10467: Merge branch 'master' into 10467-client-disconnect
[arvados.git] / services / keepstore / s3_volume_test.go
index a40b84f88d23a970017f038df20429a02390a6de..63b186220c30a562e900722d63d38be50bde05d6 100644 (file)
@@ -2,25 +2,20 @@ package main
 
 import (
        "bytes"
+       "context"
        "crypto/md5"
        "fmt"
+       "io/ioutil"
        "log"
        "os"
        "time"
 
-       "github.com/AdRoll/goamz/aws"
+       "git.curoverse.com/arvados.git/sdk/go/arvados"
        "github.com/AdRoll/goamz/s3"
        "github.com/AdRoll/goamz/s3/s3test"
        check "gopkg.in/check.v1"
 )
 
-type TestableS3Volume struct {
-       *S3Volume
-       server      *s3test.Server
-       c           *check.C
-       serverClock *fakeClock
-}
-
 const (
        TestBucketName = "testbucket"
 )
@@ -42,30 +37,6 @@ func init() {
        s3UnsafeDelete = true
 }
 
-func NewTestableS3Volume(c *check.C, raceWindow time.Duration, readonly bool, replication int) *TestableS3Volume {
-       clock := &fakeClock{}
-       srv, err := s3test.NewServer(&s3test.Config{Clock: clock})
-       c.Assert(err, check.IsNil)
-       auth := aws.Auth{}
-       region := aws.Region{
-               Name:                 "test-region-1",
-               S3Endpoint:           srv.URL(),
-               S3LocationConstraint: true,
-       }
-       bucket := &s3.Bucket{
-               S3:   s3.New(auth, region),
-               Name: TestBucketName,
-       }
-       err = bucket.PutBucket(s3.ACL("private"))
-       c.Assert(err, check.IsNil)
-
-       return &TestableS3Volume{
-               S3Volume:    NewS3Volume(auth, region, TestBucketName, raceWindow, readonly, replication),
-               server:      srv,
-               serverClock: clock,
-       }
-}
-
 var _ = check.Suite(&StubbedS3Suite{})
 
 type StubbedS3Suite struct {
@@ -76,19 +47,19 @@ func (s *StubbedS3Suite) TestGeneric(c *check.C) {
        DoGenericVolumeTests(c, func(t TB) TestableVolume {
                // Use a negative raceWindow so s3test's 1-second
                // timestamp precision doesn't confuse fixRace.
-               return NewTestableS3Volume(c, -2*time.Second, false, 2)
+               return s.newTestableVolume(c, -2*time.Second, false, 2)
        })
 }
 
 func (s *StubbedS3Suite) TestGenericReadOnly(c *check.C) {
        DoGenericVolumeTests(c, func(t TB) TestableVolume {
-               return NewTestableS3Volume(c, -2*time.Second, true, 2)
+               return s.newTestableVolume(c, -2*time.Second, true, 2)
        })
 }
 
 func (s *StubbedS3Suite) TestIndex(c *check.C) {
-       v := NewTestableS3Volume(c, 0, false, 2)
-       v.indexPageSize = 3
+       v := s.newTestableVolume(c, 0, false, 2)
+       v.IndexPageSize = 3
        for i := 0; i < 256; i++ {
                v.PutRaw(fmt.Sprintf("%02x%030x", i, i), []byte{102, 111, 111})
        }
@@ -112,41 +83,43 @@ func (s *StubbedS3Suite) TestIndex(c *check.C) {
 }
 
 func (s *StubbedS3Suite) TestBackendStates(c *check.C) {
-       defer func(tl, bs time.Duration) {
-               trashLifetime = tl
-               blobSignatureTTL = bs
-       }(trashLifetime, blobSignatureTTL)
-       trashLifetime = time.Hour
-       blobSignatureTTL = time.Hour
+       defer func(tl, bs arvados.Duration) {
+               theConfig.TrashLifetime = tl
+               theConfig.BlobSignatureTTL = bs
+       }(theConfig.TrashLifetime, theConfig.BlobSignatureTTL)
+       theConfig.TrashLifetime.Set("1h")
+       theConfig.BlobSignatureTTL.Set("1h")
 
-       v := NewTestableS3Volume(c, 5*time.Minute, false, 2)
+       v := s.newTestableVolume(c, 5*time.Minute, false, 2)
        var none time.Time
 
-       stubKey := func(t time.Time, key string, data []byte) {
+       putS3Obj := func(t time.Time, key string, data []byte) {
                if t == none {
                        return
                }
                v.serverClock.now = &t
-               v.Bucket.Put(key, data, "application/octet-stream", s3ACL, s3.Options{})
+               v.bucket.Put(key, data, "application/octet-stream", s3ACL, s3.Options{})
        }
 
        t0 := time.Now()
        nextKey := 0
-       for _, test := range []struct {
+       for _, scenario := range []struct {
                label               string
-               data                time.Time
-               recent              time.Time
-               trash               time.Time
+               dataT               time.Time
+               recentT             time.Time
+               trashT              time.Time
                canGet              bool
                canTrash            bool
                canGetAfterTrash    bool
                canUntrash          bool
                haveTrashAfterEmpty bool
+               freshAfterEmpty     bool
        }{
                {
                        "No related objects",
                        none, none, none,
-                       false, false, false, false, false},
+                       false, false, false, false, false, false,
+               },
                {
                        // Stored by older version, or there was a
                        // race between EmptyTrash and Put: Trash is a
@@ -154,103 +127,202 @@ func (s *StubbedS3Suite) TestBackendStates(c *check.C) {
                        // old
                        "No recent/X",
                        t0.Add(-48 * time.Hour), none, none,
-                       true, true, true, false, false},
+                       true, true, true, false, false, false,
+               },
                {
-                       "Not trash; old enough to trash",
+                       "Not trash, but old enough to be eligible for trash",
                        t0.Add(-24 * time.Hour), t0.Add(-2 * time.Hour), none,
-                       true, true, false, false, false},
+                       true, true, false, false, false, false,
+               },
                {
-                       "Not trash; not old enough to trash",
+                       "Not trash, and not old enough to be eligible for trash",
                        t0.Add(-24 * time.Hour), t0.Add(-30 * time.Minute), none,
-                       true, true, true, false, false},
+                       true, true, true, false, false, false,
+               },
                {
-                       "Trash + not-trash: recent race between Trash and Put",
+                       "Trashed + untrashed copies exist, due to recent race between Trash and Put",
                        t0.Add(-24 * time.Hour), t0.Add(-3 * time.Minute), t0.Add(-2 * time.Minute),
-                       true, true, true, true, true},
+                       true, true, true, true, true, false,
+               },
                {
-                       "Trash + not-trash, nearly eligible for deletion, prone to Trash race",
+                       "Trashed + untrashed copies exist, trash nearly eligible for deletion: prone to Trash race",
                        t0.Add(-24 * time.Hour), t0.Add(-12 * time.Hour), t0.Add(-59 * time.Minute),
-                       true, false, true, true, true},
+                       true, false, true, true, true, false,
+               },
                {
-                       "Trash + not-trash, eligible for deletion, prone to Trash race",
+                       "Trashed + untrashed copies exist, trash is eligible for deletion: prone to Trash race",
                        t0.Add(-24 * time.Hour), t0.Add(-12 * time.Hour), t0.Add(-61 * time.Minute),
-                       true, false, true, true, false},
-               // FIXME: old trash never gets deleted!
-               // {
-               //      "Not trash; old race between Trash and Put, or incomplete Trash",
-               //      t0.Add(-24 * time.Hour), t0.Add(-12 * time.Hour), t0.Add(-12 * time.Hour),
-               //      true, false, true, true, false},
+                       true, false, true, true, false, false,
+               },
+               {
+                       "Trashed + untrashed copies exist, due to old race between Put and unfinished Trash: emptying trash is unsafe",
+                       t0.Add(-24 * time.Hour), t0.Add(-12 * time.Hour), t0.Add(-12 * time.Hour),
+                       true, false, true, true, true, true,
+               },
+               {
+                       "Trashed + untrashed copies exist, used to be unsafe to empty, but since made safe by fixRace+Touch",
+                       t0.Add(-time.Second), t0.Add(-time.Second), t0.Add(-12 * time.Hour),
+                       true, true, true, true, false, false,
+               },
                {
-                       "Trash operation was interrupted",
+                       "Trashed + untrashed copies exist because Trash operation was interrupted (no race)",
                        t0.Add(-24 * time.Hour), t0.Add(-24 * time.Hour), t0.Add(-12 * time.Hour),
-                       true, false, true, true, false},
+                       true, false, true, true, false, false,
+               },
                {
                        "Trash, not yet eligible for deletion",
                        none, t0.Add(-12 * time.Hour), t0.Add(-time.Minute),
-                       false, false, false, true, true},
+                       false, false, false, true, true, false,
+               },
                {
                        "Trash, not yet eligible for deletion, prone to races",
                        none, t0.Add(-12 * time.Hour), t0.Add(-59 * time.Minute),
-                       false, false, false, true, true},
+                       false, false, false, true, true, false,
+               },
                {
                        "Trash, eligible for deletion",
                        none, t0.Add(-12 * time.Hour), t0.Add(-2 * time.Hour),
-                       false, false, false, true, false},
+                       false, false, false, true, false, false,
+               },
                {
-                       "Erroneously trashed during a race, detected before trashLifetime",
+                       "Erroneously trashed during a race, detected before TrashLifetime",
                        none, t0.Add(-30 * time.Minute), t0.Add(-29 * time.Minute),
-                       true, false, true, true, true},
+                       true, false, true, true, true, false,
+               },
                {
-                       "Erroneously trashed during a race, rescue during EmptyTrash despite reaching trashLifetime",
+                       "Erroneously trashed during a race, rescue during EmptyTrash despite reaching TrashLifetime",
                        none, t0.Add(-90 * time.Minute), t0.Add(-89 * time.Minute),
-                       true, false, true, true, true},
+                       true, false, true, true, true, false,
+               },
+               {
+                       "Trashed copy exists with no recent/* marker (cause unknown); repair by untrashing",
+                       none, none, t0.Add(-time.Minute),
+                       false, false, false, true, true, true,
+               },
        } {
-               c.Log("Scenario: ", test.label)
-               var loc string
-               var blk []byte
+               c.Log("Scenario: ", scenario.label)
+
+               // We have a few tests to run for each scenario, and
+               // the tests are expected to change state. By calling
+               // this setup func between tests, we (re)create the
+               // scenario as specified, using a new unique block
+               // locator to prevent interference from previous
+               // tests.
 
-               setup := func() {
+               setupScenario := func() (string, []byte) {
                        nextKey++
-                       blk = []byte(fmt.Sprintf("%d", nextKey))
-                       loc = fmt.Sprintf("%x", md5.Sum(blk))
+                       blk := []byte(fmt.Sprintf("%d", nextKey))
+                       loc := fmt.Sprintf("%x", md5.Sum(blk))
                        c.Log("\t", loc)
-                       stubKey(test.data, loc, blk)
-                       stubKey(test.recent, "recent/"+loc, nil)
-                       stubKey(test.trash, "trash/"+loc, blk)
+                       putS3Obj(scenario.dataT, loc, blk)
+                       putS3Obj(scenario.recentT, "recent/"+loc, nil)
+                       putS3Obj(scenario.trashT, "trash/"+loc, blk)
                        v.serverClock.now = &t0
+                       return loc, blk
                }
 
-               setup()
+               // Check canGet
+               loc, blk := setupScenario()
                buf := make([]byte, len(blk))
-               _, err := v.Get(loc, buf)
-               c.Check(err == nil, check.Equals, test.canGet)
+               _, err := v.Get(context.Background(), loc, buf)
+               c.Check(err == nil, check.Equals, scenario.canGet)
                if err != nil {
                        c.Check(os.IsNotExist(err), check.Equals, true)
                }
 
-               setup()
+               // Call Trash, then check canTrash and canGetAfterTrash
+               loc, blk = setupScenario()
                err = v.Trash(loc)
-               c.Check(err == nil, check.Equals, test.canTrash)
-               _, err = v.Get(loc, buf)
-               c.Check(err == nil, check.Equals, test.canGetAfterTrash)
+               c.Check(err == nil, check.Equals, scenario.canTrash)
+               _, err = v.Get(context.Background(), loc, buf)
+               c.Check(err == nil, check.Equals, scenario.canGetAfterTrash)
                if err != nil {
                        c.Check(os.IsNotExist(err), check.Equals, true)
                }
 
-               setup()
+               // Call Untrash, then check canUntrash
+               loc, blk = setupScenario()
                err = v.Untrash(loc)
-               c.Check(err == nil, check.Equals, test.canUntrash)
+               c.Check(err == nil, check.Equals, scenario.canUntrash)
+               if scenario.dataT != none || scenario.trashT != none {
+                       // In all scenarios where the data exists, we
+                       // should be able to Get after Untrash --
+                       // regardless of timestamps, errors, race
+                       // conditions, etc.
+                       _, err = v.Get(context.Background(), loc, buf)
+                       c.Check(err, check.IsNil)
+               }
 
-               setup()
+               // Call EmptyTrash, then check haveTrashAfterEmpty and
+               // freshAfterEmpty
+               loc, blk = setupScenario()
                v.EmptyTrash()
-               _, err = v.Bucket.Head("trash/"+loc, nil)
-               c.Check(err == nil, check.Equals, test.haveTrashAfterEmpty)
+               _, err = v.bucket.Head("trash/"+loc, nil)
+               c.Check(err == nil, check.Equals, scenario.haveTrashAfterEmpty)
+               if scenario.freshAfterEmpty {
+                       t, err := v.Mtime(loc)
+                       c.Check(err, check.IsNil)
+                       // new mtime must be current (with an
+                       // allowance for 1s timestamp precision)
+                       c.Check(t.After(t0.Add(-time.Second)), check.Equals, true)
+               }
+
+               // Check for current Mtime after Put (applies to all
+               // scenarios)
+               loc, blk = setupScenario()
+               err = v.Put(context.Background(), loc, blk)
+               c.Check(err, check.IsNil)
+               t, err := v.Mtime(loc)
+               c.Check(err, check.IsNil)
+               c.Check(t.After(t0.Add(-time.Second)), check.Equals, true)
+       }
+}
+
+type TestableS3Volume struct {
+       *S3Volume
+       server      *s3test.Server
+       c           *check.C
+       serverClock *fakeClock
+}
+
+func (s *StubbedS3Suite) newTestableVolume(c *check.C, raceWindow time.Duration, readonly bool, replication int) *TestableS3Volume {
+       clock := &fakeClock{}
+       srv, err := s3test.NewServer(&s3test.Config{Clock: clock})
+       c.Assert(err, check.IsNil)
+
+       tmp, err := ioutil.TempFile("", "keepstore")
+       c.Assert(err, check.IsNil)
+       defer os.Remove(tmp.Name())
+       _, err = tmp.Write([]byte("xxx\n"))
+       c.Assert(err, check.IsNil)
+       c.Assert(tmp.Close(), check.IsNil)
+
+       v := &TestableS3Volume{
+               S3Volume: &S3Volume{
+                       Bucket:             TestBucketName,
+                       AccessKeyFile:      tmp.Name(),
+                       SecretKeyFile:      tmp.Name(),
+                       Endpoint:           srv.URL(),
+                       Region:             "test-region-1",
+                       LocationConstraint: true,
+                       RaceWindow:         arvados.Duration(raceWindow),
+                       S3Replication:      replication,
+                       UnsafeDelete:       s3UnsafeDelete,
+                       ReadOnly:           readonly,
+                       IndexPageSize:      1000,
+               },
+               server:      srv,
+               serverClock: clock,
        }
+       c.Assert(v.Start(), check.IsNil)
+       err = v.bucket.PutBucket(s3.ACL("private"))
+       c.Assert(err, check.IsNil)
+       return v
 }
 
 // PutRaw skips the ContentMD5 test
 func (v *TestableS3Volume) PutRaw(loc string, block []byte) {
-       err := v.Bucket.Put(loc, block, "application/octet-stream", s3ACL, s3.Options{})
+       err := v.bucket.Put(loc, block, "application/octet-stream", s3ACL, s3.Options{})
        if err != nil {
                log.Printf("PutRaw: %+v", err)
        }
@@ -261,7 +333,7 @@ func (v *TestableS3Volume) PutRaw(loc string, block []byte) {
 // while we do this.
 func (v *TestableS3Volume) TouchWithDate(locator string, lastPut time.Time) {
        v.serverClock.now = &lastPut
-       err := v.Bucket.Put("recent/"+locator, nil, "application/octet-stream", s3ACL, s3.Options{})
+       err := v.bucket.Put("recent/"+locator, nil, "application/octet-stream", s3ACL, s3.Options{})
        if err != nil {
                panic(err)
        }