fix(infra): remove buffer/compress from llm-api for real SSE streaming
- llm-api uses llm-chain instead of common-chain - llm-chain excludes buffer & compress middlewares - Prevents Traefik from buffering streaming SSE responses - Each token is now flushed immediately
This commit is contained in:
@@ -27,7 +27,7 @@ http:
|
||||
- websecure
|
||||
tls: {}
|
||||
middlewares:
|
||||
- common-chain@file
|
||||
- llm-chain@file
|
||||
service: llm-api-service
|
||||
|
||||
# ── Tools (Document Scanner & Media Processing) ──
|
||||
|
||||
@@ -61,6 +61,14 @@ http:
|
||||
- "^/wp-login\\.php"
|
||||
- "^/config\\.php"
|
||||
|
||||
# ── LLM Stream Chain (no buffer/compress — SSE needs real-time) ──
|
||||
llm-chain:
|
||||
chain:
|
||||
middlewares:
|
||||
- secure-headers
|
||||
- retry
|
||||
- rate-limit
|
||||
|
||||
# ── Common Chain ──
|
||||
common-chain:
|
||||
chain:
|
||||
|
||||
Reference in New Issue
Block a user