Files
Aether/tests/api/handlers/base/test_stream_context.py

198 lines
6.6 KiB
Python
Raw Normal View History

feat: 流式空闲超时、健康监控查询优化、限流桶内存上限与维护清理修复 Close #233 Co-authored-by: AAEE86 <ppk0227@hotmail.com> - cli_monitor_mixin: 引入 STREAM_IDLE_TIMEOUT_SECONDS(可通过环境变量配置), 流传输开始后若超出空闲窗口无新 chunk 则提前取消并返回 504,避免长时间挂起 - stream_context: 新增 managed_recorded_bodies 上下文管理器,确保 chunks 在 telemetry 完成后及时释放;stream_telemetry 使用该接口统一管理 response body 构建 - health endpoint: 将状态聚合改为 GROUP BY 直接统计,事件列表按 api_format 单独查询,避免单次 limit 拉取大量记录导致的遗漏与性能问题;同时过滤不活跃 provider/endpoint,与公开健康接口保持一致 - endpoint health service: 修正时间线数据按 endpoint_id 而非 key_id 聚合 - token_bucket: 引入 max_buckets/bucket_expiry 上限与定时清理,防止内存无限增长; 修复 refill_rate=0 时 get_reset_time 除零异常;新增 _is_unlimited_rate_limit 判断 - maintenance_scheduler: 调整清理顺序(先删整行再按窗口清理),新增 newer_than 边界参数,避免同一行在同一轮中被重复改写 - sync_execute: 新增 create_pending_usage 开关,允许已预创建记录的调用方跳过重复创建 - quota_reader / provider_ops balance: 小幅修复与健壮性提升 - Dockerfile: 添加 MALLOC_ARENA_MAX=2 环境变量以降低 gunicorn worker RSS - 补充相关测试覆盖
2026-03-18 23:38:26 +08:00
import os
import pytest
os.environ.setdefault("JWT_SECRET_KEY", "test-secret-key")
from src.api.handlers.base import stream_context
from src.api.handlers.base.stream_context import StreamContext
def test_collected_text_append_and_property() -> None:
ctx = StreamContext(model="test-model", api_format="openai:chat")
assert ctx.collected_text == ""
assert ctx.collected_text_length == 0
ctx.append_text("hello")
ctx.append_text(" ")
ctx.append_text("world")
assert ctx.collected_text == "hello world"
assert ctx.collected_text_length == len("hello world")
def test_collected_text_is_capped_but_total_length_is_preserved() -> None:
ctx = StreamContext(model="test-model", api_format="openai:chat")
cap = stream_context._MAX_COLLECTED_TEXT_CHARS
ctx.append_text("a" * (cap - 4))
ctx.append_text("b" * 10)
assert len(ctx.collected_text) == cap
assert ctx.collected_text == ("a" * (cap - 4)) + ("b" * 4)
assert ctx.collected_text_length == cap + 6
def test_reset_for_retry_clears_state() -> None:
ctx = StreamContext(model="test-model", api_format="openai:chat")
ctx.append_text("x")
ctx.update_usage(input_tokens=10, output_tokens=5)
ctx.parsed_chunks.append({"type": "chunk"})
ctx.chunk_count = 3
ctx.data_count = 2
ctx.has_completion = True
ctx.status_code = 418
ctx.error_message = "boom"
ctx.reset_for_retry()
assert ctx.collected_text == ""
assert ctx.input_tokens == 0
assert ctx.output_tokens == 0
assert ctx.parsed_chunks == []
assert ctx.chunk_count == 0
assert ctx.data_count == 0
assert ctx.has_completion is False
assert ctx.status_code == 200
assert ctx.error_message is None
feat: 流式空闲超时、健康监控查询优化、限流桶内存上限与维护清理修复 Close #233 Co-authored-by: AAEE86 <ppk0227@hotmail.com> - cli_monitor_mixin: 引入 STREAM_IDLE_TIMEOUT_SECONDS(可通过环境变量配置), 流传输开始后若超出空闲窗口无新 chunk 则提前取消并返回 504,避免长时间挂起 - stream_context: 新增 managed_recorded_bodies 上下文管理器,确保 chunks 在 telemetry 完成后及时释放;stream_telemetry 使用该接口统一管理 response body 构建 - health endpoint: 将状态聚合改为 GROUP BY 直接统计,事件列表按 api_format 单独查询,避免单次 limit 拉取大量记录导致的遗漏与性能问题;同时过滤不活跃 provider/endpoint,与公开健康接口保持一致 - endpoint health service: 修正时间线数据按 endpoint_id 而非 key_id 聚合 - token_bucket: 引入 max_buckets/bucket_expiry 上限与定时清理,防止内存无限增长; 修复 refill_rate=0 时 get_reset_time 除零异常;新增 _is_unlimited_rate_limit 判断 - maintenance_scheduler: 调整清理顺序(先删整行再按窗口清理),新增 newer_than 边界参数,避免同一行在同一轮中被重复改写 - sync_execute: 新增 create_pending_usage 开关,允许已预创建记录的调用方跳过重复创建 - quota_reader / provider_ops balance: 小幅修复与健壮性提升 - Dockerfile: 添加 MALLOC_ARENA_MAX=2 环境变量以降低 gunicorn worker RSS - 补充相关测试覆盖
2026-03-18 23:38:26 +08:00
def test_release_recorded_chunks_clears_both_chunk_lists() -> None:
ctx = StreamContext(model="test-model", api_format="openai:chat")
ctx.parsed_chunks.append({"type": "client"})
ctx.provider_parsed_chunks.append({"type": "provider"})
ctx.release_recorded_chunks()
assert ctx.parsed_chunks == []
assert ctx.provider_parsed_chunks == []
def test_managed_recorded_bodies_builds_then_releases_chunks() -> None:
ctx = StreamContext(model="test-model", api_format="openai:chat")
ctx.parsed_chunks.append({"type": "client"})
ctx.provider_parsed_chunks.append({"type": "provider"})
ctx.data_count = 1
with ctx.managed_recorded_bodies(123) as recorded_bodies:
assert recorded_bodies.response_body is not None
assert recorded_bodies.response_body["chunks"] == [{"type": "provider"}]
assert recorded_bodies.client_response_body is not None
assert recorded_bodies.client_response_body["chunks"] == [{"type": "client"}]
assert ctx.parsed_chunks == []
assert ctx.provider_parsed_chunks == []
assert recorded_bodies.response_body is None
assert recorded_bodies.client_response_body is None
def test_record_first_byte_time(monkeypatch: pytest.MonkeyPatch) -> None:
"""测试记录首字时间"""
ctx = StreamContext(model="claude-3", api_format="claude_messages")
start_time = 100.0
monkeypatch.setattr(stream_context.time, "time", lambda: 100.0123) # 12.3ms
# 记录首字时间
ctx.record_first_byte_time(start_time)
# 验证首字时间已记录
assert ctx.first_byte_time_ms == 12
feat: 流式空闲超时、健康监控查询优化、限流桶内存上限与维护清理修复 Close #233 Co-authored-by: AAEE86 <ppk0227@hotmail.com> - cli_monitor_mixin: 引入 STREAM_IDLE_TIMEOUT_SECONDS(可通过环境变量配置), 流传输开始后若超出空闲窗口无新 chunk 则提前取消并返回 504,避免长时间挂起 - stream_context: 新增 managed_recorded_bodies 上下文管理器,确保 chunks 在 telemetry 完成后及时释放;stream_telemetry 使用该接口统一管理 response body 构建 - health endpoint: 将状态聚合改为 GROUP BY 直接统计,事件列表按 api_format 单独查询,避免单次 limit 拉取大量记录导致的遗漏与性能问题;同时过滤不活跃 provider/endpoint,与公开健康接口保持一致 - endpoint health service: 修正时间线数据按 endpoint_id 而非 key_id 聚合 - token_bucket: 引入 max_buckets/bucket_expiry 上限与定时清理,防止内存无限增长; 修复 refill_rate=0 时 get_reset_time 除零异常;新增 _is_unlimited_rate_limit 判断 - maintenance_scheduler: 调整清理顺序(先删整行再按窗口清理),新增 newer_than 边界参数,避免同一行在同一轮中被重复改写 - sync_execute: 新增 create_pending_usage 开关,允许已预创建记录的调用方跳过重复创建 - quota_reader / provider_ops balance: 小幅修复与健壮性提升 - Dockerfile: 添加 MALLOC_ARENA_MAX=2 环境变量以降低 gunicorn worker RSS - 补充相关测试覆盖
2026-03-18 23:38:26 +08:00
def test_record_first_byte_time_idempotent(monkeypatch: pytest.MonkeyPatch) -> None:
"""测试首字时间只记录一次"""
ctx = StreamContext(model="claude-3", api_format="claude_messages")
start_time = 100.0
# 第一次记录
monkeypatch.setattr(stream_context.time, "time", lambda: 100.010)
ctx.record_first_byte_time(start_time)
first_value = ctx.first_byte_time_ms
# 第二次记录(应该被忽略)
monkeypatch.setattr(stream_context.time, "time", lambda: 100.020)
ctx.record_first_byte_time(start_time)
second_value = ctx.first_byte_time_ms
# 验证值没有改变
assert first_value == second_value
feat: 流式空闲超时、健康监控查询优化、限流桶内存上限与维护清理修复 Close #233 Co-authored-by: AAEE86 <ppk0227@hotmail.com> - cli_monitor_mixin: 引入 STREAM_IDLE_TIMEOUT_SECONDS(可通过环境变量配置), 流传输开始后若超出空闲窗口无新 chunk 则提前取消并返回 504,避免长时间挂起 - stream_context: 新增 managed_recorded_bodies 上下文管理器,确保 chunks 在 telemetry 完成后及时释放;stream_telemetry 使用该接口统一管理 response body 构建 - health endpoint: 将状态聚合改为 GROUP BY 直接统计,事件列表按 api_format 单独查询,避免单次 limit 拉取大量记录导致的遗漏与性能问题;同时过滤不活跃 provider/endpoint,与公开健康接口保持一致 - endpoint health service: 修正时间线数据按 endpoint_id 而非 key_id 聚合 - token_bucket: 引入 max_buckets/bucket_expiry 上限与定时清理,防止内存无限增长; 修复 refill_rate=0 时 get_reset_time 除零异常;新增 _is_unlimited_rate_limit 判断 - maintenance_scheduler: 调整清理顺序(先删整行再按窗口清理),新增 newer_than 边界参数,避免同一行在同一轮中被重复改写 - sync_execute: 新增 create_pending_usage 开关,允许已预创建记录的调用方跳过重复创建 - quota_reader / provider_ops balance: 小幅修复与健壮性提升 - Dockerfile: 添加 MALLOC_ARENA_MAX=2 环境变量以降低 gunicorn worker RSS - 补充相关测试覆盖
2026-03-18 23:38:26 +08:00
def test_reset_for_retry_clears_first_byte_time(monkeypatch: pytest.MonkeyPatch) -> None:
"""测试重试时清除首字时间"""
ctx = StreamContext(model="claude-3", api_format="claude_messages")
start_time = 100.0
# 记录首字时间
monkeypatch.setattr(stream_context.time, "time", lambda: 100.010)
ctx.record_first_byte_time(start_time)
assert ctx.first_byte_time_ms is not None
# 重置
ctx.reset_for_retry()
# 验证首字时间已清除
assert ctx.first_byte_time_ms is None
def test_get_log_summary_with_first_byte_time() -> None:
"""测试日志摘要包含首字时间"""
ctx = StreamContext(model="claude-3", api_format="claude_messages")
ctx.provider_name = "anthropic"
ctx.input_tokens = 100
ctx.output_tokens = 50
ctx.first_byte_time_ms = 123
summary = ctx.get_log_summary("request-id-123", 456)
# 验证包含首字时间和总时间(大写格式)
assert "TTFB: 123ms" in summary
assert "Total: 456ms" in summary
assert "in:100 out:50" in summary
def test_get_log_summary_without_first_byte_time() -> None:
"""测试日志摘要在没有首字时间时的格式"""
ctx = StreamContext(model="claude-3", api_format="claude_messages")
ctx.provider_name = "anthropic"
ctx.input_tokens = 100
ctx.output_tokens = 50
# first_byte_time_ms 保持为 None
summary = ctx.get_log_summary("request-id-123", 456)
# 验证不包含首字时间标记,但有总时间(使用大写 TTFB 和 Total
assert "TTFB:" not in summary
assert "Total: 456ms" in summary
assert "in:100 out:50" in summary
def test_ensure_estimated_output_tokens_uses_collected_text() -> None:
ctx = StreamContext(model="test-model", api_format="openai:chat")
ctx.append_text("partial output")
changed = ctx.ensure_estimated_output_tokens()
assert changed is True
assert ctx.output_tokens == max(1, len("partial output") // 4)
def test_should_estimate_incomplete_tokens_for_interrupted_partial_stream() -> None:
ctx = StreamContext(model="test-model", api_format="openai:chat")
ctx.status_code = 503
ctx.chunk_count = 3
assert ctx.should_estimate_incomplete_tokens() is True
def test_should_estimate_incomplete_tokens_when_output_already_estimated() -> None:
"""ensure_estimated_output_tokens 已补了 output但 input 仍为 0 时仍需估算。"""
ctx = StreamContext(model="test-model", api_format="openai:chat")
ctx.status_code = 503
ctx.chunk_count = 3
ctx.append_text("partial")
ctx.ensure_estimated_output_tokens()
assert ctx.output_tokens > 0
assert ctx.input_tokens == 0
assert ctx.should_estimate_incomplete_tokens() is True