import (
"bytes"
+ "context"
"errors"
"flag"
"fmt"
"io"
"io/ioutil"
- "log"
+ "net/http"
"os"
"regexp"
+ "strconv"
"strings"
"sync"
"time"
+ "git.curoverse.com/arvados.git/sdk/go/arvados"
+ log "github.com/Sirupsen/logrus"
"github.com/curoverse/azure-sdk-for-go/storage"
)
+const azureDefaultRequestTimeout = arvados.Duration(10 * time.Minute)
+
var (
azureMaxGetBytes int
azureStorageAccountName string
}
type azureVolumeAdder struct {
- *volumeSet
+ *Config
}
-func (s *azureVolumeAdder) Set(containerName string) error {
- if trashLifetime <= 0 {
- log.Print("Missing required configuration parameter: trash-lifetime")
- return ErrNotImplemented
- }
+// String implements flag.Value
+func (s *azureVolumeAdder) String() string {
+ return "-"
+}
- if containerName == "" {
- return errors.New("no container name given")
- }
- if azureStorageAccountName == "" || azureStorageAccountKeyFile == "" {
- return errors.New("-azure-storage-account-name and -azure-storage-account-key-file arguments must given before -azure-storage-container-volume")
- }
- accountKey, err := readKeyFromFile(azureStorageAccountKeyFile)
- if err != nil {
- return err
- }
- azClient, err := storage.NewBasicClient(azureStorageAccountName, accountKey)
- if err != nil {
- return errors.New("creating Azure storage client: " + err.Error())
- }
- if flagSerializeIO {
- log.Print("Notice: -serialize is not supported by azure-blob-container volumes.")
- }
- v := NewAzureBlobVolume(azClient, containerName, flagReadonly, azureStorageReplication)
- if err := v.Check(); err != nil {
- return err
- }
- *s.volumeSet = append(*s.volumeSet, v)
+func (s *azureVolumeAdder) Set(containerName string) error {
+ s.Config.Volumes = append(s.Config.Volumes, &AzureBlobVolume{
+ ContainerName: containerName,
+ StorageAccountName: azureStorageAccountName,
+ StorageAccountKeyFile: azureStorageAccountKeyFile,
+ AzureReplication: azureStorageReplication,
+ ReadOnly: deprecated.flagReadonly,
+ })
return nil
}
func init() {
- flag.Var(&azureVolumeAdder{&volumes},
+ VolumeTypes = append(VolumeTypes, func() VolumeWithExamples { return &AzureBlobVolume{} })
+
+ flag.Var(&azureVolumeAdder{theConfig},
"azure-storage-container-volume",
"Use the given container as a storage volume. Can be given multiple times.")
flag.StringVar(
&azureStorageAccountKeyFile,
"azure-storage-account-key-file",
"",
- "File containing the account key used for subsequent --azure-storage-container-volume arguments.")
+ "`File` containing the account key used for subsequent --azure-storage-container-volume arguments.")
flag.IntVar(
&azureStorageReplication,
"azure-storage-replication",
// An AzureBlobVolume stores and retrieves blocks in an Azure Blob
// container.
type AzureBlobVolume struct {
- azClient storage.Client
- bsClient storage.BlobStorageClient
- containerName string
- readonly bool
- replication int
+ StorageAccountName string
+ StorageAccountKeyFile string
+ ContainerName string
+ AzureReplication int
+ ReadOnly bool
+ RequestTimeout arvados.Duration
+
+ azClient storage.Client
+ bsClient storage.BlobStorageClient
}
-// NewAzureBlobVolume returns a new AzureBlobVolume using the given
-// client and container name. The replication argument specifies the
-// replication level to report when writing data.
-func NewAzureBlobVolume(client storage.Client, containerName string, readonly bool, replication int) *AzureBlobVolume {
- return &AzureBlobVolume{
- azClient: client,
- bsClient: client.GetBlobService(),
- containerName: containerName,
- readonly: readonly,
- replication: replication,
+// Examples implements VolumeWithExamples.
+func (*AzureBlobVolume) Examples() []Volume {
+ return []Volume{
+ &AzureBlobVolume{
+ StorageAccountName: "example-account-name",
+ StorageAccountKeyFile: "/etc/azure_storage_account_key.txt",
+ ContainerName: "example-container-name",
+ AzureReplication: 3,
+ RequestTimeout: azureDefaultRequestTimeout,
+ },
}
}
-// Check returns nil if the volume is usable.
-func (v *AzureBlobVolume) Check() error {
- ok, err := v.bsClient.ContainerExists(v.containerName)
+// Type implements Volume.
+func (v *AzureBlobVolume) Type() string {
+ return "Azure"
+}
+
+// Start implements Volume.
+func (v *AzureBlobVolume) Start() error {
+ if v.ContainerName == "" {
+ return errors.New("no container name given")
+ }
+ if v.StorageAccountName == "" || v.StorageAccountKeyFile == "" {
+ return errors.New("StorageAccountName and StorageAccountKeyFile must be given")
+ }
+ accountKey, err := readKeyFromFile(v.StorageAccountKeyFile)
+ if err != nil {
+ return err
+ }
+ v.azClient, err = storage.NewBasicClient(v.StorageAccountName, accountKey)
+ if err != nil {
+ return fmt.Errorf("creating Azure storage client: %s", err)
+ }
+
+ if v.RequestTimeout == 0 {
+ v.RequestTimeout = azureDefaultRequestTimeout
+ }
+ v.azClient.HTTPClient = &http.Client{
+ Timeout: time.Duration(v.RequestTimeout),
+ }
+ v.bsClient = v.azClient.GetBlobService()
+
+ ok, err := v.bsClient.ContainerExists(v.ContainerName)
if err != nil {
return err
}
if !ok {
- return errors.New("container does not exist")
+ return fmt.Errorf("Azure container %q does not exist", v.ContainerName)
}
return nil
}
+// Return true if expires_at metadata attribute is found on the block
+func (v *AzureBlobVolume) checkTrashed(loc string) (bool, map[string]string, error) {
+ metadata, err := v.bsClient.GetBlobMetadata(v.ContainerName, loc)
+ if err != nil {
+ return false, metadata, v.translateError(err)
+ }
+ if metadata["expires_at"] != "" {
+ return true, metadata, nil
+ }
+ return false, metadata, nil
+}
+
// Get reads a Keep block that has been stored as a block blob in the
// container.
//
// If the block is younger than azureWriteRaceInterval and is
// unexpectedly empty, assume a PutBlob operation is in progress, and
// wait for it to finish writing.
-func (v *AzureBlobVolume) Get(loc string) ([]byte, error) {
+func (v *AzureBlobVolume) Get(ctx context.Context, loc string, buf []byte) (int, error) {
+ trashed, _, err := v.checkTrashed(loc)
+ if err != nil {
+ return 0, err
+ }
+ if trashed {
+ return 0, os.ErrNotExist
+ }
var deadline time.Time
haveDeadline := false
- buf, err := v.get(loc)
- for err == nil && len(buf) == 0 && loc != "d41d8cd98f00b204e9800998ecf8427e" {
+ size, err := v.get(loc, buf)
+ for err == nil && size == 0 && loc != "d41d8cd98f00b204e9800998ecf8427e" {
// Seeing a brand new empty block probably means we're
// in a race with CreateBlob, which under the hood
// (apparently) does "CreateEmpty" and "CommitData"
} else if time.Now().After(deadline) {
break
}
- bufs.Put(buf)
time.Sleep(azureWriteRacePollTime)
- buf, err = v.get(loc)
+ size, err = v.get(loc, buf)
}
if haveDeadline {
- log.Printf("Race ended with len(buf)==%d", len(buf))
+ log.Printf("Race ended with size==%d", size)
}
- return buf, err
+ return size, err
}
-func (v *AzureBlobVolume) get(loc string) ([]byte, error) {
- expectSize := BlockSize
+func (v *AzureBlobVolume) get(loc string, buf []byte) (int, error) {
+ expectSize := len(buf)
if azureMaxGetBytes < BlockSize {
// Unfortunately the handler doesn't tell us how long the blob
// is expected to be, so we have to ask Azure.
- props, err := v.bsClient.GetBlobProperties(v.containerName, loc)
+ props, err := v.bsClient.GetBlobProperties(v.ContainerName, loc)
if err != nil {
- return nil, v.translateError(err)
+ return 0, v.translateError(err)
}
if props.ContentLength > int64(BlockSize) || props.ContentLength < 0 {
- return nil, fmt.Errorf("block %s invalid size %d (max %d)", loc, props.ContentLength, BlockSize)
+ return 0, fmt.Errorf("block %s invalid size %d (max %d)", loc, props.ContentLength, BlockSize)
}
expectSize = int(props.ContentLength)
}
- buf := bufs.Get(expectSize)
if expectSize == 0 {
- return buf, nil
+ return 0, nil
}
// We'll update this actualSize if/when we get the last piece.
var rdr io.ReadCloser
var err error
if startPos == 0 && endPos == expectSize {
- rdr, err = v.bsClient.GetBlob(v.containerName, loc)
+ rdr, err = v.bsClient.GetBlob(v.ContainerName, loc)
} else {
- rdr, err = v.bsClient.GetBlobRange(v.containerName, loc, fmt.Sprintf("%d-%d", startPos, endPos-1))
+ rdr, err = v.bsClient.GetBlobRange(v.ContainerName, loc, fmt.Sprintf("%d-%d", startPos, endPos-1), nil)
}
if err != nil {
errors[p] = err
wg.Wait()
for _, err := range errors {
if err != nil {
- bufs.Put(buf)
- return nil, v.translateError(err)
+ return 0, v.translateError(err)
}
}
- return buf[:actualSize], nil
+ return actualSize, nil
}
// Compare the given data with existing stored data.
-func (v *AzureBlobVolume) Compare(loc string, expect []byte) error {
- rdr, err := v.bsClient.GetBlob(v.containerName, loc)
+func (v *AzureBlobVolume) Compare(ctx context.Context, loc string, expect []byte) error {
+ trashed, _, err := v.checkTrashed(loc)
+ if err != nil {
+ return err
+ }
+ if trashed {
+ return os.ErrNotExist
+ }
+ rdr, err := v.bsClient.GetBlob(v.ContainerName, loc)
if err != nil {
return v.translateError(err)
}
defer rdr.Close()
- return compareReaderWithBuf(rdr, expect, loc[:32])
+ return compareReaderWithBuf(ctx, rdr, expect, loc[:32])
}
// Put stores a Keep block as a block blob in the container.
-func (v *AzureBlobVolume) Put(loc string, block []byte) error {
- if v.readonly {
+func (v *AzureBlobVolume) Put(ctx context.Context, loc string, block []byte) error {
+ if v.ReadOnly {
return MethodDisabledError
}
- return v.bsClient.CreateBlockBlobFromReader(v.containerName, loc, uint64(len(block)), bytes.NewReader(block))
+ return v.bsClient.CreateBlockBlobFromReader(v.ContainerName, loc, uint64(len(block)), bytes.NewReader(block), nil)
}
// Touch updates the last-modified property of a block blob.
func (v *AzureBlobVolume) Touch(loc string) error {
- if v.readonly {
+ if v.ReadOnly {
return MethodDisabledError
}
- return v.bsClient.SetBlobMetadata(v.containerName, loc, map[string]string{
- "touch": fmt.Sprintf("%d", time.Now()),
- })
+ trashed, metadata, err := v.checkTrashed(loc)
+ if err != nil {
+ return err
+ }
+ if trashed {
+ return os.ErrNotExist
+ }
+
+ metadata["touch"] = fmt.Sprintf("%d", time.Now())
+ return v.bsClient.SetBlobMetadata(v.ContainerName, loc, metadata, nil)
}
// Mtime returns the last-modified property of a block blob.
func (v *AzureBlobVolume) Mtime(loc string) (time.Time, error) {
- props, err := v.bsClient.GetBlobProperties(v.containerName, loc)
+ trashed, _, err := v.checkTrashed(loc)
+ if err != nil {
+ return time.Time{}, err
+ }
+ if trashed {
+ return time.Time{}, os.ErrNotExist
+ }
+
+ props, err := v.bsClient.GetBlobProperties(v.ContainerName, loc)
if err != nil {
return time.Time{}, err
}
// container.
func (v *AzureBlobVolume) IndexTo(prefix string, writer io.Writer) error {
params := storage.ListBlobsParameters{
- Prefix: prefix,
+ Prefix: prefix,
+ Include: "metadata",
}
for {
- resp, err := v.bsClient.ListBlobs(v.containerName, params)
+ resp, err := v.bsClient.ListBlobs(v.ContainerName, params)
if err != nil {
return err
}
// value.
continue
}
- fmt.Fprintf(writer, "%s+%d %d\n", b.Name, b.Properties.ContentLength, t.Unix())
+ if b.Metadata["expires_at"] != "" {
+ // Trashed blob; exclude it from response
+ continue
+ }
+ fmt.Fprintf(writer, "%s+%d %d\n", b.Name, b.Properties.ContentLength, t.UnixNano())
}
if resp.NextMarker == "" {
return nil
// Trash a Keep block.
func (v *AzureBlobVolume) Trash(loc string) error {
- if v.readonly {
+ if v.ReadOnly {
return MethodDisabledError
}
+
// Ideally we would use If-Unmodified-Since, but that
// particular condition seems to be ignored by Azure. Instead,
// we get the Etag before checking Mtime, and use If-Match to
// ensure we don't delete data if Put() or Touch() happens
// between our calls to Mtime() and DeleteBlob().
- props, err := v.bsClient.GetBlobProperties(v.containerName, loc)
+ props, err := v.bsClient.GetBlobProperties(v.ContainerName, loc)
if err != nil {
return err
}
if t, err := v.Mtime(loc); err != nil {
return err
- } else if time.Since(t) < blobSignatureTTL {
+ } else if time.Since(t) < theConfig.BlobSignatureTTL.Duration() {
return nil
}
- return v.bsClient.DeleteBlob(v.containerName, loc, map[string]string{
+
+ // If TrashLifetime == 0, just delete it
+ if theConfig.TrashLifetime == 0 {
+ return v.bsClient.DeleteBlob(v.ContainerName, loc, map[string]string{
+ "If-Match": props.Etag,
+ })
+ }
+
+ // Otherwise, mark as trash
+ return v.bsClient.SetBlobMetadata(v.ContainerName, loc, map[string]string{
+ "expires_at": fmt.Sprintf("%d", time.Now().Add(theConfig.TrashLifetime.Duration()).Unix()),
+ }, map[string]string{
"If-Match": props.Etag,
})
}
// Untrash a Keep block.
-// TBD
+// Delete the expires_at metadata attribute
func (v *AzureBlobVolume) Untrash(loc string) error {
- return nil
+ // if expires_at does not exist, return NotFoundError
+ metadata, err := v.bsClient.GetBlobMetadata(v.ContainerName, loc)
+ if err != nil {
+ return v.translateError(err)
+ }
+ if metadata["expires_at"] == "" {
+ return os.ErrNotExist
+ }
+
+ // reset expires_at metadata attribute
+ metadata["expires_at"] = ""
+ err = v.bsClient.SetBlobMetadata(v.ContainerName, loc, metadata, nil)
+ return v.translateError(err)
}
// Status returns a VolumeStatus struct with placeholder data.
// String returns a volume label, including the container name.
func (v *AzureBlobVolume) String() string {
- return fmt.Sprintf("azure-storage-container:%+q", v.containerName)
+ return fmt.Sprintf("azure-storage-container:%+q", v.ContainerName)
}
// Writable returns true, unless the -readonly flag was on when the
// volume was added.
func (v *AzureBlobVolume) Writable() bool {
- return !v.readonly
+ return !v.ReadOnly
}
// Replication returns the replication level of the container, as
// specified by the -azure-storage-replication argument.
func (v *AzureBlobVolume) Replication() int {
- return v.replication
+ return v.AzureReplication
}
// If possible, translate an Azure SDK error to a recognizable error
switch {
case err == nil:
return err
- case strings.Contains(err.Error(), "404 Not Found"):
+ case strings.Contains(err.Error(), "Not Found"):
// "storage: service returned without a response body (404 Not Found)"
return os.ErrNotExist
default:
func (v *AzureBlobVolume) isKeepBlock(s string) bool {
return keepBlockRegexp.MatchString(s)
}
+
+// EmptyTrash looks for trashed blocks that exceeded TrashLifetime
+// and deletes them from the volume.
+func (v *AzureBlobVolume) EmptyTrash() {
+ var bytesDeleted, bytesInTrash int64
+ var blocksDeleted, blocksInTrash int
+ params := storage.ListBlobsParameters{Include: "metadata"}
+
+ for {
+ resp, err := v.bsClient.ListBlobs(v.ContainerName, params)
+ if err != nil {
+ log.Printf("EmptyTrash: ListBlobs: %v", err)
+ break
+ }
+ for _, b := range resp.Blobs {
+ // Check if the block is expired
+ if b.Metadata["expires_at"] == "" {
+ continue
+ }
+
+ blocksInTrash++
+ bytesInTrash += b.Properties.ContentLength
+
+ expiresAt, err := strconv.ParseInt(b.Metadata["expires_at"], 10, 64)
+ if err != nil {
+ log.Printf("EmptyTrash: ParseInt(%v): %v", b.Metadata["expires_at"], err)
+ continue
+ }
+
+ if expiresAt > time.Now().Unix() {
+ continue
+ }
+
+ err = v.bsClient.DeleteBlob(v.ContainerName, b.Name, map[string]string{
+ "If-Match": b.Properties.Etag,
+ })
+ if err != nil {
+ log.Printf("EmptyTrash: DeleteBlob(%v): %v", b.Name, err)
+ continue
+ }
+ blocksDeleted++
+ bytesDeleted += b.Properties.ContentLength
+ }
+ if resp.NextMarker == "" {
+ break
+ }
+ params.Marker = resp.NextMarker
+ }
+
+ log.Printf("EmptyTrash stats for %v: Deleted %v bytes in %v blocks. Remaining in trash: %v bytes in %v blocks.", v.String(), bytesDeleted, blocksDeleted, bytesInTrash-bytesDeleted, blocksInTrash-blocksDeleted)
+}