fix(openai): quarantine proxies after stream disconnects

This commit is contained in:
Heatherm Huang
2026-07-23 00:12:28 +08:00
parent 63cef60594
commit 47ad29db3e
17 changed files with 562 additions and 1 deletions
+3
View File
@@ -287,6 +287,9 @@ GATEWAY_OPENAI_HTTP2_ALLOW_PROXY_FALLBACK_TO_HTTP1=true
GATEWAY_OPENAI_HTTP2_FALLBACK_ERROR_THRESHOLD=2
GATEWAY_OPENAI_HTTP2_FALLBACK_WINDOW_SECONDS=60
GATEWAY_OPENAI_HTTP2_FALLBACK_TTL_SECONDS=600
GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_FAILURE_THRESHOLD=2
GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_WINDOW_SECONDS=60
GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_TTL_SECONDS=600
# 上游连接池:每主机最大连接数(默认 1024;流式/HTTP1.1 场景可调大,如 2400/4096)
GATEWAY_MAX_CONNS_PER_HOST=2048
# 上游连接池:最大空闲连接总数(默认 2560;账号/代理隔离 + 高并发场景可调大)
+5
View File
@@ -401,6 +401,11 @@ gateway:
fallback_error_threshold: 2
fallback_window_seconds: 60
fallback_ttl_seconds: 600
# OpenAI Responses SSE 代理断流熔断;按 proxy_id 跨账号隔离,仅影响下一次调度。
openai_proxy_stream_circuit:
failure_threshold: 2
window_seconds: 60
ttl_seconds: 600
# HTTP upstream connection pool settings (HTTP/2 + multi-proxy scenario defaults)
# HTTP 上游连接池配置(HTTP/2 + 多代理场景默认值)
# Max idle connections across all hosts
+3
View File
@@ -58,6 +58,9 @@ services:
- GATEWAY_OPENAI_HTTP2_FALLBACK_ERROR_THRESHOLD=${GATEWAY_OPENAI_HTTP2_FALLBACK_ERROR_THRESHOLD:-2}
- GATEWAY_OPENAI_HTTP2_FALLBACK_WINDOW_SECONDS=${GATEWAY_OPENAI_HTTP2_FALLBACK_WINDOW_SECONDS:-60}
- GATEWAY_OPENAI_HTTP2_FALLBACK_TTL_SECONDS=${GATEWAY_OPENAI_HTTP2_FALLBACK_TTL_SECONDS:-600}
- GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_FAILURE_THRESHOLD=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_FAILURE_THRESHOLD:-2}
- GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_WINDOW_SECONDS=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_WINDOW_SECONDS:-60}
- GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_TTL_SECONDS=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_TTL_SECONDS:-600}
- GATEWAY_IMAGE_STREAM_DATA_INTERVAL_TIMEOUT=${GATEWAY_IMAGE_STREAM_DATA_INTERVAL_TIMEOUT:-900}
- GATEWAY_IMAGE_STREAM_KEEPALIVE_INTERVAL=${GATEWAY_IMAGE_STREAM_KEEPALIVE_INTERVAL:-10}
- GATEWAY_IMAGE_CONCURRENCY_ENABLED=${GATEWAY_IMAGE_CONCURRENCY_ENABLED:-false}
+3
View File
@@ -166,6 +166,9 @@ services:
- GATEWAY_OPENAI_HTTP2_FALLBACK_ERROR_THRESHOLD=${GATEWAY_OPENAI_HTTP2_FALLBACK_ERROR_THRESHOLD:-2}
- GATEWAY_OPENAI_HTTP2_FALLBACK_WINDOW_SECONDS=${GATEWAY_OPENAI_HTTP2_FALLBACK_WINDOW_SECONDS:-60}
- GATEWAY_OPENAI_HTTP2_FALLBACK_TTL_SECONDS=${GATEWAY_OPENAI_HTTP2_FALLBACK_TTL_SECONDS:-600}
- GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_FAILURE_THRESHOLD=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_FAILURE_THRESHOLD:-2}
- GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_WINDOW_SECONDS=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_WINDOW_SECONDS:-60}
- GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_TTL_SECONDS=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_TTL_SECONDS:-600}
- GATEWAY_IMAGE_STREAM_DATA_INTERVAL_TIMEOUT=${GATEWAY_IMAGE_STREAM_DATA_INTERVAL_TIMEOUT:-900}
- GATEWAY_IMAGE_STREAM_KEEPALIVE_INTERVAL=${GATEWAY_IMAGE_STREAM_KEEPALIVE_INTERVAL:-10}
- GATEWAY_IMAGE_CONCURRENCY_ENABLED=${GATEWAY_IMAGE_CONCURRENCY_ENABLED:-false}
+3
View File
@@ -113,6 +113,9 @@ services:
- GATEWAY_OPENAI_HTTP2_FALLBACK_ERROR_THRESHOLD=${GATEWAY_OPENAI_HTTP2_FALLBACK_ERROR_THRESHOLD:-2}
- GATEWAY_OPENAI_HTTP2_FALLBACK_WINDOW_SECONDS=${GATEWAY_OPENAI_HTTP2_FALLBACK_WINDOW_SECONDS:-60}
- GATEWAY_OPENAI_HTTP2_FALLBACK_TTL_SECONDS=${GATEWAY_OPENAI_HTTP2_FALLBACK_TTL_SECONDS:-600}
- GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_FAILURE_THRESHOLD=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_FAILURE_THRESHOLD:-2}
- GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_WINDOW_SECONDS=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_WINDOW_SECONDS:-60}
- GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_TTL_SECONDS=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_TTL_SECONDS:-600}
- GATEWAY_IMAGE_STREAM_DATA_INTERVAL_TIMEOUT=${GATEWAY_IMAGE_STREAM_DATA_INTERVAL_TIMEOUT:-900}
- GATEWAY_IMAGE_STREAM_KEEPALIVE_INTERVAL=${GATEWAY_IMAGE_STREAM_KEEPALIVE_INTERVAL:-10}
- GATEWAY_IMAGE_CONCURRENCY_ENABLED=${GATEWAY_IMAGE_CONCURRENCY_ENABLED:-false}
+3
View File
@@ -162,6 +162,9 @@ services:
- GATEWAY_OPENAI_HTTP2_FALLBACK_ERROR_THRESHOLD=${GATEWAY_OPENAI_HTTP2_FALLBACK_ERROR_THRESHOLD:-2}
- GATEWAY_OPENAI_HTTP2_FALLBACK_WINDOW_SECONDS=${GATEWAY_OPENAI_HTTP2_FALLBACK_WINDOW_SECONDS:-60}
- GATEWAY_OPENAI_HTTP2_FALLBACK_TTL_SECONDS=${GATEWAY_OPENAI_HTTP2_FALLBACK_TTL_SECONDS:-600}
- GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_FAILURE_THRESHOLD=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_FAILURE_THRESHOLD:-2}
- GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_WINDOW_SECONDS=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_WINDOW_SECONDS:-60}
- GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_TTL_SECONDS=${GATEWAY_OPENAI_PROXY_STREAM_CIRCUIT_TTL_SECONDS:-600}
- GATEWAY_IMAGE_STREAM_DATA_INTERVAL_TIMEOUT=${GATEWAY_IMAGE_STREAM_DATA_INTERVAL_TIMEOUT:-900}
- GATEWAY_IMAGE_STREAM_KEEPALIVE_INTERVAL=${GATEWAY_IMAGE_STREAM_KEEPALIVE_INTERVAL:-10}
- GATEWAY_IMAGE_CONCURRENCY_ENABLED=${GATEWAY_IMAGE_CONCURRENCY_ENABLED:-false}