S3 objects are now flagged when they may require cleanup #59
Conversation
|
@sjones4 I'm curious about how object metadata is implemented. Is it stored separately from object data in a key/value store? Do metadata properties have an index that allows the bucket reaper task to look up objects that have the flag set, or does the task still need to iterate through all of the keys when looking for that flag? |
|
@cdonati the object storage gateway service does not store any object data, it is just a front end. The data is stored by walrus or ceph (s3). The object storage gateway object metadata is persisted in postgres. This flag is added to the main object entity table so although a full table scan may be used it is still much more efficient than the current approach (which performs some processing for every object) |
|
Cool. Thanks for the explanation. At some point, I'd love to hear more about the implementation details during a sprint planning session or something. I would be interested in learning about how the metadata is kept consistent with the underlying object data since that would be relevant for our GCS implementation. |
YAML format support for network and region configuration
This pull request adds a new flag to s3 object entities to indicate when version clean up may be needed. When an object version is set as the latest (i.e. version added or removed) we now also set the "cleanup required" flag. The clean up flag is only meaningful on the latest version of an object.
The bucket reaper task now performs version clean up only for these flagged objects rather than all objects in a bucket. This task clears the clean up flag after processing an object, allowing progress to be made even when there are many objects to be checked.
It is generally the case that periodic clean up is not required, as clean up is performed when updates to an object are made. Having a flag allows for manual triggering of object version clean up (via database change) should it ever be necessary.