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