Send partition values instead of index name in OpenTableRequest #17079
Conversation
a3e01b1
to
4672b52
Compare
| import io.crate.metadata.RelationName; | ||
|
|
||
|
|
||
| public class OpenTableRequest extends AcknowledgedRequest<OpenTableRequest> { |
There was a problem hiding this comment.
It feels a bit strange that now a OpenTableRequest is also used on the CloseTableAction. Worth adding an empty CloseTableRequest extending on this? Or we call it just TableRequest?
There was a problem hiding this comment.
Where is it used? I don't see it in TransportCloseTable
There was a problem hiding this comment.
I'll go ahead with this because I'll need to make the same change for the CloseTableRequest anyway. But please comment so I can then adapt it as part of the other PR.
There was a problem hiding this comment.
Ah my fault, I saw it at the CloseTableClusterStateExecutor, but this is now unused, so all fine.
See #17081, I've removed unused class and merge the abstract one.
| String partitionIndexName = in.readOptionalString(); | ||
| partitionValues = PartitionName.fromIndexOrTemplate(partitionIndexName).values(); |
There was a problem hiding this comment.
We can have NPE here, PartitionName.fromIndexOrTemplate doesn't work with null values whereas partitionIndexName can be null.
I think we need a test for that
There was a problem hiding this comment.
Good catch. Added a fixup
Helps #16964 To support table swap or rename without having to close and open shards it's necessary that a index has no relation name association. That means index names will be replaced by UUIDs and to identify partitions we need to keep holding onto relationName + partition values or the index uuid.
509de45
to
94f20a5
Compare
Merge the abstract class in it's afterwards only implementation. Follow up of #17079
Merge the abstract class in it's afterwards only implementation. Follow up of #17079
Same as #17079 but for close
Same as #17079 but for close
Helps #16964
To support table swap or rename without having to close and open shards
it's necessary that a index has no relation name association.
That means index names will be replaced by UUIDs and to identify
partitions we need to keep holding onto relationName + partition values
or the index uuid.