34 lines
1.1 KiB
Markdown
34 lines
1.1 KiB
Markdown
# Runtime split – Patch A
|
||
|
||
This patch introduces the runtime split scaffolding without performing the full behavioral cutover.
|
||
|
||
## Added
|
||
- `dip.runtime.mode` with values `LEGACY` / `NEW`
|
||
- `@ConditionalOnRuntimeMode`
|
||
- `DipSearchProperties` (`dip.search.*`)
|
||
- `LegacyTedProperties` (`legacy.ted.*`) as migration scaffold
|
||
- example `application-legacy.yml` and `application-new.yml`
|
||
|
||
## First gated beans
|
||
|
||
### LEGACY
|
||
- `GenericVectorizationRoute`
|
||
- `DocumentEmbeddingProcessingService`
|
||
- `ConfiguredEmbeddingModelStartupRunner`
|
||
- `GenericVectorizationStartupRunner`
|
||
|
||
### NEW
|
||
- `EmbeddingSubsystemStartupValidator`
|
||
- `PgVectorSemanticSearchEngine`
|
||
- `GenericSearchController`
|
||
|
||
## Intentional limitation of Patch A
|
||
This patch does **not** yet switch import/runtime orchestration to the new embedding job flow.
|
||
It only establishes the explicit runtime mode infrastructure and starts separating bean graphs.
|
||
|
||
## Follow-up patch (Patch B)
|
||
- add a NEW-mode embedding job scheduler
|
||
- make generic import enqueue new embedding jobs
|
||
- disable legacy vector route in NEW mode operationally
|
||
- move active search tuning from `ted.search.*` to `dip.search.*`
|