mirror of https://github.com/minio/minio.git
Allow delete marker replication from replica (#11566)
in the case of active-active replication. This PR also has the following changes: - add docs on replication design - fix corner case of completing versioned delete on a delete marker when the target is down and `mc rm --vid` is performed repeatedly. Instead the version should still be retained in the `PENDING|FAILED` state until replication sync completes. - remove `s3:Replication:OperationCompletedReplication` and `s3:Replication:OperationFailedReplication` from ObjectCreated events type
This commit is contained in:
Poorna Krishnamoorthy
GitHub
parent
95e0acbb26
commit
8e8a792d9d
|
|
@ -184,19 +184,15 @@ func checkReplicateDelete(ctx context.Context, bucket string, dobj ObjectToDelet
|
|||
if gerr != nil {
|
||||
validReplStatus := false
|
||||
switch oi.ReplicationStatus {
|
||||
// Point to note: even if two way replication with Deletemarker or Delete sync is enabled,
|
||||
// deletion of a REPLICA object/deletemarker will not trigger a replication to the other cluster.
|
||||
case replication.Pending, replication.Completed, replication.Failed:
|
||||
validReplStatus = true
|
||||
}
|
||||
if oi.DeleteMarker && validReplStatus {
|
||||
return oi.DeleteMarker, true, sync
|
||||
}
|
||||
return oi.DeleteMarker, false, sync
|
||||
}
|
||||
if oi.ReplicationStatus == replication.Replica {
|
||||
// avoid replicating a replica in bi-directional replication
|
||||
return oi.DeleteMarker, false, sync
|
||||
// can be the case that other cluster is down and duplicate `mc rm --vid`
|
||||
// is issued - this still needs to be replicated back to the other target
|
||||
return oi.DeleteMarker, oi.VersionPurgeStatus == Pending || oi.VersionPurgeStatus == Failed, sync
|
||||
}
|
||||
tgt := globalBucketTargetSys.GetRemoteTargetClient(ctx, rcfg.RoleArn)
|
||||
// the target online status should not be used here while deciding
|
||||
|
|
|
|||
|
|
@ -0,0 +1,88 @@
|
|||
# Bucket Replication Design [](https://slack.min.io) [](https://hub.docker.com/r/minio/minio/)
|
||||
|
||||
This document explains the design approach of server side bucket replication. If you're looking to get started with replication, we suggest you go through the [Bucket replication guide](https://github.com/minio/minio/blob/master/docs/bucket/replication/README.md) first.
|
||||
|
||||
## Overview
|
||||
Replication relies on immutability provided by versioning to sync objects between the configured source and replication target.
|
||||
|
||||
### Replication of object version and metadata
|
||||
If an object meets replication rules as set in the replication configuration, `X-Amz-Replication-Status` is first set to `PENDING` as the PUT operation completes and replication is queued (unless synchronous replication is in place). After replication is performed, the metadata on the source object version changes to `COMPLETED` or `FAILED` depending on whether replication succeeded. The object version on the target shows `X-Amz-Replication-Status` of `REPLICA`
|
||||
|
||||
All replication failures are picked up by the scanner which runs at a one minute frequency, each time scanning upto a sixteenth of the namespace. Object versions marked `PENDING` or `FAILED` are re-queued for replication.
|
||||
|
||||
Replication speed depends on the cluster load, number of objects in the object store as well as storage speed. In addition, any bandwidth limits set via `mc admin bucket remote add` could also contribute to replication speed. The number of workers used for replication defaults to 100. Based on network bandwidth and system load, the number of workers used in replication can be configured using `mc admin config set alias api` to set the `replication_workers`. The prometheus metrics exposed by MinIO can be used to plan resource allocation and bandwidth management to optimize replication speed.
|
||||
|
||||
If synchronous replication is configured above, replication is attempted right away prior to returning the PUT object response. In the event that the replication target is down, the `X-Amz-Replication-Status` is marked as `FAILED` and resynced with target when the scanner runs again.
|
||||
|
||||
Any metadata changes on the source object version, such as metadata updates via PutObjectTagging, PutObjectRetention, PutObjectLegalHold and COPY api are replicated in a similar manner to target version, with the `X-Amz-Replication-Status` again cycling through the same states.
|
||||
|
||||
The description above details one way replication from source to target w.r.t incoming object uploads and metadata changes to source object version. If active-active replication is configured, any incoming uploads and metadata changes to versions created on the target, will sync back to the source and be marked as `REPLICA` on the source. AWS, as well as MinIO do not by default sync metadata changes on a object version marked `REPLICA` back to source. This requires a setting in the replication configuration called [replica modification sync](https://aws.amazon.com/about-aws/whats-new/2020/12/amazon-s3-replication-adds-support-two-way-replication/) which is not yet available in MinIO.
|
||||
|
||||
For active-active replication, automatic failover occurs on `GET/HEAD` operations if object or object version requested qualifies for replication and is missing on one site, but present on the other. This allows the applications to take full advantage of two-way replication even before the two sites get fully synced.
|
||||
|
||||
### Replication of DeleteMarker and versioned Delete
|
||||
|
||||
MinIO allows DeleteMarker replication and versioned delete replication by setting `--replicate delete,delete-marker` while setting up replication configuration using `mc replicate add`. The MinIO implementation is based on V2 configuration, however it has been extended to allow both DeleteMarker replication and replication of versioned deletes with the `DeleteMarkerReplication` and `DeleteReplication` fields in the replication configuration. By default, this is set to `Disabled` unless the user specifies it while adding a replication rule.
|
||||
|
||||
Similar to object version replication, DeleteMarker replication also cycles through `PENDING` to `COMPLETED` or `FAILED` states for the `X-Amz-Replication-Status` on the source when a delete marker is set (i.e. performing `mc rm` on an object without specifying a version).After replication syncs the delete marker on the target, the DeleteMarker on the target shows `X-Amz-Replication-Status` of `REPLICA`. The status of DeleteMarker replication is returned by `X-Minio-Replication-DeleteMarker-Status` header on `HEAD/GET` calls for the delete marker version in question - i.e with `mc stat --version-id dm-version-id`
|
||||
|
||||
It must be noted that if active-active replication is set up with delete marker replication, there is potential for duplicate delete markers to be created if both source and target concurrently set a Delete Marker or if one/both of the clusters went down at tandem before the replication event was synced.This is an unavoidable side-effect in active-active replication caused by allowing delete markers set on a object version with `REPLICA` status back to source.
|
||||
|
||||
In the case of versioned deletes a.k.a permanent delete of a version by doing a `mc rm --version-id` on a object, replication implementation marks a object version permanently deleted as `PENDING` purge and deletes the version from source after syncing to the target and ensuring target version is deleted. The delete marker being deleted or object version being deleted will still be visible on listing with `mc ls --versions` until the sync is completed. Objects marked as deleted will not be accessible via `GET` or `HEAD` requests and would return a http response code of `405`. The status of versioned delete replication on the source can be queried by `HEAD` request on the delete marker versionID or object versionID in question.
|
||||
An additional header `X-Minio-Replication-Delete-Status` is returned which would show `PENDING` or `FAILED` status if the replication is still not caught up.
|
||||
|
||||
Existing object replication, replica modification sync for 2-way replication and multi site replication are currently not supported.
|
||||
|
||||
### Internal metadata for replication
|
||||
|
||||
`xl.meta` that is in use for [versioning](https://raw.githubusercontent.com/minio/minio/master/docs/bucket/versioning/DESIGN.md) has additional metadata for replication of objects,delete markers and versioned deletes.
|
||||
|
||||
### Metadata for object replication
|
||||
|
||||
```
|
||||
...
|
||||
"MetaSys": {},
|
||||
"MetaUsr": {
|
||||
"X-Amz-Replication-Status": "COMPLETED",
|
||||
"content-type": "application/octet-stream",
|
||||
"etag": "8315e643ed6a5d7c9962fc0a8ef9c11f"
|
||||
},
|
||||
"PartASizes": [
|
||||
26
|
||||
],
|
||||
...
|
||||
```
|
||||
|
||||
### Additional replication metadata for DeleteMarker
|
||||
|
||||
```
|
||||
{
|
||||
"DelObj": {
|
||||
"ID": "8+jguy20TOuzUCN2PTrESA==",
|
||||
"MTime": 1613601949645331516,
|
||||
"MetaSys": {
|
||||
"X-Amz-Replication-Status": "Q09NUExFVEVE"
|
||||
}
|
||||
},
|
||||
"Type": 2
|
||||
}
|
||||
```
|
||||
|
||||
### Additional replication metadata for versioned delete
|
||||
|
||||
```
|
||||
{
|
||||
"DelObj": {
|
||||
"ID": "8+jguy20TOuzUCN2PTrESA==",
|
||||
"MTime": 1613601949645331516,
|
||||
"MetaSys": {
|
||||
"purgestatus": "RkFJTEVE"
|
||||
}
|
||||
},
|
||||
"Type": 2
|
||||
}
|
||||
```
|
||||
|
||||
## Explore Further
|
||||
- [MinIO Bucket Versioning Implementation](https://docs.minio.io/docs/minio-bucket-versioning-guide.html)
|
||||
- [MinIO Client Quickstart Guide](https://docs.minio.io/docs/minio-client-quickstart-guide.html)
|
||||
|
|
@ -1,4 +1,4 @@
|
|||
# Bucket Replication Guide [](https://slack.min.io) [](https://hub.docker.com/r/minio/minio/)
|
||||
# Bucket Replication Guide [](https://slack.min.io) [](https://hub.docker.com/r/minio/minio/)
|
||||
|
||||
Bucket replication is designed to replicate selected objects in a bucket to a destination bucket.
|
||||
|
||||
|
|
@ -172,8 +172,7 @@ Status of delete marker replication can be viewed by doing a GET/HEAD on the obj
|
|||
|
||||
The status of replication can be monitored by configuring event notifications on the source and target buckets using `mc event add`.On the source side, the `s3:PutObject`, `s3:Replication:OperationCompletedReplication` and `s3:Replication:OperationFailedReplication` events show the status of replication in the `X-Amz-Replication-Status` metadata.
|
||||
|
||||
On the target bucket, `s3:PutObject` event shows `X-Amz-Replication-Status` status of `REPLICA` in the metadata. Additional metrics to monitor backlog state for the purpose of bandwidth management and resource allocation are
|
||||
an upcoming feature.
|
||||
On the target bucket, `s3:PutObject` event shows `X-Amz-Replication-Status` status of `REPLICA` in the metadata. Additional metrics to monitor backlog state for the purpose of bandwidth management and resource allocation are exposed via Prometheus - see https://github.com/minio/minio/blob/master/docs/metrics/prometheus/list.md for more details.
|
||||
|
||||
### Sync/Async Replication
|
||||
By default, replication is completed asynchronously. If synchronous replication is desired, set the --sync flag while adding a
|
||||
|
|
@ -183,5 +182,6 @@ remote replication target using the `mc admin bucket remote add` command
|
|||
```
|
||||
|
||||
## Explore Further
|
||||
- [MinIO Bucket Replication Design](https://raw.githubusercontent.com/minio/minio/master/docs/bucket/replication/DESIGN.md)
|
||||
- [MinIO Bucket Versioning Implementation](https://docs.minio.io/docs/minio-bucket-versioning-guide.html)
|
||||
- [MinIO Client Quickstart Guide](https://docs.minio.io/docs/minio-client-quickstart-guide.html)
|
||||
|
|
|
|||
|
|
@ -80,7 +80,6 @@ func (name Name) Expand() []Name {
|
|||
ObjectCreatedPost, ObjectCreatedPut,
|
||||
ObjectCreatedPutRetention, ObjectCreatedPutLegalHold,
|
||||
ObjectCreatedPutTagging, ObjectCreatedDeleteTagging,
|
||||
ObjectReplicationComplete, ObjectReplicationFailed,
|
||||
}
|
||||
case ObjectRemovedAll:
|
||||
return []Name{
|
||||
|
|
|
|||
|
|
@ -32,8 +32,7 @@ func TestNameExpand(t *testing.T) {
|
|||
{BucketRemoved, []Name{BucketRemoved}},
|
||||
{ObjectAccessedAll, []Name{ObjectAccessedGet, ObjectAccessedHead, ObjectAccessedGetRetention, ObjectAccessedGetLegalHold}},
|
||||
{ObjectCreatedAll, []Name{ObjectCreatedCompleteMultipartUpload, ObjectCreatedCopy, ObjectCreatedPost, ObjectCreatedPut,
|
||||
ObjectCreatedPutRetention, ObjectCreatedPutLegalHold, ObjectCreatedPutTagging, ObjectCreatedDeleteTagging,
|
||||
ObjectReplicationComplete, ObjectReplicationFailed}},
|
||||
ObjectCreatedPutRetention, ObjectCreatedPutLegalHold, ObjectCreatedPutTagging, ObjectCreatedDeleteTagging}},
|
||||
{ObjectRemovedAll, []Name{ObjectRemovedDelete, ObjectRemovedDeleteMarkerCreated}},
|
||||
{ObjectAccessedHead, []Name{ObjectAccessedHead}},
|
||||
}
|
||||
|
|
|
|||
Loading…
Reference in New Issue