[#227] Add versionID header after complete multipart

Signed-off-by: Marina Biryukova <m.biryukova@yadro.com>
This commit is contained in:
Marina Biryukova 2023-10-26 11:33:51 +03:00 committed by Alexey Vanin
parent 0bed25816c
commit 890a8ed237
11 changed files with 25 additions and 163 deletions

View file

@ -34,6 +34,7 @@ This document outlines major changes between releases.
- Add new `kludge.bypass_content_encoding_check_in_chunks` config param (#146)
- Add new `frostfs.client_cut` config param (#192)
- Add new `frostfs.buffer_max_size_for_put` config param and sync TZ hash for PUT operations (#197)
- Add `X-Amz-Version-Id` header after complete multipart upload (#227)
### Changed
- Update prometheus to v1.15.0 (#94)
@ -53,6 +54,7 @@ This document outlines major changes between releases.
### Removed
- Drop `tree.service` param (now endpoints from `peers` section are used) (#133)
- Drop sending whitespace characters during complete multipart upload and related config param `kludge.complete_multipart_keepalive` (#227)
## [0.27.0] - Karpinsky - 2023-07-12

View file

@ -40,7 +40,6 @@ type (
NotificatorEnabled() bool
ResolveZoneList() []string
IsResolveListAllow() bool
CompleteMultipartKeepalive() time.Duration
BypassContentEncodingInChunks() bool
MD5Enabled() bool
}

View file

@ -3,7 +3,6 @@ package handler
import (
"encoding/xml"
"fmt"
"io"
"net/http"
"net/url"
"strconv"
@ -399,6 +398,12 @@ func (h *handler) CompleteMultipartUploadHandler(w http.ResponseWriter, r *http.
return
}
settings, err := h.obj.GetBucketSettings(r.Context(), bktInfo)
if err != nil {
h.logAndSendError(w, "could not get bucket settings", reqInfo, err)
return
}
var (
uploadID = r.URL.Query().Get(uploadIDHeaderName)
uploadInfo = &layer.UploadInfoParams{
@ -425,29 +430,12 @@ func (h *handler) CompleteMultipartUploadHandler(w http.ResponseWriter, r *http.
Parts: reqBody.Parts,
}
// Next operations might take some time, so we want to keep client's
// connection alive. To do so, gateway sends periodic white spaces
// back to the client the same way as Amazon S3 service does.
stopPeriodicResponseWriter := periodicXMLWriter(w, h.cfg.CompleteMultipartKeepalive())
// Start complete multipart upload which may take some time to fetch object
// and re-upload it part by part.
objInfo, err := h.completeMultipartUpload(r, c, bktInfo, reqInfo)
// Stop periodic writer as complete multipart upload is finished
// successfully or not.
headerIsWritten := stopPeriodicResponseWriter()
responseWriter := middleware.EncodeToResponse
errLogger := h.logAndSendError
// Do not send XML and HTTP headers if periodic writer was invoked at this point.
if headerIsWritten {
responseWriter = middleware.EncodeToResponseNoHeader
errLogger = h.logAndSendErrorNoHeader
}
if err != nil {
errLogger(w, "complete multipart error", reqInfo, err, additional...)
h.logAndSendError(w, "complete multipart error", reqInfo, err, additional...)
return
}
@ -457,12 +445,12 @@ func (h *handler) CompleteMultipartUploadHandler(w http.ResponseWriter, r *http.
ETag: objInfo.ETag(h.cfg.MD5Enabled()),
}
// Here we previously set api.AmzVersionID header for versioned bucket.
// It is not possible after #60, because of periodic white
// space XML writer to keep connection with the client.
if settings.VersioningEnabled() {
w.Header().Set(api.AmzVersionID, objInfo.VersionID())
}
if err = responseWriter(w, response); err != nil {
errLogger(w, "something went wrong", reqInfo, err, additional...)
if err = middleware.EncodeToResponse(w, response); err != nil {
h.logAndSendError(w, "something went wrong", reqInfo, err, additional...)
}
}
@ -730,53 +718,3 @@ func encodeListPartsToResponse(info *layer.ListPartsInfo, params *layer.ListPart
StorageClass: api.DefaultStorageClass,
}
}
// periodicXMLWriter creates go routine to write xml header and whitespaces
// over time to avoid connection drop from the client. To work properly,
// pass `http.ResponseWriter` with implemented `http.Flusher` interface.
// Returns stop function which returns boolean if writer has been used
// during goroutine execution. To disable writer, pass 0 duration value.
func periodicXMLWriter(w io.Writer, dur time.Duration) (stop func() bool) {
if dur == 0 { // 0 duration disables periodic writer
return func() bool { return false }
}
whitespaceChar := []byte(" ")
closer := make(chan struct{})
done := make(chan struct{})
headerWritten := false
go func() {
defer close(done)
tick := time.NewTicker(dur)
defer tick.Stop()
for {
select {
case <-tick.C:
if !headerWritten {
_, err := w.Write([]byte(xml.Header))
headerWritten = err == nil
}
_, err := w.Write(whitespaceChar)
if err != nil {
return // is there anything we can do better than ignore error?
}
if buffered, ok := w.(http.Flusher); ok {
buffered.Flush()
}
case <-closer:
return
}
}
}()
stop = func() bool {
close(closer)
<-done // wait for goroutine to stop
return headerWritten
}
return stop
}

View file

@ -1,7 +1,6 @@
package handler
import (
"bytes"
"crypto/md5"
"encoding/hex"
"encoding/xml"
@ -10,7 +9,6 @@ import (
"net/url"
"strconv"
"testing"
"time"
"git.frostfs.info/TrueCloudLab/frostfs-s3-gw/api"
s3Errors "git.frostfs.info/TrueCloudLab/frostfs-s3-gw/api/errors"
@ -22,44 +20,6 @@ const (
partNumberMarkerQuery = "part-number-marker"
)
func TestPeriodicWriter(t *testing.T) {
const dur = 100 * time.Millisecond
const whitespaces = 8
expected := []byte(xml.Header)
for i := 0; i < whitespaces; i++ {
expected = append(expected, []byte(" ")...)
}
t.Run("writes data", func(t *testing.T) {
buf := bytes.NewBuffer(nil)
stop := periodicXMLWriter(buf, dur)
// N number of whitespaces + half durations to guarantee at least N writes in buffer
time.Sleep(whitespaces*dur + dur/2)
require.True(t, stop())
require.Equal(t, expected, buf.Bytes())
t.Run("no additional data after stop", func(t *testing.T) {
time.Sleep(2 * dur)
require.Equal(t, expected, buf.Bytes())
})
})
t.Run("does not write data", func(t *testing.T) {
buf := bytes.NewBuffer(nil)
stop := periodicXMLWriter(buf, dur)
time.Sleep(dur / 2)
require.False(t, stop())
require.Empty(t, buf.Bytes())
t.Run("disabled", func(t *testing.T) {
stop = periodicXMLWriter(buf, 0)
require.False(t, stop())
require.Empty(t, buf.Bytes())
})
})
}
func TestMultipartUploadInvalidPart(t *testing.T) {
hc := prepareHandlerContext(t)

View file

@ -56,22 +56,6 @@ func handleDeleteMarker(w http.ResponseWriter, err error) error {
return fmt.Errorf("%w: %s", s3errors.GetAPIError(target.ErrorCode), err)
}
func (h *handler) logAndSendErrorNoHeader(w http.ResponseWriter, logText string, reqInfo *middleware.ReqInfo, err error, additional ...zap.Field) {
middleware.WriteErrorResponseNoHeader(w, reqInfo, transformToS3Error(err))
fields := []zap.Field{
zap.String("request_id", reqInfo.RequestID),
zap.String("method", reqInfo.API),
zap.String("bucket", reqInfo.BucketName),
zap.String("object", reqInfo.ObjectName),
zap.String("description", logText),
zap.Error(err)}
fields = append(fields, additional...)
if traceID, err := trace.TraceIDFromHex(reqInfo.TraceID); err == nil && traceID.IsValid() {
fields = append(fields, zap.String("trace_id", reqInfo.TraceID))
}
h.log.Error(logs.RequestFailed, fields...) // consider using h.reqLogger (it requires accept context.Context or http.Request)
}
func transformToS3Error(err error) error {
err = frosterrors.UnwrapErr(err) // this wouldn't work with errors.Join
if _, ok := err.(s3errors.Error); ok {

View file

@ -141,13 +141,6 @@ func WriteErrorResponse(w http.ResponseWriter, reqInfo *ReqInfo, err error) int
return code
}
// WriteErrorResponseNoHeader writes XML encoded error to the response body.
func WriteErrorResponseNoHeader(w http.ResponseWriter, reqInfo *ReqInfo, err error) {
errorResponse := getAPIErrorResponse(reqInfo, err)
encodedErrorResponse := EncodeResponse(errorResponse)
WriteResponseBody(w, encodedErrorResponse)
}
// Write http common headers.
func setCommonHeaders(w http.ResponseWriter) {
w.Header().Set(hdrServerInfo, version.Server)

View file

@ -75,7 +75,6 @@ type (
notificatorEnabled bool
resolveZoneList []string
isResolveListAllow bool // True if ResolveZoneList contains allowed zones
completeMultipartKeepalive time.Duration
mu sync.RWMutex
defaultPolicy netmap.PlacementPolicy
@ -179,7 +178,6 @@ func newAppSettings(log *Logger, v *viper.Viper) *appSettings {
defaultXMLNS: v.GetBool(cfgKludgeUseDefaultXMLNS),
defaultMaxAge: fetchDefaultMaxAge(v, log.logger),
notificatorEnabled: v.GetBool(cfgEnableNATS),
completeMultipartKeepalive: v.GetDuration(cfgKludgeCompleteMultipartUploadKeepalive),
}
settings.resolveZoneList = v.GetStringSlice(cfgResolveBucketAllow)
@ -310,10 +308,6 @@ func (s *appSettings) IsResolveListAllow() bool {
return s.isResolveListAllow
}
func (s *appSettings) CompleteMultipartKeepalive() time.Duration {
return s.completeMultipartKeepalive
}
func (s *appSettings) MD5Enabled() bool {
s.mu.RLock()
defer s.mu.RUnlock()

View file

@ -134,7 +134,6 @@ const ( // Settings.
// Kludge.
cfgKludgeUseDefaultXMLNS = "kludge.use_default_xmlns"
cfgKludgeCompleteMultipartUploadKeepalive = "kludge.complete_multipart_keepalive"
cfgKludgeBypassContentEncodingCheckInChunks = "kludge.bypass_content_encoding_check_in_chunks"
// Web.
@ -547,7 +546,6 @@ func newSettings() *viper.Viper {
// kludge
v.SetDefault(cfgKludgeUseDefaultXMLNS, false)
v.SetDefault(cfgKludgeCompleteMultipartUploadKeepalive, 10*time.Second)
v.SetDefault(cfgKludgeBypassContentEncodingCheckInChunks, false)
// web

View file

@ -140,8 +140,6 @@ S3_GW_RESOLVE_BUCKET_ALLOW=container
# Enable using default xml namespace `http://s3.amazonaws.com/doc/2006-03-01/` when parse xml bodies.
S3_GW_KLUDGE_USE_DEFAULT_XMLNS=false
# Set timeout between whitespace transmissions during CompleteMultipartUpload processing.
S3_GW_KLUDGE_COMPLETE_MULTIPART_KEEPALIVE=10s
# Use this flag to be able to use chunked upload approach without having `aws-chunked` value in `Content-Encoding` header.
S3_GW_BYPASS_CONTENT_ENCODING_CHECK_IN_CHUNKS=false

View file

@ -169,8 +169,6 @@ resolve_bucket:
kludge:
# Enable using default xml namespace `http://s3.amazonaws.com/doc/2006-03-01/` when parse xml bodies.
use_default_xmlns: false
# Set timeout between whitespace transmissions during CompleteMultipartUpload processing.
complete_multipart_keepalive: 10s
# Use this flag to be able to use chunked upload approach without having `aws-chunked` value in `Content-Encoding` header.
bypass_content_encoding_check_in_chunks: false

View file

@ -543,14 +543,12 @@ Workarounds for non-standard use cases.
```yaml
kludge:
use_default_xmlns: false
complete_multipart_keepalive: 10s
bypass_content_encoding_check_in_chunks: false
```
| Parameter | Type | SIGHUP reload | Default value | Description |
|-------------------------------------------|------------|---------------|---------------|---------------------------------------------------------------------------------------------------------------------------------|
| `use_default_xmlns` | `bool` | yes | false | Enable using default xml namespace `http://s3.amazonaws.com/doc/2006-03-01/` when parse xml bodies. |
| `complete_multipart_keepalive` | `duration` | no | 10s | Set timeout between whitespace transmissions during CompleteMultipartUpload processing. |
| `bypass_content_encoding_check_in_chunks` | `bool` | yes | false | Use this flag to be able to use [chunked upload approach](https://docs.aws.amazon.com/AmazonS3/latest/API/sigv4-streaming.html) without having `aws-chunked` value in `Content-Encoding` header. |
# `runtime` section