7159: Fix error handling when reading full size block. refs #7159
[arvados.git] / services / keepstore / azure_blob_volume_test.go
index dc1a7e46e48f365001a27045729f040feaa21537..a4c6e62a7d5f6f02c938ae27872cb96bb6a8e3e6 100644 (file)
@@ -1,11 +1,14 @@
 package main
 
 import (
+       "bytes"
        "encoding/base64"
        "encoding/xml"
        "flag"
+       "fmt"
        "io/ioutil"
        "log"
+       "math/rand"
        "net"
        "net/http"
        "net/http/httptest"
@@ -17,13 +20,13 @@ import (
        "testing"
        "time"
 
-       "github.com/Azure/azure-sdk-for-go/storage"
+       "github.com/curoverse/azure-sdk-for-go/storage"
 )
 
 const (
        // The same fake credentials used by Microsoft's Azure emulator
        emulatorAccountName = "devstoreaccount1"
-       emulatorAccountKey = "Eby8vdM02xNOcqFlqUwJPLlmEtlCDXJ1OUzFT50uSRZ6IFsuFq2UVErCz4I6tq/K1SZFPTOtr/KBHBeksoGMGw=="
+       emulatorAccountKey  = "Eby8vdM02xNOcqFlqUwJPLlmEtlCDXJ1OUzFT50uSRZ6IFsuFq2UVErCz4I6tq/K1SZFPTOtr/KBHBeksoGMGw=="
 )
 
 var azureTestContainer string
@@ -36,15 +39,18 @@ func init() {
                "Name of Azure container to use for testing. Do not use a container with real data! Use -azure-storage-account-name and -azure-storage-key-file arguments to supply credentials.")
 }
 
-type azBlob struct{
+type azBlob struct {
        Data        []byte
+       Etag        string
+       Metadata    map[string]string
        Mtime       time.Time
        Uncommitted map[string][]byte
 }
 
 type azStubHandler struct {
        sync.Mutex
-       blobs  map[string]*azBlob
+       blobs map[string]*azBlob
+       race  chan chan struct{}
 }
 
 func newAzStubHandler() *azStubHandler {
@@ -54,7 +60,7 @@ func newAzStubHandler() *azStubHandler {
 }
 
 func (h *azStubHandler) TouchWithDate(container, hash string, t time.Time) {
-       if blob, ok := h.blobs[container + "|" + hash]; !ok {
+       if blob, ok := h.blobs[container+"|"+hash]; !ok {
                return
        } else {
                blob.Mtime = t
@@ -64,13 +70,28 @@ func (h *azStubHandler) TouchWithDate(container, hash string, t time.Time) {
 func (h *azStubHandler) PutRaw(container, hash string, data []byte) {
        h.Lock()
        defer h.Unlock()
-       h.blobs[container + "|" + hash] = &azBlob{
-               Data: data,
-               Mtime: time.Now(),
+       h.blobs[container+"|"+hash] = &azBlob{
+               Data:        data,
+               Mtime:       time.Now(),
                Uncommitted: make(map[string][]byte),
        }
 }
 
+func (h *azStubHandler) unlockAndRace() {
+       if h.race == nil {
+               return
+       }
+       h.Unlock()
+       // Signal caller that race is starting by reading from
+       // h.race. If we get a channel, block until that channel is
+       // ready to receive. If we get nil (or h.race is closed) just
+       // proceed.
+       if c := <-h.race; c != nil {
+               c <- struct{}{}
+       }
+       h.Lock()
+}
+
 func (h *azStubHandler) ServeHTTP(rw http.ResponseWriter, r *http.Request) {
        h.Lock()
        defer h.Unlock()
@@ -99,17 +120,32 @@ func (h *azStubHandler) ServeHTTP(rw http.ResponseWriter, r *http.Request) {
                Uncommitted []string
        }
 
-       blob, blobExists := h.blobs[container + "|" + hash]
+       blob, blobExists := h.blobs[container+"|"+hash]
 
        switch {
-       case r.Method == "PUT" && r.Form.Get("comp") == "" && r.Header.Get("Content-Length") == "0":
-               rw.WriteHeader(http.StatusCreated)
-               h.blobs[container + "|" + hash] = &azBlob{
-                       Data:  body,
-                       Mtime: time.Now(),
+       case r.Method == "PUT" && r.Form.Get("comp") == "":
+               // "Put Blob" API
+               if _, ok := h.blobs[container+"|"+hash]; !ok {
+                       // Like the real Azure service, we offer a
+                       // race window during which other clients can
+                       // list/get the new blob before any data is
+                       // committed.
+                       h.blobs[container+"|"+hash] = &azBlob{
+                               Mtime:       time.Now(),
+                               Uncommitted: make(map[string][]byte),
+                               Etag:        makeEtag(),
+                       }
+                       h.unlockAndRace()
+               }
+               h.blobs[container+"|"+hash] = &azBlob{
+                       Data:        body,
+                       Mtime:       time.Now(),
                        Uncommitted: make(map[string][]byte),
+                       Etag:        makeEtag(),
                }
+               rw.WriteHeader(http.StatusCreated)
        case r.Method == "PUT" && r.Form.Get("comp") == "block":
+               // "Put Block" API
                if !blobExists {
                        log.Printf("Got block for nonexistent blob: %+v", r)
                        rw.WriteHeader(http.StatusBadRequest)
@@ -124,6 +160,7 @@ func (h *azStubHandler) ServeHTTP(rw http.ResponseWriter, r *http.Request) {
                blob.Uncommitted[string(blockID)] = body
                rw.WriteHeader(http.StatusCreated)
        case r.Method == "PUT" && r.Form.Get("comp") == "blocklist":
+               // "Put Block List" API
                bl := &blockListRequestBody{}
                if err := xml.Unmarshal(body, bl); err != nil {
                        log.Printf("xml Unmarshal: %s", err)
@@ -138,10 +175,31 @@ func (h *azStubHandler) ServeHTTP(rw http.ResponseWriter, r *http.Request) {
                                return
                        }
                        blob.Data = blob.Uncommitted[string(blockID)]
-                       log.Printf("body %+q, bl %+v, blockID %+q, data %+q", body, bl, blockID, blob.Data)
+                       blob.Etag = makeEtag()
+                       blob.Mtime = time.Now()
+                       delete(blob.Uncommitted, string(blockID))
                }
                rw.WriteHeader(http.StatusCreated)
+       case r.Method == "PUT" && r.Form.Get("comp") == "metadata":
+               // "Set Metadata Headers" API. We don't bother
+               // stubbing "Get Metadata Headers": AzureBlobVolume
+               // sets metadata headers only as a way to bump Etag
+               // and Last-Modified.
+               if !blobExists {
+                       log.Printf("Got metadata for nonexistent blob: %+v", r)
+                       rw.WriteHeader(http.StatusBadRequest)
+                       return
+               }
+               blob.Metadata = make(map[string]string)
+               for k, v := range r.Header {
+                       if strings.HasPrefix(strings.ToLower(k), "x-ms-meta-") {
+                               blob.Metadata[k] = v[0]
+                       }
+               }
+               blob.Mtime = time.Now()
+               blob.Etag = makeEtag()
        case (r.Method == "GET" || r.Method == "HEAD") && hash != "":
+               // "Get Blob" API
                if !blobExists {
                        rw.WriteHeader(http.StatusNotFound)
                        return
@@ -153,14 +211,17 @@ func (h *azStubHandler) ServeHTTP(rw http.ResponseWriter, r *http.Request) {
                                log.Printf("write %+q: %s", blob.Data, err)
                        }
                }
+               h.unlockAndRace()
        case r.Method == "DELETE" && hash != "":
+               // "Delete Blob" API
                if !blobExists {
                        rw.WriteHeader(http.StatusNotFound)
                        return
                }
-               delete(h.blobs, container + "|" + hash)
+               delete(h.blobs, container+"|"+hash)
                rw.WriteHeader(http.StatusAccepted)
        case r.Method == "GET" && r.Form.Get("comp") == "list" && r.Form.Get("restype") == "container":
+               // "List Blobs" API
                prefix := container + "|" + r.Form.Get("prefix")
                marker := r.Form.Get("marker")
 
@@ -170,7 +231,7 @@ func (h *azStubHandler) ServeHTTP(rw http.ResponseWriter, r *http.Request) {
                }
 
                resp := storage.BlobListResponse{
-                       Marker: marker,
+                       Marker:     marker,
                        NextMarker: "",
                        MaxResults: int64(maxResults),
                }
@@ -187,12 +248,13 @@ func (h *azStubHandler) ServeHTTP(rw http.ResponseWriter, r *http.Request) {
                                break
                        }
                        if len(resp.Blobs) > 0 || marker == "" || marker == hash {
-                               blob := h.blobs[container + "|" + hash]
+                               blob := h.blobs[container+"|"+hash]
                                resp.Blobs = append(resp.Blobs, storage.Blob{
                                        Name: hash,
                                        Properties: storage.BlobProperties{
-                                               LastModified: blob.Mtime.Format(time.RFC1123),
+                                               LastModified:  blob.Mtime.Format(time.RFC1123),
                                                ContentLength: int64(len(blob.Data)),
+                                               Etag:          blob.Etag,
                                        },
                                })
                        }
@@ -217,6 +279,7 @@ type azStubDialer struct {
 }
 
 var localHostPortRe = regexp.MustCompile(`(127\.0\.0\.1|localhost|\[::1\]):\d+`)
+
 func (d *azStubDialer) Dial(network, address string) (net.Conn, error) {
        if hp := localHostPortRe.FindString(address); hp != "" {
                log.Println("azStubDialer: dial", hp, "instead of", address)
@@ -232,7 +295,7 @@ type TestableAzureBlobVolume struct {
        t         *testing.T
 }
 
-func NewTestableAzureBlobVolume(t *testing.T, readonly bool, replication int) TestableVolume {
+func NewTestableAzureBlobVolume(t *testing.T, readonly bool, replication int) *TestableAzureBlobVolume {
        azHandler := newAzStubHandler()
        azStub := httptest.NewServer(azHandler)
 
@@ -263,9 +326,9 @@ func NewTestableAzureBlobVolume(t *testing.T, readonly bool, replication int) Te
 
        return &TestableAzureBlobVolume{
                AzureBlobVolume: v,
-               azHandler: azHandler,
-               azStub: azStub,
-               t: t,
+               azHandler:       azHandler,
+               azStub:          azStub,
+               t:               t,
        }
 }
 
@@ -276,6 +339,8 @@ func TestAzureBlobVolumeWithGeneric(t *testing.T) {
        http.DefaultTransport = &http.Transport{
                Dial: (&azStubDialer{}).Dial,
        }
+       azureWriteRaceInterval = time.Millisecond
+       azureWriteRacePollTime = time.Nanosecond
        DoGenericVolumeTests(t, func(t *testing.T) TestableVolume {
                return NewTestableAzureBlobVolume(t, false, azureStorageReplication)
        })
@@ -288,6 +353,8 @@ func TestReadonlyAzureBlobVolumeWithGeneric(t *testing.T) {
        http.DefaultTransport = &http.Transport{
                Dial: (&azStubDialer{}).Dial,
        }
+       azureWriteRaceInterval = time.Millisecond
+       azureWriteRacePollTime = time.Nanosecond
        DoGenericVolumeTests(t, func(t *testing.T) TestableVolume {
                return NewTestableAzureBlobVolume(t, true, azureStorageReplication)
        })
@@ -303,6 +370,99 @@ func TestAzureBlobVolumeReplication(t *testing.T) {
        }
 }
 
+func TestAzureBlobVolumeCreateBlobRace(t *testing.T) {
+       defer func(t http.RoundTripper) {
+               http.DefaultTransport = t
+       }(http.DefaultTransport)
+       http.DefaultTransport = &http.Transport{
+               Dial: (&azStubDialer{}).Dial,
+       }
+
+       v := NewTestableAzureBlobVolume(t, false, 3)
+       defer v.Teardown()
+
+       azureWriteRaceInterval = time.Second
+       azureWriteRacePollTime = time.Millisecond
+
+       allDone := make(chan struct{})
+       v.azHandler.race = make(chan chan struct{})
+       go func() {
+               err := v.Put(TestHash, TestBlock)
+               if err != nil {
+                       t.Error(err)
+               }
+       }()
+       continuePut := make(chan struct{})
+       // Wait for the stub's Put to create the empty blob
+       v.azHandler.race <- continuePut
+       go func() {
+               buf, err := v.Get(TestHash)
+               if err != nil {
+                       t.Error(err)
+               } else {
+                       bufs.Put(buf)
+               }
+               close(allDone)
+       }()
+       // Wait for the stub's Get to get the empty blob
+       close(v.azHandler.race)
+       // Allow stub's Put to continue, so the real data is ready
+       // when the volume's Get retries
+       <-continuePut
+       // Wait for volume's Get to return the real data
+       <-allDone
+}
+
+func TestAzureBlobVolumeCreateBlobRaceDeadline(t *testing.T) {
+       defer func(t http.RoundTripper) {
+               http.DefaultTransport = t
+       }(http.DefaultTransport)
+       http.DefaultTransport = &http.Transport{
+               Dial: (&azStubDialer{}).Dial,
+       }
+
+       v := NewTestableAzureBlobVolume(t, false, 3)
+       defer v.Teardown()
+
+       azureWriteRaceInterval = 2 * time.Second
+       azureWriteRacePollTime = 5 * time.Millisecond
+
+       v.PutRaw(TestHash, []byte{})
+
+       buf := new(bytes.Buffer)
+       v.IndexTo("", buf)
+       if buf.Len() != 0 {
+               t.Errorf("Index %+q should be empty", buf.Bytes())
+       }
+
+       v.TouchWithDate(TestHash, time.Now().Add(-1982 * time.Millisecond))
+
+       allDone := make(chan struct{})
+       go func() {
+               defer close(allDone)
+               buf, err := v.Get(TestHash)
+               if err != nil {
+                       t.Error(err)
+                       return
+               }
+               if len(buf) != 0 {
+                       t.Errorf("Got %+q, expected empty buf", buf)
+               }
+               bufs.Put(buf)
+       }()
+       select {
+       case <-allDone:
+       case <-time.After(time.Second):
+               t.Error("Get should have stopped waiting for race when block was 2s old")
+       }
+
+       buf.Reset()
+       v.IndexTo("", buf)
+       if !bytes.HasPrefix(buf.Bytes(), []byte(TestHash+"+0")) {
+               t.Errorf("Index %+q should have %+q", buf.Bytes(), TestHash+"+0")
+       }
+}
+
 func (v *TestableAzureBlobVolume) PutRaw(locator string, data []byte) {
        v.azHandler.PutRaw(v.containerName, locator, data)
 }
@@ -314,3 +474,7 @@ func (v *TestableAzureBlobVolume) TouchWithDate(locator string, lastPut time.Tim
 func (v *TestableAzureBlobVolume) Teardown() {
        v.azStub.Close()
 }
+
+func makeEtag() string {
+       return fmt.Sprintf("0x%x", rand.Int63())
+}