1 // Copyright (C) The Arvados Authors. All rights reserved.
3 // SPDX-License-Identifier: AGPL-3.0
21 "git.arvados.org/arvados.git/sdk/go/arvados"
22 "git.arvados.org/arvados.git/sdk/go/ctxlog"
24 "github.com/aws/aws-sdk-go-v2/aws"
25 "github.com/aws/aws-sdk-go-v2/feature/s3/manager"
26 "github.com/aws/aws-sdk-go-v2/service/s3"
28 "github.com/johannesboyne/gofakes3"
29 "github.com/johannesboyne/gofakes3/backend/s3mem"
30 "github.com/prometheus/client_golang/prometheus"
31 "github.com/sirupsen/logrus"
32 check "gopkg.in/check.v1"
35 type s3fakeClock struct {
39 func (c *s3fakeClock) Now() time.Time {
41 return time.Now().UTC()
46 func (c *s3fakeClock) Since(t time.Time) time.Duration {
50 var _ = check.Suite(&stubbedS3Suite{})
52 var srv httptest.Server
54 type stubbedS3Suite struct {
55 s3server *httptest.Server
56 s3fakeClock *s3fakeClock
57 metadata *httptest.Server
58 cluster *arvados.Cluster
59 volumes []*testableS3Volume
62 func (s *stubbedS3Suite) SetUpTest(c *check.C) {
64 s.s3fakeClock = &s3fakeClock{}
66 s.cluster = testCluster(c)
67 s.cluster.Volumes = map[string]arvados.Volume{
68 "zzzzz-nyw5e-000000000000000": {Driver: "S3"},
69 "zzzzz-nyw5e-111111111111111": {Driver: "S3"},
73 func (s *stubbedS3Suite) TearDownTest(c *check.C) {
74 if s.s3server != nil {
79 func (s *stubbedS3Suite) TestGeneric(c *check.C) {
80 DoGenericVolumeTests(c, false, func(t TB, params newVolumeParams) TestableVolume {
81 // Use a negative raceWindow so s3test's 1-second
82 // timestamp precision doesn't confuse fixRace.
83 return s.newTestableVolume(c, params, -2*time.Second)
87 func (s *stubbedS3Suite) TestGenericReadOnly(c *check.C) {
88 DoGenericVolumeTests(c, true, func(t TB, params newVolumeParams) TestableVolume {
89 return s.newTestableVolume(c, params, -2*time.Second)
93 func (s *stubbedS3Suite) TestGenericWithPrefix(c *check.C) {
94 DoGenericVolumeTests(c, false, func(t TB, params newVolumeParams) TestableVolume {
95 v := s.newTestableVolume(c, params, -2*time.Second)
101 func (s *stubbedS3Suite) TestIndex(c *check.C) {
102 v := s.newTestableVolume(c, newVolumeParams{
104 ConfigVolume: arvados.Volume{Replication: 2},
105 MetricsVecs: newVolumeMetricsVecs(prometheus.NewRegistry()),
106 BufferPool: newBufferPool(ctxlog.TestLogger(c), 8, prometheus.NewRegistry()),
109 for i := 0; i < 256; i++ {
110 err := v.blockWriteWithoutMD5Check(fmt.Sprintf("%02x%030x", i, i), []byte{102, 111, 111})
111 c.Assert(err, check.IsNil)
113 for _, spec := range []struct {
122 buf := new(bytes.Buffer)
123 err := v.Index(context.Background(), spec.prefix, buf)
124 c.Check(err, check.IsNil)
126 idx := bytes.SplitAfter(buf.Bytes(), []byte{10})
127 c.Check(len(idx), check.Equals, spec.expectMatch+1)
128 c.Check(len(idx[len(idx)-1]), check.Equals, 0)
132 func (s *stubbedS3Suite) TestSignature(c *check.C) {
133 var header http.Header
134 stub := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
139 // The aws-sdk-go-v2 driver only supports S3 V4 signatures. S3 v2 signatures are being phased out
140 // as of June 24, 2020. Cf. https://forums.aws.amazon.com/ann.jspa?annID=5816
142 S3VolumeDriverParameters: arvados.S3VolumeDriverParameters{
144 SecretAccessKey: "xxx",
146 Region: "test-region-1",
147 Bucket: "test-bucket-name",
151 logger: ctxlog.TestLogger(c),
152 metrics: newVolumeMetricsVecs(prometheus.NewRegistry()),
156 c.Check(err, check.IsNil)
157 err = vol.BlockWrite(context.Background(), "acbd18db4cc2f85cedef654fccc4a4d8", []byte("foo"))
158 c.Check(err, check.IsNil)
159 c.Check(header.Get("Authorization"), check.Matches, `AWS4-HMAC-SHA256 .*`)
162 func (s *stubbedS3Suite) TestIAMRoleCredentials(c *check.C) {
163 var reqHeader http.Header
164 stub := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
169 retrievedMetadata := false
170 s.metadata = httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
171 retrievedMetadata = true
172 upd := time.Now().UTC().Add(-time.Hour).Format(time.RFC3339)
173 exp := time.Now().UTC().Add(time.Hour).Format(time.RFC3339)
174 c.Logf("metadata stub received request: %s %s", r.Method, r.URL.Path)
176 case r.URL.Path == "/latest/meta-data/iam/security-credentials/":
177 io.WriteString(w, "testcredential\n")
178 case r.URL.Path == "/latest/api/token",
179 r.URL.Path == "/latest/meta-data/iam/security-credentials/testcredential":
180 // Literal example from
181 // https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/iam-roles-for-amazon-ec2.html#instance-metadata-security-credentials
182 // but with updated timestamps
183 io.WriteString(w, `{"Code":"Success","LastUpdated":"`+upd+`","Type":"AWS-HMAC","AccessKeyId":"ASIAIOSFODNN7EXAMPLE","SecretAccessKey":"wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY","Token":"token","Expiration":"`+exp+`"}`)
185 w.WriteHeader(http.StatusNotFound)
188 defer s.metadata.Close()
191 S3VolumeDriverParameters: arvados.S3VolumeDriverParameters{
193 Region: "test-region-1",
194 Bucket: "test-bucket-name",
197 logger: ctxlog.TestLogger(c),
198 metrics: newVolumeMetricsVecs(prometheus.NewRegistry()),
200 err := v.check(s.metadata.URL + "/latest")
201 c.Check(err, check.IsNil)
202 resp, err := v.bucket.svc.ListBuckets(context.Background(), &s3.ListBucketsInput{})
203 c.Check(err, check.IsNil)
204 c.Check(resp.Buckets, check.HasLen, 0)
205 c.Check(retrievedMetadata, check.Equals, true)
206 c.Check(reqHeader.Get("Authorization"), check.Matches, `AWS4-HMAC-SHA256 Credential=ASIAIOSFODNN7EXAMPLE/\d+/test-region-1/s3/aws4_request, SignedHeaders=.*`)
208 retrievedMetadata = false
209 s.metadata = httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
210 retrievedMetadata = true
211 c.Logf("metadata stub received request: %s %s", r.Method, r.URL.Path)
212 w.WriteHeader(http.StatusNotFound)
215 S3VolumeDriverParameters: arvados.S3VolumeDriverParameters{
216 Endpoint: "http://localhost:9",
217 Region: "test-region-1",
218 Bucket: "test-bucket-name",
221 logger: ctxlog.TestLogger(c),
222 metrics: newVolumeMetricsVecs(prometheus.NewRegistry()),
224 err = deadv.check(s.metadata.URL + "/latest")
225 c.Check(err, check.IsNil)
226 _, err = deadv.bucket.svc.ListBuckets(context.Background(), &s3.ListBucketsInput{})
227 c.Check(err, check.ErrorMatches, `(?s).*failed to refresh cached credentials, no EC2 IMDS role found.*`)
228 c.Check(err, check.ErrorMatches, `(?s).*404.*`)
229 c.Check(retrievedMetadata, check.Equals, true)
232 func (s *stubbedS3Suite) TestStats(c *check.C) {
233 v := s.newTestableVolume(c, newVolumeParams{
235 ConfigVolume: arvados.Volume{Replication: 2},
236 MetricsVecs: newVolumeMetricsVecs(prometheus.NewRegistry()),
237 BufferPool: newBufferPool(ctxlog.TestLogger(c), 8, prometheus.NewRegistry()),
239 stats := func() string {
240 buf, err := json.Marshal(v.InternalStats())
241 c.Check(err, check.IsNil)
245 c.Check(stats(), check.Matches, `.*"Ops":0,.*`)
247 loc := "acbd18db4cc2f85cedef654fccc4a4d8"
248 err := v.BlockRead(context.Background(), loc, brdiscard)
249 c.Check(err, check.NotNil)
250 c.Check(stats(), check.Matches, `.*"Ops":[^0],.*`)
251 c.Check(stats(), check.Matches, `.*"\*smithy.OperationError 404 NoSuchKey":[^0].*`)
252 c.Check(stats(), check.Matches, `.*"InBytes":0,.*`)
254 err = v.BlockWrite(context.Background(), loc, []byte("foo"))
255 c.Check(err, check.IsNil)
256 c.Check(stats(), check.Matches, `.*"OutBytes":3,.*`)
257 c.Check(stats(), check.Matches, `.*"PutOps":2,.*`)
259 err = v.BlockRead(context.Background(), loc, brdiscard)
260 c.Check(err, check.IsNil)
261 err = v.BlockRead(context.Background(), loc, brdiscard)
262 c.Check(err, check.IsNil)
263 c.Check(stats(), check.Matches, `.*"InBytes":6,.*`)
266 type s3AWSBlockingHandler struct {
267 requested chan *http.Request
268 unblock chan struct{}
271 func (h *s3AWSBlockingHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
272 if r.Method == "PUT" && !strings.Contains(strings.Trim(r.URL.Path, "/"), "/") {
273 // Accept PutBucket ("PUT /bucketname/"), called by
277 if h.requested != nil {
280 if h.unblock != nil {
283 http.Error(w, "nothing here", http.StatusNotFound)
286 func (s *stubbedS3Suite) TestGetContextCancel(c *check.C) {
287 s.testContextCancel(c, func(ctx context.Context, v *testableS3Volume) error {
288 return v.BlockRead(ctx, fooHash, brdiscard)
292 func (s *stubbedS3Suite) TestPutContextCancel(c *check.C) {
293 s.testContextCancel(c, func(ctx context.Context, v *testableS3Volume) error {
294 return v.BlockWrite(ctx, fooHash, []byte("foo"))
298 func (s *stubbedS3Suite) testContextCancel(c *check.C, testFunc func(context.Context, *testableS3Volume) error) {
299 handler := &s3AWSBlockingHandler{}
300 s.s3server = httptest.NewServer(handler)
301 defer s.s3server.Close()
303 v := s.newTestableVolume(c, newVolumeParams{
305 ConfigVolume: arvados.Volume{Replication: 2},
306 MetricsVecs: newVolumeMetricsVecs(prometheus.NewRegistry()),
307 BufferPool: newBufferPool(ctxlog.TestLogger(c), 8, prometheus.NewRegistry()),
310 ctx, cancel := context.WithCancel(context.Background())
312 handler.requested = make(chan *http.Request)
313 handler.unblock = make(chan struct{})
314 defer close(handler.unblock)
316 doneFunc := make(chan struct{})
318 err := testFunc(ctx, v)
319 c.Check(err, check.Equals, context.Canceled)
323 timeout := time.After(10 * time.Second)
325 // Wait for the stub server to receive a request, meaning
326 // Get() is waiting for an s3 operation.
329 c.Fatal("timed out waiting for test func to call our handler")
331 c.Fatal("test func finished without even calling our handler!")
332 case <-handler.requested:
344 func (s *stubbedS3Suite) TestBackendStates(c *check.C) {
345 s.cluster.Collections.BlobTrashLifetime.Set("1h")
346 s.cluster.Collections.BlobSigningTTL.Set("1h")
348 v := s.newTestableVolume(c, newVolumeParams{
350 ConfigVolume: arvados.Volume{Replication: 2},
351 Logger: ctxlog.TestLogger(c),
352 MetricsVecs: newVolumeMetricsVecs(prometheus.NewRegistry()),
353 BufferPool: newBufferPool(ctxlog.TestLogger(c), 8, prometheus.NewRegistry()),
357 putS3Obj := func(t time.Time, key string, data []byte) {
361 s.s3fakeClock.now = &t
362 uploader := manager.NewUploader(v.bucket.svc)
363 _, err := uploader.Upload(context.Background(), &s3.PutObjectInput{
364 Bucket: aws.String(v.bucket.bucket),
365 Key: aws.String(key),
366 Body: bytes.NewReader(data),
371 s.s3fakeClock.now = nil
380 for _, scenario := range []struct {
387 canGetAfterTrash bool
389 haveTrashAfterEmpty bool
393 "No related objects",
395 false, false, false, false, false, false,
398 // Stored by older version, or there was a
399 // race between EmptyTrash and Put: Trash is a
400 // no-op even though the data object is very
403 t0.Add(-48 * time.Hour), none, none,
404 true, true, true, false, false, false,
407 "Not trash, but old enough to be eligible for trash",
408 t0.Add(-24 * time.Hour), t0.Add(-2 * time.Hour), none,
409 true, true, false, false, false, false,
412 "Not trash, and not old enough to be eligible for trash",
413 t0.Add(-24 * time.Hour), t0.Add(-30 * time.Minute), none,
414 true, true, true, false, false, false,
417 "Trashed + untrashed copies exist, due to recent race between Trash and Put",
418 t0.Add(-24 * time.Hour), t0.Add(-3 * time.Minute), t0.Add(-2 * time.Minute),
419 true, true, true, true, true, false,
422 "Trashed + untrashed copies exist, trash nearly eligible for deletion: prone to Trash race",
423 t0.Add(-24 * time.Hour), t0.Add(-12 * time.Hour), t0.Add(-59 * time.Minute),
424 true, false, true, true, true, false,
427 "Trashed + untrashed copies exist, trash is eligible for deletion: prone to Trash race",
428 t0.Add(-24 * time.Hour), t0.Add(-12 * time.Hour), t0.Add(-61 * time.Minute),
429 true, false, true, true, false, false,
432 "Trashed + untrashed copies exist, due to old race between Put and unfinished Trash: emptying trash is unsafe",
433 t0.Add(-24 * time.Hour), t0.Add(-12 * time.Hour), t0.Add(-12 * time.Hour),
434 true, false, true, true, true, true,
437 "Trashed + untrashed copies exist, used to be unsafe to empty, but since made safe by fixRace+Touch",
438 t0.Add(-time.Second), t0.Add(-time.Second), t0.Add(-12 * time.Hour),
439 true, true, true, true, false, false,
442 "Trashed + untrashed copies exist because Trash operation was interrupted (no race)",
443 t0.Add(-24 * time.Hour), t0.Add(-24 * time.Hour), t0.Add(-12 * time.Hour),
444 true, false, true, true, false, false,
447 "Trash, not yet eligible for deletion",
448 none, t0.Add(-12 * time.Hour), t0.Add(-time.Minute),
449 false, false, false, true, true, false,
452 "Trash, not yet eligible for deletion, prone to races",
453 none, t0.Add(-12 * time.Hour), t0.Add(-59 * time.Minute),
454 false, false, false, true, true, false,
457 "Trash, eligible for deletion",
458 none, t0.Add(-12 * time.Hour), t0.Add(-2 * time.Hour),
459 false, false, false, true, false, false,
462 "Erroneously trashed during a race, detected before BlobTrashLifetime",
463 none, t0.Add(-30 * time.Minute), t0.Add(-29 * time.Minute),
464 true, false, true, true, true, false,
467 "Erroneously trashed during a race, rescue during EmptyTrash despite reaching BlobTrashLifetime",
468 none, t0.Add(-90 * time.Minute), t0.Add(-89 * time.Minute),
469 true, false, true, true, true, false,
472 "Trashed copy exists with no recent/* marker (cause unknown); repair by untrashing",
473 none, none, t0.Add(-time.Minute),
474 false, false, false, true, true, true,
477 for _, prefixLength := range []int{0, 3} {
478 v.PrefixLength = prefixLength
479 c.Logf("Scenario: %q (prefixLength=%d)", scenario.label, prefixLength)
481 // We have a few tests to run for each scenario, and
482 // the tests are expected to change state. By calling
483 // this setup func between tests, we (re)create the
484 // scenario as specified, using a new unique block
485 // locator to prevent interference from previous
488 setupScenario := func() (string, []byte) {
490 blk := []byte(fmt.Sprintf("%d", nextKey))
491 loc := fmt.Sprintf("%x", md5.Sum(blk))
493 if prefixLength > 0 {
494 key = loc[:prefixLength] + "/" + loc
496 c.Log("\t", loc, "\t", key)
497 putS3Obj(scenario.dataT, key, blk)
498 putS3Obj(scenario.recentT, "recent/"+key, nil)
499 putS3Obj(scenario.trashT, "trash/"+key, blk)
500 v.s3fakeClock.now = &t0
505 loc, blk := setupScenario()
506 err := v.BlockRead(context.Background(), loc, brdiscard)
507 c.Check(err == nil, check.Equals, scenario.canGet, check.Commentf("err was %+v", err))
509 c.Check(os.IsNotExist(err), check.Equals, true)
512 // Call Trash, then check canTrash and canGetAfterTrash
513 loc, _ = setupScenario()
514 err = v.BlockTrash(loc)
515 c.Check(err == nil, check.Equals, scenario.canTrash)
516 err = v.BlockRead(context.Background(), loc, brdiscard)
517 c.Check(err == nil, check.Equals, scenario.canGetAfterTrash)
519 c.Check(os.IsNotExist(err), check.Equals, true)
522 // Call Untrash, then check canUntrash
523 loc, _ = setupScenario()
524 err = v.BlockUntrash(loc)
525 c.Check(err == nil, check.Equals, scenario.canUntrash)
526 if scenario.dataT != none || scenario.trashT != none {
527 // In all scenarios where the data exists, we
528 // should be able to Get after Untrash --
529 // regardless of timestamps, errors, race
531 err = v.BlockRead(context.Background(), loc, brdiscard)
532 c.Check(err, check.IsNil)
535 // Call EmptyTrash, then check haveTrashAfterEmpty and
537 loc, _ = setupScenario()
539 _, err = v.head("trash/" + v.key(loc))
540 c.Check(err == nil, check.Equals, scenario.haveTrashAfterEmpty)
541 if scenario.freshAfterEmpty {
542 t, err := v.Mtime(loc)
543 c.Check(err, check.IsNil)
544 // new mtime must be current (with an
545 // allowance for 1s timestamp precision)
546 c.Check(t.After(t0.Add(-time.Second)), check.Equals, true)
549 // Check for current Mtime after Put (applies to all
551 loc, blk = setupScenario()
552 err = v.BlockWrite(context.Background(), loc, blk)
553 c.Check(err, check.IsNil)
554 t, err := v.Mtime(loc)
555 c.Check(err, check.IsNil)
556 c.Check(t.After(t0.Add(-time.Second)), check.Equals, true)
561 type testableS3Volume struct {
563 server *httptest.Server
565 s3fakeClock *s3fakeClock
568 type gofakes3logger struct {
572 func (l gofakes3logger) Print(level gofakes3.LogLevel, v ...interface{}) {
574 case gofakes3.LogErr:
576 case gofakes3.LogWarn:
578 case gofakes3.LogInfo:
581 panic("unknown level")
585 var testBucketSerial atomic.Int64
587 func (s *stubbedS3Suite) newTestableVolume(c *check.C, params newVolumeParams, raceWindow time.Duration) *testableS3Volume {
588 if params.Logger == nil {
589 params.Logger = ctxlog.TestLogger(c)
591 if s.s3server == nil {
592 backend := s3mem.New(s3mem.WithTimeSource(s.s3fakeClock))
593 logger := ctxlog.TestLogger(c)
594 faker := gofakes3.New(backend,
595 gofakes3.WithTimeSource(s.s3fakeClock),
596 gofakes3.WithLogger(gofakes3logger{FieldLogger: logger}),
597 gofakes3.WithTimeSkewLimit(0))
598 s.s3server = httptest.NewServer(faker.Server())
600 endpoint := s.s3server.URL
601 bucketName := fmt.Sprintf("testbucket%d", testBucketSerial.Add(1))
603 var metadataURL, accessKey, secretKey string
604 if s.metadata != nil {
605 metadataURL = s.metadata.URL
607 accessKey, secretKey = "xxx", "xxx"
610 v := &testableS3Volume{
612 S3VolumeDriverParameters: arvados.S3VolumeDriverParameters{
613 AccessKeyID: accessKey,
614 SecretAccessKey: secretKey,
617 Region: "test-region-1",
618 LocationConstraint: true,
623 cluster: params.Cluster,
624 volume: params.ConfigVolume,
625 logger: params.Logger,
626 metrics: params.MetricsVecs,
627 bufferPool: params.BufferPool,
630 s3fakeClock: s.s3fakeClock,
632 c.Assert(v.s3Volume.check(metadataURL), check.IsNil)
633 // Create the testbucket
634 input := &s3.CreateBucketInput{
635 Bucket: aws.String(bucketName),
637 _, err := v.s3Volume.bucket.svc.CreateBucket(context.Background(), input)
638 c.Assert(err, check.IsNil)
639 // We couldn't set RaceWindow until now because check()
640 // rejects negative values.
641 v.s3Volume.RaceWindow = arvados.Duration(raceWindow)
645 func (v *testableS3Volume) blockWriteWithoutMD5Check(loc string, block []byte) error {
647 r := newCountingReader(bytes.NewReader(block), v.bucket.stats.TickOutBytes)
649 uploader := manager.NewUploader(v.bucket.svc, func(u *manager.Uploader) {
650 u.PartSize = 5 * 1024 * 1024
654 _, err := uploader.Upload(context.Background(), &s3.PutObjectInput{
655 Bucket: aws.String(v.bucket.bucket),
656 Key: aws.String(key),
663 empty := bytes.NewReader([]byte{})
664 _, err = uploader.Upload(context.Background(), &s3.PutObjectInput{
665 Bucket: aws.String(v.bucket.bucket),
666 Key: aws.String("recent/" + key),
672 // TouchWithDate turns back the clock while doing a Touch(). We assume
673 // there are no other operations happening on the same s3test server
675 func (v *testableS3Volume) TouchWithDate(loc string, lastPut time.Time) {
676 v.s3fakeClock.now = &lastPut
678 uploader := manager.NewUploader(v.bucket.svc)
679 empty := bytes.NewReader([]byte{})
680 _, err := uploader.Upload(context.Background(), &s3.PutObjectInput{
681 Bucket: aws.String(v.bucket.bucket),
682 Key: aws.String("recent/" + v.key(loc)),
689 v.s3fakeClock.now = nil
692 func (v *testableS3Volume) Teardown() {
695 func (v *testableS3Volume) ReadWriteOperationLabelValues() (r, w string) {