hz4th_coder/llm-speed-tester
1
0 Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

v2.8.0 清空历史真正物理删除 + 请求超时可配修复慢接口被掐断

Browse Source 
1. 清空历史=真正清空（不再只是删行留痕）:
   - 清空前先停止所有运行中测试线程，避免其继续写入孤儿数据
   - DELETE 后重置自增序列(sqlite_sequence)，新测试 ID 从 1 开始
   - WAL checkpoint(TRUNCATE) + VACUUM 物理回收文件空间，数据不可恢复
   - 单条删除删光时同样重置自增序列

2. 慢接口被掐断修复（预处理/首字耗时过长 Read timed out）:
   - 读取超时默认 300s→1800s(30分钟)，连接 60s
   - 「速度测试配置」新增 请求超时(秒) 输入框(默认1800)，测试连接+速度采样共用，按次可调(如3600)
   - 超时错误改为友好提示：等待响应超时(超过N秒未收到数据)，引导调大请求超时，不再裸抛 Read timed out
   - 测试启动日志打印当前超时配置
   - 修复案例: 18008 Qwen3.8-FP8 131072 长上下文预填充被 300s 掐断
This commit is contained in:
hz4th_coder
2026-09-13 12:38:40 +08:00
parent 31e5bbb873
commit ccfd7eb689
7 changed files with 94 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
llm_providers.py
+39 -3
View File
@@ -116,7 +116,7 @@ def stream_openai(cfg, prompt, gen, log, should_stop=None):
if should_stop and should_stop():
raise StopRequested()
resp = requests.post(url, json=build_payload(), headers=headers, stream=True,
timeout=(config.CONNECT_TIMEOUT, config.STREAM_READ_TIMEOUT))
timeout=_timeouts(gen))
if resp.status_code == 200:
break
err = resp.text[:400]
@@ -152,6 +152,10 @@ def stream_openai(cfg, prompt, gen, log, should_stop=None):
cached_tokens = details.get("cached_tokens") or 0
except StopRequested:
raise
except requests.exceptions.ReadTimeout:
raise ProviderError(_read_timeout_err(gen))
except requests.exceptions.ConnectTimeout:
raise ProviderError(_connect_timeout_err(gen))
except Exception as e:
raise ProviderError("流式请求异常: %s" % e)
finally:
@@ -171,6 +175,30 @@ def _bad_stream_options(err: str):
or "additional properties" in err)
def _timeouts(gen):
"""从测试参数 gen 里取超时配置（秒），未配置则用全局默认。
返回 (connect, read) 元组，供 requests timeout 使用。"""
try:
connect = float(gen.get("connect_timeout") or config.CONNECT_TIMEOUT)
except (TypeError, ValueError):
connect = config.CONNECT_TIMEOUT
try:
read = float(gen.get("read_timeout") or config.STREAM_READ_TIMEOUT)
except (TypeError, ValueError):
read = config.STREAM_READ_TIMEOUT
return (max(connect, 5), max(read, 10))
def _read_timeout_err(gen):
"""读超时的友好报错：提示这是等待数据超时，可调大请求超时"""
return ("等待响应超时：超过 %.0f 秒未收到数据（接口预处理/首字耗时过长或网络过慢），"
"请在「速度测试配置」中调大“请求超时（秒）”" % _timeouts(gen)[1])
def _connect_timeout_err(gen):
return "连接超时：超过 %.0f 秒未建立连接（请检查地址/网络）" % _timeouts(gen)[0]
def _iter_json(resp):
"""解析 SSE data: 行，逐个返回 JSON 对象"""
for raw in resp.iter_lines(decode_unicode=True):
@@ -206,7 +234,7 @@ def stream_anthropic(cfg, prompt, gen, log, should_stop=None):
if should_stop and should_stop():
raise StopRequested()
resp = requests.post(url, json=payload, headers=headers, stream=True,
timeout=(config.CONNECT_TIMEOUT, config.STREAM_READ_TIMEOUT))
timeout=_timeouts(gen))
if resp.status_code != 200:
err = resp.text[:400]
resp.close()
@@ -233,6 +261,10 @@ def stream_anthropic(cfg, prompt, gen, log, should_stop=None):
output_tokens = usage.get("output_tokens") or output_tokens
except StopRequested:
raise
except requests.exceptions.ReadTimeout:
raise ProviderError(_read_timeout_err(gen))
except requests.exceptions.ConnectTimeout:
raise ProviderError(_connect_timeout_err(gen))
except Exception as e:
raise ProviderError("流式请求异常: %s" % e)
finally:
@@ -272,7 +304,7 @@ def stream_google(cfg, prompt, gen, log, should_stop=None):
if should_stop and should_stop():
raise StopRequested()
resp = requests.post(url, params=params, json=payload, headers=headers, stream=True,
timeout=(config.CONNECT_TIMEOUT, config.STREAM_READ_TIMEOUT))
timeout=_timeouts(gen))
if resp.status_code != 200:
err = resp.text[:400]
resp.close()
@@ -299,6 +331,10 @@ def stream_google(cfg, prompt, gen, log, should_stop=None):
cached_tokens = um.get("cachedContentTokenCount") or 0
except StopRequested:
raise
except requests.exceptions.ReadTimeout:
raise ProviderError(_read_timeout_err(gen))
except requests.exceptions.ConnectTimeout:
raise ProviderError(_connect_timeout_err(gen))
except Exception as e:
raise ProviderError("流式请求异常: %s" % e)
finally: