priority", 500, "priority for test container (1..1000, or 0 to skip)")
  37         f.DurationVar(&diag.timeout, "timeout", 10*time.Second, "timeout for http requests")
  38         if ok, code := cmd.ParseFlags(f, prog, args, "", stderr); !ok {
  39                 return code
  40         }
  41         diag.logger = ctxlog.New(stdout, "text", diag.logLevel)
  42         diag.logger.SetFormatter(&logrus.TextFormatter{DisableTimestamp: true, DisableLevelTruncation: true, PadLevelText: true})
  43         diag.runtests()
  44         if len(diag.errors) == 0 {
  45                 diag.logger.Info("--- no errors ---")
  46                 return 0
  47         } else {
  48                 if diag.logger.Level > logrus.ErrorLevel {
  49                         fmt.Fprint(stdout, "\n--- cut here --- error summary ---\n\n")
  50                         for _, e := range diag.errors {
  51                                 diag.logger.Error(e)
  52                         }
  53                 }
  54                 return 1
  55         }
  56 }
  57
  58 type diagnoser struct {
  59         stdout        io.Writer
  60         stderr        io.Writer
  61         logLevel      string
  62         priority      int
  63         projectName   string
  64         dockerImage   string
  65         checkInternal bool
  66         checkExternal bool
  67         timeout       time.Duration
  68         logger        *logrus.Logger
  69         errors        []string
  70         done          map[int]bool
  71 }
  72
  73 func (diag *diagnoser) debugf(f string, args ...interface{}) {
  74         diag.logger.Debugf("  ... "+f, args...)
  75 }
  76
  77 func (diag *diagnoser) infof(f string, args ...interface{}) {
  78         diag.logger.Infof("  ... "+f, args...)
  79 }
  80
  81 func (diag *diagnoser) warnf(f string, args ...interface{}) {
  82         diag.logger.Warnf("  ... "+f, args...)
  83 }
  84
  85 func (diag *diagnoser) errorf(f string, args ...interface{}) {
  86         diag.logger.Errorf(f, args...)
  87         diag.errors = append(diag.errors, fmt.Sprintf(f, args...))
  88 }
  89
  90 // Run the given func, logging appropriate messages before and after,
  91 // adding timing info, etc.
  92 //
  93 // The id argument should be unique among tests, and shouldn't change
  94 // when other tests are added/removed.
  95 func (diag *diagnoser) dotest(id int, title string, fn func() error) {
  96         if diag.done == nil {
  97                 diag.done = map[int]bool{}
  98         } else if diag.done[id] {
  99                 diag.errorf("(bug) reused test id %d", id)
 100         }
 101         diag.done[id] = true
 102
 103         diag.logger.Infof("%4d: %s", id, title)
 104         t0 := time.Now()
 105         err := fn()
 106         elapsed := fmt.Sprintf("%d ms", time.Now().Sub(t0)/time.Millisecond)
 107         if err != nil {
 108                 diag.errorf("%4d: %s (%s): %s", id, title, elapsed, err)
 109         } else {
 110                 diag.logger.Debugf("%4d: %s (%s): ok", id, title, elapsed)
 111         }
 112 }
 113
 114 func (diag *diagnoser) runtests() {
 115         client := arvados.NewClientFromEnv()
 116
 117         if client.APIHost == "" || client.AuthToken == "" {
 118                 diag.errorf("ARVADOS_API_HOST and ARVADOS_API_TOKEN environment variables are not set -- aborting without running any tests")
 119                 return
 120         }
 121
 122         var dd arvados.DiscoveryDocument
 123         ddpath := "discovery/v1/apis/arvados/v1/rest"
 124         diag.dotest(10, fmt.Sprintf("getting discovery document from https://%s/%s", client.APIHost, ddpath), func() error {
 125                 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 126                 defer cancel()
 127                 err := client.RequestAndDecodeContext(ctx, &dd, "GET", ddpath, nil, nil)
 128                 if err != nil {
 129                         return err
 130                 }
 131                 diag.debugf("BlobSignatureTTL = %d", dd.BlobSignatureTTL)
 132                 return nil
 133         })
 134
 135         var cluster arvados.Cluster
 136         cfgpath := "arvados/v1/config"
 137         diag.dotest(20, fmt.Sprintf("getting exported config from https://%s/%s", client.APIHost, cfgpath), func() error {
 138                 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 139                 defer cancel()
 140                 err := client.RequestAndDecodeContext(ctx, &cluster, "GET", cfgpath, nil, nil)
 141                 if err != nil {
 142                         return err
 143                 }
 144                 diag.debugf("Collections.BlobSigning = %v", cluster.Collections.BlobSigning)
 145                 return nil
 146         })
 147
 148         var user arvados.User
 149         diag.dotest(30, "getting current user record", func() error {
 150                 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 151                 defer cancel()
 152                 err := client.RequestAndDecodeContext(ctx, &user, "GET", "arvados/v1/users/current", nil, nil)
 153                 if err != nil {
 154                         return err
 155                 }
 156                 diag.debugf("user uuid = %s", user.UUID)
 157                 return nil
 158         })
 159
 160         // uncomment to create some spurious errors
 161         // cluster.Services.WebDAVDownload.ExternalURL.Host = "0.0.0.0:9"
 162
 163         // TODO: detect routing errors here, like finding wb2 at the
 164         // wb1 address.
 165         for i, svc := range []*arvados.Service{
 166                 &cluster.Services.Keepproxy,
 167                 &cluster.Services.WebDAV,
 168                 &cluster.Services.WebDAVDownload,
 169                 &cluster.Services.Websocket,
 170                 &cluster.Services.Workbench1,
 171                 &cluster.Services.Workbench2,
 172         } {
 173                 diag.dotest(40+i, fmt.Sprintf("connecting to service endpoint %s", svc.ExternalURL), func() error {
 174                         ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 175                         defer cancel()
 176                         u := svc.ExternalURL
 177                         if strings.HasPrefix(u.Scheme, "ws") {
 178                                 // We can do a real websocket test elsewhere,
 179                                 // but for now we'll just check the https
 180                                 // connection.
 181                                 u.Scheme = "http" + u.Scheme[2:]
 182                         }
 183                         if svc == &cluster.Services.WebDAV && strings.HasPrefix(u.Host, "*") {
 184                                 u.Host = "d41d8cd98f00b204e9800998ecf8427e-0" + u.Host[1:]
 185                         }
 186                         req, err := http.NewRequestWithContext(ctx, http.MethodGet, u.String(), nil)
 187                         if err != nil {
 188                                 return err
 189                         }
 190                         resp, err := http.DefaultClient.Do(req)
 191                         if err != nil {
 192                                 return err
 193                         }
 194                         resp.Body.Close()
 195                         return nil
 196                 })
 197         }
 198
 199         for i, url := range []string{
 200                 cluster.Services.Controller.ExternalURL.String(),
 201                 cluster.Services.Keepproxy.ExternalURL.String() + "d41d8cd98f00b204e9800998ecf8427e+0",
 202                 cluster.Services.WebDAVDownload.ExternalURL.String(),
 203         } {
 204                 diag.dotest(50+i, fmt.Sprintf("checking CORS headers at %s", url), func() error {
 205                         ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 206                         defer cancel()
 207                         req, err := http.NewRequestWithContext(ctx, "GET", url, nil)
 208                         if err != nil {
 209                                 return err
 210                         }
 211                         req.Header.Set("Origin", "https://example.com")
 212                         resp, err := http.DefaultClient.Do(req)
 213                         if err != nil {
 214                                 return err
 215                         }
 216                         if hdr := resp.Header.Get("Access-Control-Allow-Origin"); hdr != "*" {
 217                                 return fmt.Errorf("expected \"Access-Control-Allow-Origin: *\", got %q", hdr)
 218                         }
 219                         return nil
 220                 })
 221         }
 222
 223         var keeplist arvados.KeepServiceList
 224         diag.dotest(60, "checking internal/external client detection", func() error {
 225                 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 226                 defer cancel()
 227                 err := client.RequestAndDecodeContext(ctx, &keeplist, "GET", "arvados/v1/keep_services/accessible", nil, arvados.ListOptions{Limit: 999999})
 228                 if err != nil {
 229                         return fmt.Errorf("error getting keep services list: %s", err)
 230                 } else if len(keeplist.Items) == 0 {
 231                         return fmt.Errorf("controller did not return any keep services")
 232                 }
 233                 found := map[string]int{}
 234                 for _, ks := range keeplist.Items {
 235                         found[ks.ServiceType]++
 236                 }
 237                 isInternal := found["proxy"] == 0 && len(keeplist.Items) > 0
 238                 isExternal := found["proxy"] > 0 && found["proxy"] == len(keeplist.Items)
 239                 if isExternal {
 240                         diag.debugf("controller returned only proxy services, this host is treated as \"external\"")
 241                 } else if isInternal {
 242                         diag.debugf("controller returned only non-proxy services, this host is treated as \"internal\"")
 243                 }
 244                 if (diag.checkInternal && !isInternal) || (diag.checkExternal && !isExternal) {
 245                         return fmt.Errorf("expecting internal=%v external=%v, but found internal=%v external=%v", diag.checkInternal, diag.checkExternal, isInternal, isExternal)
 246                 }
 247                 return nil
 248         })
 249
 250         for i, ks := range keeplist.Items {
 251                 u := url.URL{
 252                         Scheme: "http",
 253                         Host:   net.JoinHostPort(ks.ServiceHost, fmt.Sprintf("%d", ks.ServicePort)),
 254                         Path:   "/",
 255                 }
 256                 if ks.ServiceSSLFlag {
 257                         u.Scheme = "https"
 258                 }
 259                 diag.dotest(61+i, fmt.Sprintf("reading+writing via keep service at %s", u.String()), func() error {
 260                         ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 261                         defer cancel()
 262                         req, err := http.NewRequestWithContext(ctx, "PUT", u.String()+"d41d8cd98f00b204e9800998ecf8427e", nil)
 263                         if err != nil {
 264                                 return err
 265                         }
 266                         req.Header.Set("Authorization", "Bearer "+client.AuthToken)
 267                         resp, err := http.DefaultClient.Do(req)
 268                         if err != nil {
 269                                 return err
 270                         }
 271                         defer resp.Body.Close()
 272                         body, err := ioutil.ReadAll(resp.Body)
 273                         if err != nil {
 274                                 return fmt.Errorf("reading response body: %s", err)
 275                         }
 276                         loc := strings.TrimSpace(string(body))
 277                         if !strings.HasPrefix(loc, "d41d8") {
 278                                 return fmt.Errorf("unexpected response from write: %q", body)
 279                         }
 280
 281                         req, err = http.NewRequestWithContext(ctx, "GET", u.String()+loc, nil)
 282                         if err != nil {
 283                                 return err
 284                         }
 285                         req.Header.Set("Authorization", "Bearer "+client.AuthToken)
 286                         resp, err = http.DefaultClient.Do(req)
 287                         if err != nil {
 288                                 return err
 289                         }
 290                         defer resp.Body.Close()
 291                         body, err = ioutil.ReadAll(resp.Body)
 292                         if err != nil {
 293                                 return fmt.Errorf("reading response body: %s", err)
 294                         }
 295                         if len(body) != 0 {
 296                                 return fmt.Errorf("unexpected response from read: %q", body)
 297                         }
 298
 299                         return nil
 300                 })
 301         }
 302
 303         var project arvados.Group
 304         diag.dotest(80, fmt.Sprintf("finding/creating %q project", diag.projectName), func() error {
 305                 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 306                 defer cancel()
 307                 var grplist arvados.GroupList
 308                 err := client.RequestAndDecodeContext(ctx, &grplist, "GET", "arvados/v1/groups", nil, arvados.ListOptions{
 309                         Filters: []arvados.Filter{
 310                                 {"name", "=", diag.projectName},
 311                                 {"group_class", "=", "project"},
 312                                 {"owner_uuid", "=", user.UUID}},
 313                         Limit: 999999})
 314                 if err != nil {
 315                         return fmt.Errorf("list groups: %s", err)
 316                 }
 317                 if len(grplist.Items) > 0 {
 318                         project = grplist.Items[0]
 319                         diag.debugf("using existing project, uuid = %s", project.UUID)
 320                         return nil
 321                 }
 322                 diag.debugf("list groups: ok, no results")
 323                 err = client.RequestAndDecodeContext(ctx, &project, "POST", "arvados/v1/groups", nil, map[string]interface{}{"group": map[string]interface{}{
 324                         "name":        diag.projectName,
 325                         "group_class": "project",
 326                 }})
 327                 if err != nil {
 328                         return fmt.Errorf("create project: %s", err)
 329                 }
 330                 diag.debugf("created project, uuid = %s", project.UUID)
 331                 return nil
 332         })
 333
 334         var collection arvados.Collection
 335         diag.dotest(90, "creating temporary collection", func() error {
 336                 if project.UUID == "" {
 337                         return fmt.Errorf("skipping, no project to work in")
 338                 }
 339                 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 340                 defer cancel()
 341                 err := client.RequestAndDecodeContext(ctx, &collection, "POST", "arvados/v1/collections", nil, map[string]interface{}{
 342                         "ensure_unique_name": true,
 343                         "collection": map[string]interface{}{
 344                                 "owner_uuid": project.UUID,
 345                                 "name":       "test collection",
 346                                 "trash_at":   time.Now().Add(time.Hour)}})
 347                 if err != nil {
 348                         return err
 349                 }
 350                 diag.debugf("ok, uuid = %s", collection.UUID)
 351                 return nil
 352         })
 353
 354         if collection.UUID != "" {
 355                 defer func() {
 356                         diag.dotest(9990, "deleting temporary collection", func() error {
 357                                 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 358                                 defer cancel()
 359                                 return client.RequestAndDecodeContext(ctx, nil, "DELETE", "arvados/v1/collections/"+collection.UUID, nil, nil)
 360                         })
 361                 }()
 362         }
 363
 364         diag.dotest(100, "uploading file via webdav", func() error {
 365                 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 366                 defer cancel()
 367                 if collection.UUID == "" {
 368                         return fmt.Errorf("skipping, no test collection")
 369                 }
 370                 req, err := http.NewRequestWithContext(ctx, "PUT", cluster.Services.WebDAVDownload.ExternalURL.String()+"c="+collection.UUID+"/testfile", bytes.NewBufferString("testfiledata"))
 371                 if err != nil {
 372                         return fmt.Errorf("BUG? http.NewRequest: %s", err)
 373                 }
 374                 req.Header.Set("Authorization", "Bearer "+client.AuthToken)
 375                 resp, err := http.DefaultClient.Do(req)
 376                 if err != nil {
 377                         return fmt.Errorf("error performing http request: %s", err)
 378                 }
 379                 resp.Body.Close()
 380                 if resp.StatusCode != http.StatusCreated {
 381                         return fmt.Errorf("status %s", resp.Status)
 382                 }
 383                 diag.debugf("ok, status %s", resp.Status)
 384                 err = client.RequestAndDecodeContext(ctx, &collection, "GET", "arvados/v1/collections/"+collection.UUID, nil, nil)
 385                 if err != nil {
 386                         return fmt.Errorf("get updated collection: %s", err)
 387                 }
 388                 diag.debugf("ok, pdh %s", collection.PortableDataHash)
 389                 return nil
 390         })
 391
 392         davurl := cluster.Services.WebDAV.ExternalURL
 393         diag.dotest(110, fmt.Sprintf("checking WebDAV ExternalURL wildcard (%s)", davurl), func() error {
 394                 if davurl.Host == "" {
 395                         return fmt.Errorf("host missing - content previews will not work")
 396                 }
 397                 if !strings.HasPrefix(davurl.Host, "*--") && !strings.HasPrefix(davurl.Host, "*.") && !cluster.Collections.TrustAllContent {
 398                         diag.warnf("WebDAV ExternalURL has no leading wildcard and TrustAllContent==false - content previews will not work")
 399                 }
 400                 return nil
 401         })
 402
 403         for i, trial := range []struct {
 404                 needcoll bool
 405                 status   int
 406                 fileurl  string
 407         }{
 408                 {false, http.StatusNotFound, strings.Replace(davurl.String(), "*", "d41d8cd98f00b204e9800998ecf8427e-0", 1) + "foo"},
 409                 {false, http.StatusNotFound, strings.Replace(davurl.String(), "*", "d41d8cd98f00b204e9800998ecf8427e-0", 1) + "testfile"},
 410                 {false, http.StatusNotFound, cluster.Services.WebDAVDownload.ExternalURL.String() + "c=d41d8cd98f00b204e9800998ecf8427e+0/_/foo"},
 411                 {false, http.StatusNotFound, cluster.Services.WebDAVDownload.ExternalURL.String() + "c=d41d8cd98f00b204e9800998ecf8427e+0/_/testfile"},
 412                 {true, http.StatusOK, strings.Replace(davurl.String(), "*", strings.Replace(collection.PortableDataHash, "+", "-", -1), 1) + "testfile"},
 413                 {true, http.StatusOK, cluster.Services.WebDAVDownload.ExternalURL.String() + "c=" + collection.UUID + "/_/testfile"},
 414         } {
 415                 diag.dotest(120+i, fmt.Sprintf("downloading from webdav (%s)", trial.fileurl), func() error {
 416                         ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 417                         defer cancel()
 418                         if trial.needcoll && collection.UUID == "" {
 419                                 return fmt.Errorf("skipping, no test collection")
 420                         }
 421                         req, err := http.NewRequestWithContext(ctx, "GET", trial.fileurl, nil)
 422                         if err != nil {
 423                                 return err
 424                         }
 425                         req.Header.Set("Authorization", "Bearer "+client.AuthToken)
 426                         resp, err := http.DefaultClient.Do(req)
 427                         if err != nil {
 428                                 return err
 429                         }
 430                         defer resp.Body.Close()
 431                         body, err := ioutil.ReadAll(resp.Body)
 432                         if err != nil {
 433                                 return fmt.Errorf("reading response: %s", err)
 434                         }
 435                         if resp.StatusCode != trial.status {
 436                                 return fmt.Errorf("unexpected response status: %s", resp.Status)
 437                         }
 438                         if trial.status == http.StatusOK && string(body) != "testfiledata" {
 439                                 return fmt.Errorf("unexpected response content: %q", body)
 440                         }
 441                         return nil
 442                 })
 443         }
 444
 445         var vm arvados.VirtualMachine
 446         diag.dotest(130, "getting list of virtual machines", func() error {
 447                 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 448                 defer cancel()
 449                 var vmlist arvados.VirtualMachineList
 450                 err := client.RequestAndDecodeContext(ctx, &vmlist, "GET", "arvados/v1/virtual_machines", nil, arvados.ListOptions{Limit: 999999})
 451                 if err != nil {
 452                         return err
 453                 }
 454                 if len(vmlist.Items) < 1 {
 455                         return fmt.Errorf("no VMs found")
 456                 }
 457                 vm = vmlist.Items[0]
 458                 return nil
 459         })
 460
 461         diag.dotest(140, "getting workbench1 webshell page", func() error {
 462                 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 463                 defer cancel()
 464                 if vm.UUID == "" {
 465                         return fmt.Errorf("skipping, no vm available")
 466                 }
 467                 webshelltermurl := cluster.Services.Workbench1.ExternalURL.String() + "virtual_machines/" + vm.UUID + "/webshell/testusername"
 468                 diag.debugf("url %s", webshelltermurl)
 469                 req, err := http.NewRequestWithContext(ctx, "GET", webshelltermurl, nil)
 470                 if err != nil {
 471                         return err
 472                 }
 473                 req.Header.Set("Authorization", "Bearer "+client.AuthToken)
 474                 resp, err := http.DefaultClient.Do(req)
 475                 if err != nil {
 476                         return err
 477                 }
 478                 defer resp.Body.Close()
 479                 body, err := ioutil.ReadAll(resp.Body)
 480                 if err != nil {
 481                         return fmt.Errorf("reading response: %s", err)
 482                 }
 483                 if resp.StatusCode != http.StatusOK {
 484                         return fmt.Errorf("unexpected response status: %s %q", resp.Status, body)
 485                 }
 486                 return nil
 487         })
 488
 489         diag.dotest(150, "connecting to webshell service", func() error {
 490                 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 491                 defer cancel()
 492                 if vm.UUID == "" {
 493                         return fmt.Errorf("skipping, no vm available")
 494                 }
 495                 u := cluster.Services.WebShell.ExternalURL
 496                 webshellurl := u.String() + vm.Hostname + "?"
 497                 if strings.HasPrefix(u.Host, "*") {
 498                         u.Host = vm.Hostname + u.Host[1:]
 499                         webshellurl = u.String() + "?"
 500                 }
 501                 diag.debugf("url %s", webshellurl)
 502                 req, err := http.NewRequestWithContext(ctx, "POST", webshellurl, bytes.NewBufferString(url.Values{
 503                         "width":   {"80"},
 504                         "height":  {"25"},
 505                         "session": {"xyzzy"},
 506                         "rooturl": {webshellurl},
 507                 }.Encode()))
 508                 if err != nil {
 509                         return err
 510                 }
 511                 req.Header.Set("Content-Type", "application/x-www-form-urlencoded; charset=UTF-8")
 512                 resp, err := http.DefaultClient.Do(req)
 513                 if err != nil {
 514                         return err
 515                 }
 516                 defer resp.Body.Close()
 517                 diag.debugf("response status %s", resp.Status)
 518                 body, err := ioutil.ReadAll(resp.Body)
 519                 if err != nil {
 520                         return fmt.Errorf("reading response: %s", err)
 521                 }
 522                 diag.debugf("response body %q", body)
 523                 // We don't speak the protocol, so we get a 400 error
 524                 // from the webshell server even if everything is
 525                 // OK. Anything else (404, 502, ???) indicates a
 526                 // problem.
 527                 if resp.StatusCode != http.StatusBadRequest {
 528                         return fmt.Errorf("unexpected response status: %s, %q", resp.Status, body)
 529                 }
 530                 return nil
 531         })
 532
 533         diag.dotest(160, "running a container", func() error {
 534                 if diag.priority < 1 {
 535                         diag.infof("skipping (use priority > 0 if you want to run a container)")
 536                         return nil
 537                 }
 538                 if project.UUID == "" {
 539                         return fmt.Errorf("skipping, no project to work in")
 540                 }
 541
 542                 var cr arvados.ContainerRequest
 543                 ctx, cancel := context.WithDeadline(context.Background(), time.Now().Add(diag.timeout))
 544                 defer cancel()
 545
 546                 timestamp := time.Now().Format(time.RFC3339)
 547                 err := client.RequestAndDecodeContext(ctx, &cr, "POST", "arvados/v1/container_requests", nil, map[string]interface{}{"container_request": map[string]interface{}{
 548                         "owner_uuid":      project.UUID,
 549                         "name":            fmt.Sprintf("diagnostics container request %s", timestamp),
 550                         "container_image": diag.dockerImage,
 551                         "command":         []string{"echo", timestamp},
 552                         "use_existing":    false,
 553                         "output_path":     "/mnt/output",
 554                         "output_name":     fmt.Sprintf("diagnostics output %s", timestamp),
 555                         "priority":        diag.priority,
 556                         "state":           arvados.ContainerRequestStateCommitted,
 557                         "mounts": map[string]map[string]interface{}{
 558                                 "/mnt/output": {
 559                                         "kind":     "collection",
 560                                         "writable": true,
 561                                 },
 562                         },
 563                         "runtime_constraints": arvados.RuntimeConstraints{
 564                                 VCPUs:        1,
 565                                 RAM:          1 << 26,
 566                                 KeepCacheRAM: 1 << 26,
 567                         },
 568                 }})
 569                 if err != nil {
 570                         return err
 571                 }
 572                 diag.debugf("container request uuid = %s", cr.UUID)
 573                 diag.debugf("container uuid = %s", cr.ContainerUUID)
 574
 575                 timeout := 10 * time.Minute
 576                 diag.infof("container request submitted, waiting up to %v for container to run", arvados.Duration(timeout))
 577                 ctx, cancel = context.WithDeadline(context.Background(), time.Now().Add(timeout))
 578                 defer cancel()
 579
 580                 var c arvados.Container
 581                 for ; cr.State != arvados.ContainerRequestStateFinal; time.Sleep(2 * time.Second) {
 582                         ctx, cancel := context.WithDeadline(ctx, time.Now().Add(diag.timeout))
 583                         defer cancel()
 584
 585                         crStateWas := cr.State
 586                         err := client.RequestAndDecodeContext(ctx, &cr, "GET", "arvados/v1/container_requests/"+cr.UUID, nil, nil)
 587                         if err != nil {
 588                                 return err
 589                         }
 590                         if cr.State != crStateWas {
 591                                 diag.debugf("container request state = %s", cr.State)
 592                         }
 593
 594                         cStateWas := c.State
 595                         err = client.RequestAndDecodeContext(ctx, &c, "GET", "arvados/v1/containers/"+cr.ContainerUUID, nil, nil)
 596                         if err != nil {
 597                                 return err
 598                         }
 599                         if c.State != cStateWas {
 600                                 diag.debugf("container state = %s", c.State)
 601                         }
 602                 }
 603
 604                 if c.State != arvados.ContainerStateComplete {
 605                         return fmt.Errorf("container request %s is final but container %s did not complete: container state = %q", cr.UUID, cr.ContainerUUID, c.State)
 606                 } else if c.ExitCode != 0 {
 607                         return fmt.Errorf("container exited %d", c.ExitCode)
 608                 }
 609                 return nil
 610         })
 611 }