feat(types): Embedding-Abschnitt der Server-Config (Provider, Failover-Gruppen, goldener Pfad) + Status-Payloads
index.d.ts: EmbeddingProviderConfig/EmbeddingGroupConfig (onSingleQuery, onBulkWrite, health), EmbeddingConfig für config.server().embedding (Roh-JSON von config.Server. Embedding) sowie die Status-Payloads von GET …/admin/embedding/status (Projekt- und Server-Scope, Zustände live|degraded|quarantined|offline, Breaker closed|open|half-open). VectorConfig.provider dokumentiert jetzt, dass dort auch ein Gruppenname stehen darf. schemas/config/embedding.schema.json: validierbares Schema des Server-Abschnitts (Enums für Policies/Distributionen, Go-Duration-Muster, Pflichtfelder, keine erfundenen Properties); Beispiel examples/server/embedding.yml, per scripts/validate-schemas.mjs mitgeprüft (npm test: 5 YAML gegen 17 Schemas). Quelle der Wahrheit: internal/config/server.go + internal/config/embedding_group.go (tibi-server). Typecheck: tsc --noEmit --strict über index.d.ts grün.
This commit is contained in:
1 parent
4c55f82446
commit
5a45c4dd3f
4 files changed
+427
No files matched your search
@@ -0,0 +1,46 @@
|
||||
# Beispiel für den embedding-Abschnitt der Server-config.yml
|
||||
# (tibi-server, server.embedding) — validiert gegen schemas/config/embedding.schema.json.
|
||||
providers:
|
||||
- name: bge-m3-external
|
||||
type: flagembedding
|
||||
model: BAAI/bge-m3
|
||||
dimensions: 1024
|
||||
baseURL: https://embed.intern/embed
|
||||
timeoutPerAttempt: 8s
|
||||
slowThreshold: 1500ms
|
||||
- name: bge-m3-local
|
||||
type: native
|
||||
model: BAAI/bge-m3
|
||||
dimensions: 1024
|
||||
modelPath: /models/bge-m3
|
||||
timeoutPerAttempt: 60s
|
||||
groups:
|
||||
- name: bge-m3-main
|
||||
description: Bevorzugt extern, sonst lokal, kehrt automatisch zurück.
|
||||
members:
|
||||
- bge-m3-external
|
||||
- bge-m3-local
|
||||
onSingleQuery:
|
||||
policy: prefer-in-order
|
||||
preferredOrder:
|
||||
- bge-m3-external
|
||||
- bge-m3-local
|
||||
timeoutPerAttempt: 8s
|
||||
timeoutTotal: 25s
|
||||
autoFailback: true
|
||||
retryAttempts: 1
|
||||
retryBackoff: 200ms
|
||||
retryBackoffMax: 2s
|
||||
onBulkWrite:
|
||||
distribution: lowest-latency
|
||||
maxConcurrentTotal: 6
|
||||
maxConcurrentPerEndpoint: 2
|
||||
timeoutPerAttempt: 15s
|
||||
timeoutTotal: 120s
|
||||
health:
|
||||
checkInterval: 30s
|
||||
checkTimeout: 3s
|
||||
unhealthyAfterFailures: 3
|
||||
healthyAfterSuccesses: 2
|
||||
cooldownAfterUnhealthy: 60s
|
||||
goldenPath: testdata/embedding-golden
|
||||
Reference in new issue
Block a user