17 "github.com/curoverse/azure-sdk-for-go/storage"
22 azureStorageAccountName string
23 azureStorageAccountKeyFile string
24 azureStorageReplication int
25 azureWriteRaceInterval = 15 * time.Second
26 azureWriteRacePollTime = time.Second
29 func readKeyFromFile(file string) (string, error) {
30 buf, err := ioutil.ReadFile(file)
32 return "", errors.New("reading key from " + file + ": " + err.Error())
34 accountKey := strings.TrimSpace(string(buf))
36 return "", errors.New("empty account key in " + file)
38 return accountKey, nil
41 type azureVolumeAdder struct {
45 func (s *azureVolumeAdder) Set(containerName string) error {
46 if containerName == "" {
47 return errors.New("no container name given")
49 if azureStorageAccountName == "" || azureStorageAccountKeyFile == "" {
50 return errors.New("-azure-storage-account-name and -azure-storage-account-key-file arguments must given before -azure-storage-container-volume")
52 accountKey, err := readKeyFromFile(azureStorageAccountKeyFile)
56 azClient, err := storage.NewBasicClient(azureStorageAccountName, accountKey)
58 return errors.New("creating Azure storage client: " + err.Error())
61 log.Print("Notice: -serialize is not supported by azure-blob-container volumes.")
63 v := NewAzureBlobVolume(azClient, containerName, flagReadonly, azureStorageReplication)
64 if err := v.Check(); err != nil {
67 *s.volumeSet = append(*s.volumeSet, v)
72 flag.Var(&azureVolumeAdder{&volumes},
73 "azure-storage-container-volume",
74 "Use the given container as a storage volume. Can be given multiple times.")
76 &azureStorageAccountName,
77 "azure-storage-account-name",
79 "Azure storage account name used for subsequent --azure-storage-container-volume arguments.")
81 &azureStorageAccountKeyFile,
82 "azure-storage-account-key-file",
84 "File containing the account key used for subsequent --azure-storage-container-volume arguments.")
86 &azureStorageReplication,
87 "azure-storage-replication",
89 "Replication level to report to clients when data is stored in an Azure container.")
92 "azure-max-get-bytes",
94 fmt.Sprintf("Maximum bytes to request in a single GET request. If smaller than %d, use multiple concurrent range requests to retrieve a block.", BlockSize))
97 // An AzureBlobVolume stores and retrieves blocks in an Azure Blob
99 type AzureBlobVolume struct {
100 azClient storage.Client
101 bsClient storage.BlobStorageClient
107 // NewAzureBlobVolume returns a new AzureBlobVolume using the given
108 // client and container name. The replication argument specifies the
109 // replication level to report when writing data.
110 func NewAzureBlobVolume(client storage.Client, containerName string, readonly bool, replication int) *AzureBlobVolume {
111 return &AzureBlobVolume{
113 bsClient: client.GetBlobService(),
114 containerName: containerName,
116 replication: replication,
120 // Check returns nil if the volume is usable.
121 func (v *AzureBlobVolume) Check() error {
122 ok, err := v.bsClient.ContainerExists(v.containerName)
127 return errors.New("container does not exist")
132 // Get reads a Keep block that has been stored as a block blob in the
135 // If the block is younger than azureWriteRaceInterval and is
136 // unexpectedly empty, assume a PutBlob operation is in progress, and
137 // wait for it to finish writing.
138 func (v *AzureBlobVolume) Get(loc string) ([]byte, error) {
139 var deadline time.Time
140 haveDeadline := false
141 buf, err := v.get(loc)
142 for err == nil && len(buf) == 0 && loc != "d41d8cd98f00b204e9800998ecf8427e" {
143 // Seeing a brand new empty block probably means we're
144 // in a race with CreateBlob, which under the hood
145 // (apparently) does "CreateEmpty" and "CommitData"
146 // with no additional transaction locking.
148 t, err := v.Mtime(loc)
150 log.Print("Got empty block (possible race) but Mtime failed: ", err)
153 deadline = t.Add(azureWriteRaceInterval)
154 if time.Now().After(deadline) {
157 log.Printf("Race? Block %s is 0 bytes, %s old. Polling until %s", loc, time.Since(t), deadline)
159 } else if time.Now().After(deadline) {
163 time.Sleep(azureWriteRacePollTime)
164 buf, err = v.get(loc)
167 log.Printf("Race ended with len(buf)==%d", len(buf))
172 func (v *AzureBlobVolume) get(loc string) ([]byte, error) {
173 expectSize := BlockSize
174 if azureMaxGetBytes < BlockSize {
175 // Unfortunately the handler doesn't tell us how long the blob
176 // is expected to be, so we have to ask Azure.
177 props, err := v.bsClient.GetBlobProperties(v.containerName, loc)
179 return nil, v.translateError(err)
181 if props.ContentLength > int64(BlockSize) || props.ContentLength < 0 {
182 return nil, fmt.Errorf("block %s invalid size %d (max %d)", loc, props.ContentLength, BlockSize)
184 expectSize = int(props.ContentLength)
187 buf := bufs.Get(expectSize)
192 // We'll update this actualSize if/when we get the last piece.
194 pieces := (expectSize + azureMaxGetBytes - 1) / azureMaxGetBytes
195 errors := make([]error, pieces)
196 var wg sync.WaitGroup
198 for p := 0; p < pieces; p++ {
201 startPos := p * azureMaxGetBytes
202 endPos := startPos + azureMaxGetBytes
203 if endPos > expectSize {
206 var rdr io.ReadCloser
208 if startPos == 0 && endPos == expectSize {
209 rdr, err = v.bsClient.GetBlob(v.containerName, loc)
211 rdr, err = v.bsClient.GetBlobRange(v.containerName, loc, fmt.Sprintf("%d-%d", startPos, endPos-1))
218 n, err := io.ReadFull(rdr, buf[startPos:endPos])
219 if pieces == 1 && (err == io.ErrUnexpectedEOF || err == io.EOF) {
220 // If we don't know the actual size,
221 // and just tried reading 64 MiB, it's
222 // normal to encounter EOF.
223 } else if err != nil {
227 actualSize = startPos + n
232 for _, err := range errors {
235 return nil, v.translateError(err)
238 return buf[:actualSize], nil
241 // Compare the given data with existing stored data.
242 func (v *AzureBlobVolume) Compare(loc string, expect []byte) error {
243 rdr, err := v.bsClient.GetBlob(v.containerName, loc)
245 return v.translateError(err)
248 return compareReaderWithBuf(rdr, expect, loc[:32])
251 // Put stores a Keep block as a block blob in the container.
252 func (v *AzureBlobVolume) Put(loc string, block []byte) error {
254 return MethodDisabledError
256 return v.bsClient.CreateBlockBlobFromReader(v.containerName, loc, uint64(len(block)), bytes.NewReader(block))
259 // Touch updates the last-modified property of a block blob.
260 func (v *AzureBlobVolume) Touch(loc string) error {
262 return MethodDisabledError
264 return v.bsClient.SetBlobMetadata(v.containerName, loc, map[string]string{
265 "touch": fmt.Sprintf("%d", time.Now()),
269 // Mtime returns the last-modified property of a block blob.
270 func (v *AzureBlobVolume) Mtime(loc string) (time.Time, error) {
271 props, err := v.bsClient.GetBlobProperties(v.containerName, loc)
273 return time.Time{}, err
275 return time.Parse(time.RFC1123, props.LastModified)
278 // IndexTo writes a list of Keep blocks that are stored in the
280 func (v *AzureBlobVolume) IndexTo(prefix string, writer io.Writer) error {
281 params := storage.ListBlobsParameters{
285 resp, err := v.bsClient.ListBlobs(v.containerName, params)
289 for _, b := range resp.Blobs {
290 t, err := time.Parse(time.RFC1123, b.Properties.LastModified)
294 if !v.isKeepBlock(b.Name) {
297 if b.Properties.ContentLength == 0 && t.Add(azureWriteRaceInterval).After(time.Now()) {
298 // A new zero-length blob is probably
299 // just a new non-empty blob that
300 // hasn't committed its data yet (see
301 // Get()), and in any case has no
305 fmt.Fprintf(writer, "%s+%d %d\n", b.Name, b.Properties.ContentLength, t.Unix())
307 if resp.NextMarker == "" {
310 params.Marker = resp.NextMarker
314 // Trash a Keep block.
315 func (v *AzureBlobVolume) Trash(loc string) error {
317 return MethodDisabledError
319 // Ideally we would use If-Unmodified-Since, but that
320 // particular condition seems to be ignored by Azure. Instead,
321 // we get the Etag before checking Mtime, and use If-Match to
322 // ensure we don't delete data if Put() or Touch() happens
323 // between our calls to Mtime() and DeleteBlob().
324 props, err := v.bsClient.GetBlobProperties(v.containerName, loc)
328 if t, err := v.Mtime(loc); err != nil {
330 } else if time.Since(t) < blobSignatureTTL {
333 return v.bsClient.DeleteBlob(v.containerName, loc, map[string]string{
334 "If-Match": props.Etag,
338 // Untrash a Keep block.
340 func (v *AzureBlobVolume) Untrash(loc string) error {
344 // Status returns a VolumeStatus struct with placeholder data.
345 func (v *AzureBlobVolume) Status() *VolumeStatus {
346 return &VolumeStatus{
348 BytesFree: BlockSize * 1000,
353 // String returns a volume label, including the container name.
354 func (v *AzureBlobVolume) String() string {
355 return fmt.Sprintf("azure-storage-container:%+q", v.containerName)
358 // Writable returns true, unless the -readonly flag was on when the
360 func (v *AzureBlobVolume) Writable() bool {
364 // Replication returns the replication level of the container, as
365 // specified by the -azure-storage-replication argument.
366 func (v *AzureBlobVolume) Replication() int {
370 // If possible, translate an Azure SDK error to a recognizable error
371 // like os.ErrNotExist.
372 func (v *AzureBlobVolume) translateError(err error) error {
376 case strings.Contains(err.Error(), "404 Not Found"):
377 // "storage: service returned without a response body (404 Not Found)"
378 return os.ErrNotExist
384 var keepBlockRegexp = regexp.MustCompile(`^[0-9a-f]{32}$`)
386 func (v *AzureBlobVolume) isKeepBlock(s string) bool {
387 return keepBlockRegexp.MatchString(s)