Description
When increasing the replica count, without increasing the partition count, the service currently ingests from DCP and re-builds the partition.
This can be avoided if the new replica partition transfers files from existing replica partitions, similar to file transfer rebalance.
While this is applicable to all types of indexes, it will be especially important in case of vector indexes where each re-training/reconstruction, etc. are expensive and can be avoided.
Attachments
Issue Links
- relates to
-
MB-60727 FTS re-ingest data on dropping replicas from 1 -> 0
- Resolved