Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Resolving 500 server timeouts on append blob appends #2430

Merged
merged 22 commits into from
Nov 29, 2023
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions common/fe-ste-models.go
Original file line number Diff line number Diff line change
Expand Up @@ -1069,6 +1069,7 @@ const (
MaxAppendBlobBlockSize = 4 * 1024 * 1024
DefaultPageBlobChunkSize = 4 * 1024 * 1024
DefaultAzureFileChunkSize = 4 * 1024 * 1024
MaxRangeGetSize = 4 * 1024 * 1024
MaxNumberOfBlocksPerBlob = 50000
BlockSizeThreshold = 256 * 1024 * 1024
MinParallelChunkCountThreshold = 4 /* minimum number of chunks in parallel for AzCopy to be performant. */
Expand Down
62 changes: 60 additions & 2 deletions ste/sender-appendBlob.go
Original file line number Diff line number Diff line change
Expand Up @@ -21,12 +21,16 @@
package ste

import (
"bytes"
"context"
"crypto/md5"
"fmt"
"time"

"github.com/Azure/azure-sdk-for-go/sdk/azcore/to"
"github.com/Azure/azure-sdk-for-go/sdk/storage/azblob/appendblob"
"github.com/Azure/azure-sdk-for-go/sdk/storage/azblob/blob"
"github.com/Azure/azure-sdk-for-go/sdk/storage/azblob/bloberror"
"io"
"time"

"golang.org/x/sync/semaphore"

Expand Down Expand Up @@ -208,3 +212,57 @@ func (s *appendBlobSenderBase) GetDestinationLength() (int64, error) {
}
return *prop.ContentLength, nil
}

func (s *appendBlobSenderBase) GetMD5(offset, count int64) ([]byte, error) {
var rangeGetContentMD5 *bool
if count <= common.MaxRangeGetSize {
rangeGetContentMD5 = to.Ptr(true)
}
response, err := s.destAppendBlobClient.DownloadStream(s.jptm.Context(),
&blob.DownloadStreamOptions{
Range: blob.HTTPRange{Offset: offset, Count: count},
RangeGetContentMD5: rangeGetContentMD5,
CPKInfo: s.jptm.CpkInfo(),
CPKScopeInfo: s.jptm.CpkScopeInfo(),
})
if err != nil {
return nil, err
}
if response.ContentMD5 != nil && len(response.ContentMD5) > 0 {
return response.ContentMD5, nil
} else {
// compute md5
body := response.NewRetryReader(s.jptm.Context(), &blob.RetryReaderOptions{MaxRetries: MaxRetryPerDownloadBody})
defer body.Close()
h := md5.New()
if _, err = io.Copy(h, body); err != nil {
return nil, err
}
return h.Sum(nil), nil
}
}

func (s *appendBlobSenderBase) transformAppendConditionMismatchError(timeoutFromCtx bool, offset, count int64, err error) (string, error) {
if err != nil && bloberror.HasCode(err, bloberror.AppendPositionConditionNotMet) && timeoutFromCtx {
if _, ok := s.sip.(benchmarkSourceInfoProvider); ok {
// If the source is a benchmark, then we don't need to check MD5 since the data is constantly changing. This is OK.
return "", nil
}
// Download Range of last append
destMD5, destErr := s.GetMD5(offset, count)
if destErr != nil {
return ", get destination md5 after timeout", destErr
}
sourceMD5, sourceErr := s.sip.GetMD5(offset, count)
if sourceErr != nil {
return ", get source md5 after timeout", sourceErr
}
if destMD5 != nil && sourceMD5 != nil && len(destMD5) > 0 && len(sourceMD5) > 0 {
// Compare MD5
if bytes.Equal(destMD5, sourceMD5) {
return "", nil
}
}
}
return "", err
}
12 changes: 8 additions & 4 deletions ste/sender-appendBlobFromLocal.go
Original file line number Diff line number Diff line change
Expand Up @@ -68,14 +68,18 @@ func (u *appendBlobUploader) GenerateUploadFunc(id common.ChunkID, blockIndex in
u.jptm.LogChunkStatus(id, common.EWaitReason.Body())
body := newPacedRequestBody(u.jptm.Context(), reader, u.pacer)
offset := id.OffsetInFile()
_, err := u.destAppendBlobClient.AppendBlock(u.jptm.Context(), body,
var timeoutFromCtx bool
ctx := withTimeoutNotification(u.jptm.Context(), &timeoutFromCtx)
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why do we need context with timeout here, in older code its different right ?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

this is actually a notifier that a 500 status code was returned during the append block retries so we can check below. This is not adding a request timeout

_, err := u.destAppendBlobClient.AppendBlock(ctx, body,
&appendblob.AppendBlockOptions{
AppendPositionAccessConditions: &appendblob.AppendPositionAccessConditions{AppendPosition: &offset},
CPKInfo: u.jptm.CpkInfo(),
CPKScopeInfo: u.jptm.CpkScopeInfo(),
CPKInfo: u.jptm.CpkInfo(),
CPKScopeInfo: u.jptm.CpkScopeInfo(),
})
errString, err := u.transformAppendConditionMismatchError(timeoutFromCtx, offset, reader.Length(), err)
if err != nil {
u.jptm.FailActiveUpload("Appending block", err)
errString = "Appending block" + errString
u.jptm.FailActiveUpload(errString, err)
return
}
}
Expand Down
16 changes: 10 additions & 6 deletions ste/sender-appendBlobFromURL.go
Original file line number Diff line number Diff line change
Expand Up @@ -62,16 +62,20 @@ func (c *urlToAppendBlobCopier) GenerateCopyFunc(id common.ChunkID, blockIndex i
c.jptm.FailActiveS2SCopy("Getting source token credential", err)
return
}
_, err = c.destAppendBlobClient.AppendBlockFromURL(c.jptm.Context(), c.srcURL,
var timeoutFromCtx bool
ctx := withTimeoutNotification(c.jptm.Context(), &timeoutFromCtx)
_, err = c.destAppendBlobClient.AppendBlockFromURL(ctx, c.srcURL,
&appendblob.AppendBlockFromURLOptions{
Range: blob.HTTPRange{Offset: offset, Count: adjustedChunkSize},
Range: blob.HTTPRange{Offset: offset, Count: adjustedChunkSize},
AppendPositionAccessConditions: &appendblob.AppendPositionAccessConditions{AppendPosition: &offset},
CPKInfo: c.jptm.CpkInfo(),
CPKScopeInfo: c.jptm.CpkScopeInfo(),
CopySourceAuthorization: token,
CPKInfo: c.jptm.CpkInfo(),
CPKScopeInfo: c.jptm.CpkScopeInfo(),
CopySourceAuthorization: token,
})
errString, err := c.transformAppendConditionMismatchError(timeoutFromCtx, offset, adjustedChunkSize, err)
if err != nil {
c.jptm.FailActiveS2SCopy("Appending block from URL", err)
errString = "Appending block from URL" + errString
c.jptm.FailActiveS2SCopy(errString, err)
return
}
}
Expand Down
110 changes: 110 additions & 0 deletions ste/sender-appendBlob_test.go
Original file line number Diff line number Diff line change
@@ -0,0 +1,110 @@
package ste

import (
"context"
"fmt"
"github.com/Azure/azure-sdk-for-go/sdk/azcore"
"github.com/Azure/azure-sdk-for-go/sdk/azcore/policy"
"github.com/Azure/azure-sdk-for-go/sdk/azcore/streaming"
"github.com/Azure/azure-sdk-for-go/sdk/azcore/to"
"github.com/Azure/azure-sdk-for-go/sdk/storage/azblob/appendblob"
"github.com/Azure/azure-sdk-for-go/sdk/storage/azblob/blob"
blobsas "github.com/Azure/azure-sdk-for-go/sdk/storage/azblob/sas"
blobservice "github.com/Azure/azure-sdk-for-go/sdk/storage/azblob/service"
"github.com/Azure/azure-storage-azcopy/v10/common"
"github.com/stretchr/testify/assert"
"net/http"
"testing"
"time"
)

type appendErrorInjectionPolicy struct {
timedOut bool
}

func (r *appendErrorInjectionPolicy) Do(req *policy.Request) (*http.Response, error) {
if req.Raw().URL.Query().Get("comp") == "appendblock" && !r.timedOut {
req.Next()
var headers http.Header = make(map[string][]string)
headers.Add("x-ms-error-code", "OperationTimedOut")
r.timedOut = true
return &http.Response{StatusCode: 500, Header: headers}, nil
}
return req.Next()
}

func Test500FollowedBy412Logic(t *testing.T) {
a := assert.New(t)

// Setup source and destination
accountName, accountKey := getAccountAndKey()
rawURL := fmt.Sprintf("https://%s.blob.core.windows.net/", accountName)

credential, err := blob.NewSharedKeyCredential(accountName, accountKey)
a.Nil(err)

client, err := blobservice.NewClientWithSharedKeyCredential(rawURL, credential, &blobservice.ClientOptions{
ClientOptions: azcore.ClientOptions{
Transport: NewAzcopyHTTPClient(0),
}})
a.Nil(err)

cName := generateContainerName()
cc := client.NewContainerClient(cName)
_, err = cc.Create(context.Background(), nil)
a.Nil(err)
defer cc.Delete(context.Background(), nil)

sourceName := generateBlobName()
sourceClient := cc.NewBlockBlobClient(sourceName)
size := 1024 * 1024 * 10
dataReader, _ := getDataAndReader(t.Name(), size)
_, err = sourceClient.Upload(context.Background(), streaming.NopCloser(dataReader), nil)
a.Nil(err)

sasURL, err := cc.NewBlobClient(sourceName).GetSASURL(
blobsas.BlobPermissions{Read: true},
time.Now().Add(1*time.Hour),
nil)
a.Nil(err)

destName := generateBlobName()
destClient := cc.NewAppendBlobClient(destName)
destClient.Create(context.Background(), nil)
a.Nil(err)

jptm := testJobPartTransferManager{
info: to.Ptr(TransferInfo{
Source: sasURL,
SrcContainer: cName,
SrcFilePath: sourceName,
}),
fromTo: common.EFromTo.BlobBlob(),
}
blobSIP, err := newBlobSourceInfoProvider(jptm)
a.Nil(err)

injectionPolicy := &appendErrorInjectionPolicy{timedOut: false}
destClient, err = appendblob.NewClientWithSharedKeyCredential(destClient.URL(), credential, &appendblob.ClientOptions{
ClientOptions: azcore.ClientOptions{
PerRetryPolicies: []policy.Policy{newRetryNotificationPolicy(), injectionPolicy},
Transport: NewAzcopyHTTPClient(0),
},
})
a.Nil(err)
base := appendBlobSenderBase{jptm: jptm, destAppendBlobClient: destClient, sip: blobSIP}

// Get MD5 range within service calculation
offset := int64(0)
count := int64(common.MaxRangeGetSize)
var timeoutFromCtx bool
ctx := withTimeoutNotification(context.Background(), &timeoutFromCtx)
_, err = base.destAppendBlobClient.AppendBlockFromURL(ctx, sasURL,
&appendblob.AppendBlockFromURLOptions{
Range: blob.HTTPRange{Offset: offset, Count: count},
AppendPositionAccessConditions: &appendblob.AppendPositionAccessConditions{AppendPosition: &offset},
})
errString, err := base.transformAppendConditionMismatchError(timeoutFromCtx, offset, count, err)
a.Nil(err)
a.Empty(errString)
}
6 changes: 6 additions & 0 deletions ste/sourceInfoProvider-Benchmark.go
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,7 @@
package ste

import (
"errors"
"time"

"github.com/Azure/azure-storage-azcopy/v10/common"
Expand Down Expand Up @@ -57,3 +58,8 @@ func (b benchmarkSourceInfoProvider) GetFreshFileLastModifiedTime() (time.Time,
func (b benchmarkSourceInfoProvider) EntityType() common.EntityType {
return common.EEntityType.File() // no folders in benchmark
}

// Tell append blob it doesn't matter if the MD5 is wrong since this data doesn't matter
func (b benchmarkSourceInfoProvider) GetMD5(_, _ int64) ([]byte, error) {
return nil, errors.New("benchmark does not support retrieving MD5")
}
37 changes: 33 additions & 4 deletions ste/sourceInfoProvider-Blob.go
Original file line number Diff line number Diff line change
Expand Up @@ -21,18 +21,18 @@
package ste

import (
"io"
"time"

"crypto/md5"
"github.com/Azure/azure-sdk-for-go/sdk/azcore/to"
"github.com/Azure/azure-sdk-for-go/sdk/storage/azblob/blob"
"github.com/Azure/azure-storage-azcopy/v10/common"
"io"
"time"
)

// Source info provider for Azure blob
type blobSourceInfoProvider struct {
defaultRemoteSourceInfoProvider
source *blob.Client
source *blob.Client
}

func (p *blobSourceInfoProvider) IsDFSSource() bool {
Expand Down Expand Up @@ -147,3 +147,32 @@ func (p *blobSourceInfoProvider) GetFreshFileLastModifiedTime() (time.Time, erro
}
return common.IffNotNil(properties.LastModified, time.Time{}), nil
}

func (p *blobSourceInfoProvider) GetMD5(offset, count int64) ([]byte, error) {
var rangeGetContentMD5 *bool
if count <= common.MaxRangeGetSize {
rangeGetContentMD5 = to.Ptr(true)
}
response, err := p.source.DownloadStream(p.jptm.Context(),
&blob.DownloadStreamOptions{
Range: blob.HTTPRange{Offset: offset, Count: count},
RangeGetContentMD5: rangeGetContentMD5,
CPKInfo: p.jptm.CpkInfo(),
CPKScopeInfo: p.jptm.CpkScopeInfo(),
})
if err != nil {
return nil, err
}
if response.ContentMD5 != nil && len(response.ContentMD5) > 0 {
return response.ContentMD5, nil
} else {
// compute md5
body := response.NewRetryReader(p.jptm.Context(), &blob.RetryReaderOptions{MaxRetries: MaxRetryPerDownloadBody})
defer body.Close()
h := md5.New()
if _, err = io.Copy(h, body); err != nil {
return nil, err
}
return h.Sum(nil), nil
}
}
55 changes: 49 additions & 6 deletions ste/sourceInfoProvider-File.go
Original file line number Diff line number Diff line change
Expand Up @@ -22,11 +22,15 @@ package ste

import (
"context"
"crypto/md5"
"fmt"
"github.com/Azure/azure-sdk-for-go/sdk/azcore/to"
"github.com/Azure/azure-sdk-for-go/sdk/storage/azfile/directory"
"github.com/Azure/azure-sdk-for-go/sdk/storage/azfile/file"
"io"
"sync"
"time"

"github.com/Azure/azure-sdk-for-go/sdk/storage/azfile/directory"
"github.com/Azure/azure-sdk-for-go/sdk/storage/azfile/file"
"github.com/Azure/azure-sdk-for-go/sdk/storage/azfile/share"

"github.com/Azure/azure-storage-azcopy/v10/common"
Expand Down Expand Up @@ -180,10 +184,10 @@ func newFileSourceInfoProvider(jptm IJobPartTransferMgr) (ISourceInfoProvider, e
//TODO: Should we do that?
return &fileSourceInfoProvider{
defaultRemoteSourceInfoProvider: *base,
ctx: jptm.Context(),
cacheOnce: &sync.Once{},
srcShareClient: s.NewShareClient(jptm.Info().SrcContainer),
sourceURL: source.URL()}, nil
ctx: jptm.Context(),
cacheOnce: &sync.Once{},
srcShareClient: s.NewShareClient(jptm.Info().SrcContainer),
sourceURL: source.URL()}, nil
}

func (p *fileSourceInfoProvider) PreSignedSourceURL() (string, error) {
Expand Down Expand Up @@ -303,3 +307,42 @@ func (p *fileSourceInfoProvider) GetFreshFileLastModifiedTime() (time.Time, erro
// We ignore smblastwrite because otherwise the tx will fail s2s
return properties.LastModified(), nil
}

func (p *fileSourceInfoProvider) GetMD5(offset, count int64) ([]byte, error) {
switch p.EntityType() {
case common.EEntityType.File():
var rangeGetContentMD5 *bool
if count <= common.MaxRangeGetSize {
rangeGetContentMD5 = to.Ptr(true)
}
fsc, err := p.jptm.SrcServiceClient().FileServiceClient()
if err != nil {
return nil, err
}
shareClient := fsc.NewShareClient(p.transferInfo.SrcContainer)
fileClient := shareClient.NewRootDirectoryClient().NewFileClient(p.transferInfo.SrcFilePath)
response, err := fileClient.DownloadStream(p.ctx, &file.DownloadStreamOptions{
Range: file.HTTPRange{Offset: offset, Count: count},
RangeGetContentMD5: rangeGetContentMD5,
})
if err != nil {
return nil, err
}
if response.ContentMD5 != nil && len(response.ContentMD5) > 0 {
return response.ContentMD5, nil
} else {
// compute md5
body := response.NewRetryReader(p.jptm.Context(), &file.RetryReaderOptions{MaxRetries: MaxRetryPerDownloadBody})
defer body.Close()
h := md5.New()
if _, err = io.Copy(h, body); err != nil {
return nil, err
}
return h.Sum(nil), nil
}
case common.EEntityType.Folder():
return nil, fmt.Errorf("cannot get body or md5 of a folder")
default:
panic("unexpected case")
}
}
Loading