A full re-index (e.g. embedding-model change, schema change) needs a zero-downtime strategy — build into a new index and alias-cutover rather than mutating the live one.
Tracked as an ADR Open Question — rationale, constraints, and direction live there: https://github.com/openmrs/openmrs-module-querystore/blob/main/docs/adr.md#re-index--alias-strategy. Filed from the querystore roadmap survey; this issue is the GitHub-side tracker.
A full re-index (e.g. embedding-model change, schema change) needs a zero-downtime strategy — build into a new index and alias-cutover rather than mutating the live one.
Tracked as an ADR Open Question — rationale, constraints, and direction live there: https://github.com/openmrs/openmrs-module-querystore/blob/main/docs/adr.md#re-index--alias-strategy. Filed from the querystore roadmap survey; this issue is the GitHub-side tracker.