S3 跨区域复制 - 反向复制

Question

I currently have replication configured such that my S3 documents on us-east-1 are replicated to a bucket on us-west-2. In light of today's (gasp) AWS outage, I considered failing over to us-west-2 (which appears to be online at the moment). So, I have several questions about this:

1. Would documents uploaded to us-west-2 be replicated back to us-east-1 once services are restored? I suspect the answer is no since I have not found any documentation regarding bi-directional replication.
2. If bi-directional replication does not happen and I decide to failover to us-west-2, what's the process for recovering once us-east-1 comes back online? I assume this would require writing a script to copy all missing documents back to us-east-1. Any other ideas or suggestions?

Answer 1

The issue with bi-directional replication always boils down to "who has the correct version" of data. It's not so hard right now when there is only one source of data but what about when there are two? If one process writes to one region and another writes the same thing to another, what adjudicates the conflict?

This is why you don't see documentation about bi-directional replication - it's hard to solve in a generic way. I would look at something like s3 sync or a custom script to get the buckets back in sync.

Answer 2

We setup cross region mutual replication rules on both regions so that they could replicate each other. Seems to be working fine over the years.

The reason it does not do infinite loop is the principal is different for upload and replication. If you listen to your s3 event in Lambda , this is more obvious:

For original upload, it is:

            "userIdentity": {
                "principalId": "AWS:xxx:YourBucketName"
            },

For replication, it is

            "userIdentity": {
                "principalId": "AWS:xxx:s3-replication"
            },

As for the replication itself internally, quote:

For an object uploaded by you
    Amazon S3 triggers the rule you configured to replicate it to another bucket
    And sets Replication status to COMPLETED
For an object replicated from another bucket
    Amazon S3 knows not to re-replicate the object
    And sets Replication status to REPLICA

Answer 3

Implementing Bi-Directional Cross-Region Replication for Amazon Simple Storage Service (Amazon S3)

https://wellarchitectedlabs.com/Reliability/200_Bidirectional_Replication_for_S3/README.html