lib/service/cmd_test.go

   1 // Copyright (C) The Arvados Authors. All rights reserved.
   2 //
   3 // SPDX-License-Identifier: Apache-2.0
   4
   5 // package service provides a cmd.Handler that brings up a system service.
   6 package service
   7
   8 import (
   9         "bytes"
  10         "context"
  11         "crypto/tls"
  12         "encoding/json"
  13         "fmt"
  14         "io/ioutil"
  15         "net"
  16         "net/http"
  17         "net/url"
  18         "os"
  19         "strings"
  20         "sync"
  21         "sync/atomic"
  22         "testing"
  23         "time"
  24
  25         "git.arvados.org/arvados.git/sdk/go/arvados"
  26         "git.arvados.org/arvados.git/sdk/go/ctxlog"
  27         "github.com/prometheus/client_golang/prometheus"
  28         check "gopkg.in/check.v1"
  29 )
  30
  31 func Test(t *testing.T) {
  32         check.TestingT(t)
  33 }
  34
  35 var _ = check.Suite(&Suite{})
  36
  37 type Suite struct{}
  38 type key int
  39
  40 const (
  41         contextKey key = iota
  42 )
  43
  44 func unusedPort(c *check.C) string {
  45         // Find an available port on the testing host, so the test
  46         // cases don't get confused by "already in use" errors.
  47         listener, err := net.Listen("tcp", ":")
  48         c.Assert(err, check.IsNil)
  49         listener.Close()
  50         _, port, err := net.SplitHostPort(listener.Addr().String())
  51         c.Assert(err, check.IsNil)
  52         return port
  53 }
  54
  55 func (*Suite) TestGetListenAddress(c *check.C) {
  56         port := unusedPort(c)
  57         defer os.Unsetenv("ARVADOS_SERVICE_INTERNAL_URL")
  58         for idx, trial := range []struct {
  59                 // internalURL => listenURL, both with trailing "/"
  60                 // because config loader always adds it
  61                 internalURLs     map[string]string
  62                 envVar           string
  63                 expectErrorMatch string
  64                 expectLogsMatch  string
  65                 expectListen     string
  66                 expectInternal   string
  67         }{
  68                 {
  69                         internalURLs:   map[string]string{"http://localhost:" + port + "/": ""},
  70                         expectListen:   "http://localhost:" + port + "/",
  71                         expectInternal: "http://localhost:" + port + "/",
  72                 },
  73                 { // implicit port 80 in InternalURLs
  74                         internalURLs:     map[string]string{"http://localhost/": ""},
  75                         expectErrorMatch: `.*:80: bind: permission denied`,
  76                 },
  77                 { // implicit port 443 in InternalURLs
  78                         internalURLs:   map[string]string{"https://host.example/": "http://localhost:" + port + "/"},
  79                         expectListen:   "http://localhost:" + port + "/",
  80                         expectInternal: "https://host.example/",
  81                 },
  82                 { // implicit port 443 in ListenURL
  83                         internalURLs:     map[string]string{"wss://host.example/": "wss://localhost/"},
  84                         expectErrorMatch: `.*:443: bind: permission denied`,
  85                 },
  86                 {
  87                         internalURLs:   map[string]string{"https://hostname.example/": "http://localhost:8000/"},
  88                         expectListen:   "http://localhost:8000/",
  89                         expectInternal: "https://hostname.example/",
  90                 },
  91                 {
  92                         internalURLs: map[string]string{
  93                                 "https://hostname1.example/": "http://localhost:12435/",
  94                                 "https://hostname2.example/": "http://localhost:" + port + "/",
  95                         },
  96                         envVar:         "https://hostname2.example", // note this works despite missing trailing "/"
  97                         expectListen:   "http://localhost:" + port + "/",
  98                         expectInternal: "https://hostname2.example/",
  99                 },
 100                 { // cannot listen on any of the ListenURLs
 101                         internalURLs: map[string]string{
 102                                 "https://hostname1.example/": "http://1.2.3.4:" + port + "/",
 103                                 "https://hostname2.example/": "http://1.2.3.4:" + port + "/",
 104                         },
 105                         expectErrorMatch: "configuration does not enable the \"arvados-controller\" service on this host",
 106                 },
 107                 { // cannot listen on any of the (implied) ListenURLs
 108                         internalURLs: map[string]string{
 109                                 "https://1.2.3.4/": "",
 110                                 "https://1.2.3.5/": "",
 111                         },
 112                         expectErrorMatch: "configuration does not enable the \"arvados-controller\" service on this host",
 113                 },
 114                 { // impossible port number
 115                         internalURLs: map[string]string{
 116                                 "https://host.example/": "http://0.0.0.0:1234567",
 117                         },
 118                         expectErrorMatch: `.*:1234567: listen tcp: address 1234567: invalid port`,
 119                 },
 120                 {
 121                         // env var URL not mentioned in config = obey env var, with warning
 122                         internalURLs:    map[string]string{"https://hostname1.example/": "http://localhost:8000/"},
 123                         envVar:          "https://hostname2.example",
 124                         expectListen:    "https://hostname2.example/",
 125                         expectInternal:  "https://hostname2.example/",
 126                         expectLogsMatch: `.*\Qpossible configuration error: listening on https://hostname2.example/ (from $ARVADOS_SERVICE_INTERNAL_URL) even though configuration does not have a matching InternalURLs entry\E.*\n`,
 127                 },
 128                 {
 129                         // env var + empty config = obey env var, with warning
 130                         envVar:          "https://hostname.example",
 131                         expectListen:    "https://hostname.example/",
 132                         expectInternal:  "https://hostname.example/",
 133                         expectLogsMatch: `.*\Qpossible configuration error: listening on https://hostname.example/ (from $ARVADOS_SERVICE_INTERNAL_URL) even though configuration does not have a matching InternalURLs entry\E.*\n`,
 134                 },
 135         } {
 136                 c.Logf("trial %d %+v", idx, trial)
 137                 os.Setenv("ARVADOS_SERVICE_INTERNAL_URL", trial.envVar)
 138                 var logbuf bytes.Buffer
 139                 log := ctxlog.New(&logbuf, "text", "info")
 140                 services := arvados.Services{Controller: arvados.Service{InternalURLs: map[arvados.URL]arvados.ServiceInstance{}}}
 141                 for k, v := range trial.internalURLs {
 142                         u, err := url.Parse(k)
 143                         c.Assert(err, check.IsNil)
 144                         si := arvados.ServiceInstance{}
 145                         if v != "" {
 146                                 u, err := url.Parse(v)
 147                                 c.Assert(err, check.IsNil)
 148                                 si.ListenURL = arvados.URL(*u)
 149                         }
 150                         services.Controller.InternalURLs[arvados.URL(*u)] = si
 151                 }
 152                 listenURL, internalURL, err := getListenAddr(services, "arvados-controller", log)
 153                 if trial.expectLogsMatch != "" {
 154                         c.Check(logbuf.String(), check.Matches, trial.expectLogsMatch)
 155                 }
 156                 if trial.expectErrorMatch != "" {
 157                         c.Check(err, check.ErrorMatches, trial.expectErrorMatch)
 158                         continue
 159                 }
 160                 if !c.Check(err, check.IsNil) {
 161                         continue
 162                 }
 163                 c.Check(listenURL.String(), check.Equals, trial.expectListen)
 164                 c.Check(internalURL.String(), check.Equals, trial.expectInternal)
 165         }
 166 }
 167
 168 func (*Suite) TestCommand(c *check.C) {
 169         cf, err := ioutil.TempFile("", "cmd_test.")
 170         c.Assert(err, check.IsNil)
 171         defer os.Remove(cf.Name())
 172         defer cf.Close()
 173         fmt.Fprintf(cf, "Clusters:\n zzzzz:\n  SystemRootToken: abcde\n  NodeProfiles: {\"*\": {\"arvados-controller\": {Listen: \":1234\"}}}")
 174
 175         healthCheck := make(chan bool, 1)
 176         ctx, cancel := context.WithCancel(context.Background())
 177         defer cancel()
 178
 179         cmd := Command(arvados.ServiceNameController, func(ctx context.Context, _ *arvados.Cluster, token string, reg *prometheus.Registry) Handler {
 180                 c.Check(ctx.Value(contextKey), check.Equals, "bar")
 181                 c.Check(token, check.Equals, "abcde")
 182                 return &testHandler{ctx: ctx, healthCheck: healthCheck}
 183         })
 184         cmd.(*command).ctx = context.WithValue(ctx, contextKey, "bar")
 185
 186         done := make(chan bool)
 187         var stdin, stdout, stderr bytes.Buffer
 188
 189         go func() {
 190                 cmd.RunCommand("arvados-controller", []string{"-config", cf.Name()}, &stdin, &stdout, &stderr)
 191                 close(done)
 192         }()
 193         select {
 194         case <-healthCheck:
 195         case <-done:
 196                 c.Error("command exited without health check")
 197         }
 198         cancel()
 199         c.Check(stdout.String(), check.Equals, "")
 200         c.Check(stderr.String(), check.Matches, `(?ms).*"msg":"CheckHealth called".*`)
 201 }
 202
 203 func (s *Suite) TestRequestLimitsAndDumpRequests_Keepweb(c *check.C) {
 204         s.testRequestLimitAndDumpRequests(c, arvados.ServiceNameKeepweb, "MaxConcurrentRequests")
 205 }
 206
 207 func (s *Suite) TestRequestLimitsAndDumpRequests_Controller(c *check.C) {
 208         s.testRequestLimitAndDumpRequests(c, arvados.ServiceNameController, "MaxConcurrentRailsRequests")
 209 }
 210
 211 func (*Suite) testRequestLimitAndDumpRequests(c *check.C, serviceName arvados.ServiceName, maxReqsConfigKey string) {
 212         defer func(orig time.Duration) { requestQueueDumpCheckInterval = orig }(requestQueueDumpCheckInterval)
 213         requestQueueDumpCheckInterval = time.Second / 10
 214
 215         port := unusedPort(c)
 216         tmpdir := c.MkDir()
 217         cf, err := ioutil.TempFile(tmpdir, "cmd_test.")
 218         c.Assert(err, check.IsNil)
 219         defer os.Remove(cf.Name())
 220         defer cf.Close()
 221
 222         max := 24
 223         maxTunnels := 30
 224         fmt.Fprintf(cf, `
 225 Clusters:
 226  zzzzz:
 227   SystemRootToken: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
 228   ManagementToken: bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb
 229   API:
 230    `+maxReqsConfigKey+`: %d
 231    MaxQueuedRequests: 1
 232    MaxGatewayTunnels: %d
 233   SystemLogs: {RequestQueueDumpDirectory: %q}
 234   Services:
 235    Controller:
 236     ExternalURL: "http://localhost:`+port+`"
 237     InternalURLs: {"http://localhost:`+port+`": {}}
 238    WebDAV:
 239     ExternalURL: "http://localhost:`+port+`"
 240     InternalURLs: {"http://localhost:`+port+`": {}}
 241 `, max, maxTunnels, tmpdir)
 242         cf.Close()
 243
 244         started := make(chan bool, max+1)
 245         hold := make(chan bool)
 246         handler := http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 247                 if strings.HasPrefix(r.URL.Path, "/arvados/v1/connect") {
 248                         <-hold
 249                 } else {
 250                         started <- true
 251                         <-hold
 252                 }
 253         })
 254         healthCheck := make(chan bool, 1)
 255         ctx, cancel := context.WithCancel(context.Background())
 256         defer cancel()
 257
 258         cmd := Command(serviceName, func(ctx context.Context, _ *arvados.Cluster, token string, reg *prometheus.Registry) Handler {
 259                 return &testHandler{ctx: ctx, handler: handler, healthCheck: healthCheck}
 260         })
 261         cmd.(*command).ctx = context.WithValue(ctx, contextKey, "bar")
 262
 263         exited := make(chan bool)
 264         var stdin, stdout, stderr bytes.Buffer
 265
 266         go func() {
 267                 cmd.RunCommand(string(serviceName), []string{"-config", cf.Name()}, &stdin, &stdout, &stderr)
 268                 close(exited)
 269         }()
 270         select {
 271         case <-healthCheck:
 272         case <-exited:
 273                 c.Logf("%s", stderr.String())
 274                 c.Error("command exited without health check")
 275         }
 276         client := http.Client{}
 277         deadline := time.Now().Add(time.Second * 2)
 278         var activeReqs sync.WaitGroup
 279
 280         // Start some API reqs
 281         var apiResp200, apiResp503 int64
 282         for i := 0; i < max+1; i++ {
 283                 activeReqs.Add(1)
 284                 go func() {
 285                         defer activeReqs.Done()
 286                         resp, err := client.Get("http://localhost:" + port + "/testpath")
 287                         for err != nil && strings.Contains(err.Error(), "dial tcp") && deadline.After(time.Now()) {
 288                                 time.Sleep(time.Second / 100)
 289                                 resp, err = client.Get("http://localhost:" + port + "/testpath")
 290                         }
 291                         if c.Check(err, check.IsNil) {
 292                                 if resp.StatusCode == http.StatusOK {
 293                                         atomic.AddInt64(&apiResp200, 1)
 294                                 } else if resp.StatusCode == http.StatusServiceUnavailable {
 295                                         atomic.AddInt64(&apiResp503, 1)
 296                                 }
 297                         }
 298                 }()
 299         }
 300
 301         // Start some gateway tunnel reqs that don't count toward our
 302         // API req limit
 303         extraTunnelReqs := 20
 304         var tunnelResp200, tunnelResp503 int64
 305         for i := 0; i < maxTunnels+extraTunnelReqs; i++ {
 306                 activeReqs.Add(1)
 307                 go func() {
 308                         defer activeReqs.Done()
 309                         resp, err := client.Get("http://localhost:" + port + "/arvados/v1/connect/...")
 310                         for err != nil && strings.Contains(err.Error(), "dial tcp") && deadline.After(time.Now()) {
 311                                 time.Sleep(time.Second / 100)
 312                                 resp, err = client.Get("http://localhost:" + port + "/arvados/v1/connect/...")
 313                         }
 314                         if c.Check(err, check.IsNil) {
 315                                 if resp.StatusCode == http.StatusOK {
 316                                         atomic.AddInt64(&tunnelResp200, 1)
 317                                 } else if resp.StatusCode == http.StatusServiceUnavailable {
 318                                         atomic.AddInt64(&tunnelResp503, 1)
 319                                 } else {
 320                                         c.Errorf("tunnel response code %d", resp.StatusCode)
 321                                 }
 322                         }
 323                 }()
 324         }
 325         for i := 0; i < max; i++ {
 326                 select {
 327                 case <-started:
 328                 case <-time.After(time.Second):
 329                         c.Logf("%s", stderr.String())
 330                         c.Fatal("timed out")
 331                 }
 332         }
 333         for delay := time.Second / 100; ; delay = delay * 2 {
 334                 time.Sleep(delay)
 335                 j, err := os.ReadFile(tmpdir + "/" + string(serviceName) + "-requests.json")
 336                 if os.IsNotExist(err) && deadline.After(time.Now()) {
 337                         continue
 338                 }
 339                 c.Assert(err, check.IsNil)
 340                 c.Logf("stderr:\n%s", stderr.String())
 341                 c.Logf("json:\n%s", string(j))
 342
 343                 var loaded []struct{ URL string }
 344                 err = json.Unmarshal(j, &loaded)
 345                 c.Check(err, check.IsNil)
 346
 347                 for i := 0; i < len(loaded); i++ {
 348                         if strings.HasPrefix(loaded[i].URL, "/arvados/v1/connect/") {
 349                                 // Filter out a gateway tunnel req
 350                                 // that doesn't count toward our API
 351                                 // req limit
 352                                 if i < len(loaded)-1 {
 353                                         copy(loaded[i:], loaded[i+1:])
 354                                         i--
 355                                 }
 356                                 loaded = loaded[:len(loaded)-1]
 357                         }
 358                 }
 359
 360                 if len(loaded) < max {
 361                         // Dumped when #requests was >90% but <100% of
 362                         // limit. If we stop now, we won't be able to
 363                         // confirm (below) that management endpoints
 364                         // are still accessible when normal requests
 365                         // are at 100%.
 366                         c.Logf("loaded dumped requests, but len %d < max %d -- still waiting", len(loaded), max)
 367                         continue
 368                 }
 369                 c.Check(loaded, check.HasLen, max+1)
 370                 c.Check(loaded[0].URL, check.Equals, "/testpath")
 371                 break
 372         }
 373
 374         for _, path := range []string{"/_inspect/requests", "/metrics"} {
 375                 req, err := http.NewRequest("GET", "http://localhost:"+port+""+path, nil)
 376                 c.Assert(err, check.IsNil)
 377                 req.Header.Set("Authorization", "Bearer bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb")
 378                 resp, err := client.Do(req)
 379                 if !c.Check(err, check.IsNil) {
 380                         break
 381                 }
 382                 c.Logf("got response for %s", path)
 383                 c.Check(resp.StatusCode, check.Equals, http.StatusOK)
 384                 buf, err := ioutil.ReadAll(resp.Body)
 385                 c.Check(err, check.IsNil)
 386                 switch path {
 387                 case "/metrics":
 388                         c.Check(string(buf), check.Matches, `(?ms).*arvados_concurrent_requests{queue="api"} `+fmt.Sprintf("%d", max)+`\n.*`)
 389                         c.Check(string(buf), check.Matches, `(?ms).*arvados_queued_requests{priority="normal",queue="api"} 1\n.*`)
 390                 case "/_inspect/requests":
 391                         c.Check(string(buf), check.Matches, `(?ms).*"URL":"/testpath".*`)
 392                 default:
 393                         c.Error("oops, testing bug")
 394                 }
 395         }
 396         close(hold)
 397         activeReqs.Wait()
 398         c.Check(int(apiResp200), check.Equals, max+1)
 399         c.Check(int(apiResp503), check.Equals, 0)
 400         c.Check(int(tunnelResp200), check.Equals, maxTunnels)
 401         c.Check(int(tunnelResp503), check.Equals, extraTunnelReqs)
 402         cancel()
 403 }
 404
 405 func (*Suite) TestTLS(c *check.C) {
 406         port := unusedPort(c)
 407         cwd, err := os.Getwd()
 408         c.Assert(err, check.IsNil)
 409
 410         stdin := bytes.NewBufferString(`
 411 Clusters:
 412  zzzzz:
 413   SystemRootToken: abcde
 414   Services:
 415    Controller:
 416     ExternalURL: "https://localhost:` + port + `"
 417     InternalURLs: {"https://localhost:` + port + `": {}}
 418   TLS:
 419    Key: file://` + cwd + `/../../services/api/tmp/self-signed.key
 420    Certificate: file://` + cwd + `/../../services/api/tmp/self-signed.pem
 421 `)
 422
 423         called := make(chan bool)
 424         cmd := Command(arvados.ServiceNameController, func(ctx context.Context, _ *arvados.Cluster, token string, reg *prometheus.Registry) Handler {
 425                 return &testHandler{handler: http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 426                         w.Write([]byte("ok"))
 427                         close(called)
 428                 })}
 429         })
 430
 431         exited := make(chan bool)
 432         var stdout, stderr bytes.Buffer
 433         go func() {
 434                 cmd.RunCommand("arvados-controller", []string{"-config", "-"}, stdin, &stdout, &stderr)
 435                 close(exited)
 436         }()
 437         got := make(chan bool)
 438         go func() {
 439                 defer close(got)
 440                 client := &http.Client{Transport: &http.Transport{TLSClientConfig: &tls.Config{InsecureSkipVerify: true}}}
 441                 for range time.NewTicker(time.Millisecond).C {
 442                         resp, err := client.Get("https://localhost:" + port)
 443                         if err != nil {
 444                                 c.Log(err)
 445                                 continue
 446                         }
 447                         body, err := ioutil.ReadAll(resp.Body)
 448                         c.Check(err, check.IsNil)
 449                         c.Logf("status %d, body %s", resp.StatusCode, string(body))
 450                         c.Check(resp.StatusCode, check.Equals, http.StatusOK)
 451                         break
 452                 }
 453         }()
 454         select {
 455         case <-called:
 456         case <-exited:
 457                 c.Error("command exited without calling handler")
 458         case <-time.After(time.Second):
 459                 c.Error("timed out")
 460         }
 461         select {
 462         case <-got:
 463         case <-exited:
 464                 c.Error("command exited before client received response")
 465         case <-time.After(time.Second):
 466                 c.Error("timed out")
 467         }
 468         c.Log(stderr.String())
 469 }
 470
 471 type testHandler struct {
 472         ctx         context.Context
 473         handler     http.Handler
 474         healthCheck chan bool
 475 }
 476
 477 func (th *testHandler) Done() <-chan struct{}                            { return nil }
 478 func (th *testHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) { th.handler.ServeHTTP(w, r) }
 479 func (th *testHandler) CheckHealth() error {
 480         ctxlog.FromContext(th.ctx).Info("CheckHealth called")
 481         select {
 482         case th.healthCheck <- true:
 483         default:
 484         }
 485         return nil
 486 }