mirror of
https://github.com/Ed1s0nZ/CyberStrikeAI.git
synced 2026-10-20 06:27:53 +02:00
Merge upstream main and resolve context memory asset conflicts
This commit is contained in:
Sycun committed
commit
0fb5484657
45 files changed
+1699
-74
No files matched your search
@@ -0,0 +1,57 @@
|
||||
# Task timeout classification fix
|
||||
|
||||
## F0 — Intake and gates
|
||||
|
||||
- Branch: `codex/task-timeout-classification`; base: `deda642b`, clean and synchronized with origin/main on 2026-10-15.
|
||||
- Workflow: P10 feedback → P8 backend correction → P9 verification. User authorized fixing the diagnosed bug and merging main.
|
||||
- Existing evidence: customer screenshot and local handler analysis; customer deployed revision and raw error remain unknown.
|
||||
- Workflow gate: implementation allowed. No UI/API/schema redesign required; required upstream contract is the existing task deadline and user-approved correction.
|
||||
- Maintainability gate: narrow_fix; handlers are 318–673 lines, mix orchestration and persistence, moderate coupling. No broad refactor needed. Extract one small timeout predicate and cover context semantics and persisted batch outcome.
|
||||
- Scope: single-agent, multi-agent, workflow, and batch task deadline classification. Unrelated robot request handling remains outside scope.
|
||||
|
||||
## F1 — Confirmed requirements
|
||||
|
||||
Confirmed by the user's request to fix the identified defect and merge main.
|
||||
|
||||
- R1: A live task must not be classified as globally timed out solely because a child operation returns DeadlineExceeded.
|
||||
- R2: An expired task deadline must retain the existing timeout behavior, even if the returned error is wrapped or unrelated.
|
||||
- R3: Cancellation must not be reclassified as deadline expiry, including cancellation whose cause wraps DeadlineExceeded.
|
||||
- R4: Keep existing failure diagnostics, bounded execution limits, and cancellation behavior. This fix does not promise automatic continuation after every local failure or blindly replay tools.
|
||||
- Acceptance: regression tests distinguish child timeout, task expiry, cancellation and wrapped HTTP errors; affected packages pass; main includes the reviewed change.
|
||||
|
||||
## F2 — Design
|
||||
|
||||
Introduce a package-local predicate using the task context's Err(), which reports DeadlineExceeded only when that context's deadline expires. Do not use the returned operation error or an arbitrary cancellation cause as evidence of task budget exhaustion.
|
||||
|
||||
Replace the four task-level OR checks with this predicate. Child timeouts follow the existing failed/error path with original error diagnostics and existing trace persistence. Real task expiry follows the unchanged timeout path. No new event schema or retry policy. No change to the 600-minute interactive or 6-hour batch budgets.
|
||||
|
||||
## F3 — Implementation plan
|
||||
|
||||
1. Add the shared predicate and update four handler call sites.
|
||||
2. Add deterministic context tests (past deadlines, child contexts, cancellation causes) plus batch handler regression coverage for status/persistence.
|
||||
3. Run targeted regression tests, affected package suites, and git diff --check; review the diff and update CHANGELOG.md.
|
||||
4. Commit and push each phase; merge only after checks pass and remote main is rechecked. No release or deployment requested.
|
||||
|
||||
Rollback: revert the fix commit; no data migration. Customer-specific fault attribution remains a separate evidence gap.
|
||||
|
||||
## F4 — Implementation
|
||||
|
||||
Added `taskDeadlineExceeded`, based on task context Err rather than operation error or cancellation cause. Single-agent, multi-agent, role workflow and batch task paths now use it. Existing terminal error details, trace persistence and retry policy are unchanged.
|
||||
|
||||
Regression proof: before the fix, `TestBatchTaskDeadlineClassification` failed for both child DeadlineExceeded and wrapped HTTP timeout (got timeout, wanted failed). After the fix, targeted tests pass, including real expiry, unrelated returned error after expiry, custom deadline causes and user cancellation.
|
||||
|
||||
## F5 — Verification
|
||||
|
||||
- `go test ./internal/handler -run 'Test(BatchTaskDeadlineClassification|TaskDeadlineExceededContextOwnership)' -count=1`: passed after previously demonstrating the defect.
|
||||
- `go test ./internal/handler ./internal/multiagent ./internal/workflow`: all three packages passed (17.789s, 17.230s, 2.545s).
|
||||
- `git diff --check`: passed.
|
||||
- Batch regression verifies finish status, queue failure and preservation of the original diagnostic in SQLite messages; context tests verify child timeout isolation, deadline expiry and cancellation semantics.
|
||||
- No customer environment reproduction or deployment performed. The customer-specific original timeout source is still unknown; the deterministic classification defect is fixed.
|
||||
|
||||
## F6 — Review and merge readiness
|
||||
|
||||
- Reviewed all four call sites: each receives the task context created for its own execution budget; checks occur before timeout cleanup. Local errors now follow existing failed/error paths.
|
||||
- Actual deadline expiry remains recognized even with a custom cause; a cancellation carrying a child timeout is no longer treated as exhausted task budget.
|
||||
- Scope is four predicate substitutions plus a small shared helper and regression tests. No dependency, schema, public API, replay policy or task budget change.
|
||||
- Added Unreleased/Fixed changelog entry. Implementation commit: `5ff781bd`; verification commit: `54213f82`.
|
||||
- Merge is explicitly authorized by the user. Plan: re-fetch origin, fast-forward main if unchanged, push without force and verify remote equality. No deployment or release tag.
|
||||
@@ -0,0 +1,68 @@
|
||||
# Token 用量监控：调研与实现
|
||||
|
||||
调研时间：2026-10-11。
|
||||
|
||||
## 调研结论
|
||||
|
||||
| 参考 | 官方能力 | 本项目采用的设计 |
|
||||
| --- | --- | --- |
|
||||
| [ZCode 使用统计](https://zcode.z.ai/cn/docs/usage-stats) | 本地会话统计与远端编程套餐额度分别展示；支持时间范围、日趋势和模型占比 | 首页增加可展开的本地用量面板，提供 7/30/90 天、日趋势、模型占比与明细 |
|
||||
| [DeepSeek 上下文缓存](https://api-docs.deepseek.com/guides/kv_cache/) | 响应 usage 的 `prompt_cache_hit_tokens` / `prompt_cache_miss_tokens` 区分输入缓存命中和未命中 | 将命中字段补入 OpenAI 标准缓存读取字段；不把未命中错误映射为缓存写入 |
|
||||
| [Eino v0.9.21](https://github.com/cloudwego/eino/releases/tag/v0.9.21)、[schema PR #1309](https://github.com/cloudwego/eino/pull/1309) | 稳定版新增 `PromptTokenDetails.CacheWriteTokens`，原有缓存读取、输入、输出、推理字段继续保留 | 使用稳定版的标准字段贯穿流式归并、运行汇总、数据库和 UI |
|
||||
| [Eino-ext PR #1015](https://github.com/cloudwego/eino-ext/pull/1015) | 模型适配器传递缓存写入计数 | 升级 agenticclaude 0.1.7、agenticopenai 0.2.4 与对应 OpenAI ACL，而非仅升级框架 |
|
||||
| [Eino 回调与 Trace](https://www.cloudwego.io/docs/eino/quick_start/chapter_06_callback_and_trace/) | 组件回调可收集模型生命周期、用量与耗时 | 保留现有运行事件采集；未来要覆盖所有内部调用，应另外设计统一模型调用账本 |
|
||||
|
||||
版本已通过 Go module 解析和本地源码确认；未采用 0.10 alpha。DeepSeek 原始字段当前不会自动进入 Eino 的标准缓存读取计数，因此在项目现有兼容 HTTP 层补齐，覆盖普通 JSON 和 SSE 流式响应。已有标准 `cached_tokens` 优先，未知字段使用 RawMessage 保存，避免浮点转换损失数字精度。
|
||||
|
||||
## 统计口径
|
||||
|
||||
- `promptTokens`：模型返回的全部输入。
|
||||
- `completionTokens`：全部输出，包括供应商计入其中的推理。
|
||||
- `totalTokens`：使用供应商的总量；缺省时由输入加输出补齐。
|
||||
- `cachedTokens`：输入中的缓存读取部分。
|
||||
- `cacheWriteTokens`：输入中的缓存写入部分。
|
||||
- `reasoningTokens`：输出中的推理部分。
|
||||
- 缓存读取、写入、推理都是细分，不能再加到总量。缓存读取占输入的比例为 `cachedTokens / promptTokens`；没有输入时显示破折号。
|
||||
- 一个流式模型输出可能多次返回累计计数，沿用按字段取最大值的归并方式，整条输出结束后只加入一次运行汇总。部分输出发生错误时，仍可保留已返回用量。
|
||||
|
||||
DeepSeek 的未命中输入不等于 Claude 的缓存写入。不能根据文本长度推导账单级计数，也不能根据缺失字段断言供应商未使用缓存。旧记录的新增字段默认为 0，表示未记录该细分。
|
||||
|
||||
## 功能入口与数据流
|
||||
|
||||
点击首页「Token 用量」卡片进入「系统设置 → Token 用量」，也可直接从设置二级菜单打开。首页卡片显示近 7 天总用量，副标仅显示缓存命中率（缓存读取 Token / 输入 Token），调用次数和每日用量保留在详情页。支持当前项目范围、7/30/90 天切换、八项指标、最多 20 项运行模型排行、每日趋势与完整日明细、最多 20 项最近运行。详情页独立加载，支持手动刷新与请求取消，切换周期不会刷新整个仪表盘；刷新页面可保持该设置分区。
|
||||
|
||||
详情页采用四项核心指标（总量、输入、输出、缓存读取占输入）与四项次级指标（调用、缓存读取、缓存写入、推理）的层级布局。趋势可切换总 Token、模型调用与缓存命中率，鼠标悬停或键盘聚焦柱形可查看日期与精确数字；切换图表指标不再请求 API。缓存环形图拆分缓存读取与其余输入，不把缓存或推理再次叠加到总量。模型明细展示占比条，每日数字明细默认折叠，最近运行区分完整与部分记录。统计口径可展开查看。
|
||||
|
||||
每日图表沿用 API 返回的服务端日期，只补齐已返回日期之间无记录的空档，不根据浏览器时区猜测查询终点。小屏图表和宽表在自己的区域内横向滚动。没有记录和无法获取统计分别展示空状态与不可用状态。
|
||||
|
||||
### UI 调研依据（2026-10-12）
|
||||
|
||||
| 参考 | 采用的设计 |
|
||||
| --- | --- |
|
||||
| [DeepSeek Context Caching](https://api-docs.deepseek.com/guides/kv_cache/) | 使用其明确的缓存命中/未命中输入定义，设计缓存构成视图；比例按 Token 而非请求次数计算 |
|
||||
| [LiteLLM Usage](https://docs.litellm.ai/docs/proxy/ui/chat_ui) | 周期选择、核心统计与使用日志分层；本项目保留已有数据范围，不新增无法准确计量的费用或成功率 |
|
||||
| [Langfuse Custom Dashboards](https://langfuse.com/docs/metrics/features/custom-dashboards) | 将汇总、时间趋势、模型分布拆为独立可读区域，并让趋势指标可切换 |
|
||||
|
||||
DeepSeek 和 GLM 控制台的完整页面本轮未能公开读取，因此本轮视觉设计以能核实的开源面板为依据。没有声称复刻它们当前登录后的界面。项目也没有新增供应商账单、预算或节省金额推算。
|
||||
|
||||
运行事件 → 流式/非流式用量累加 → `eino_usage_summary` → `process_details` 与 `model_token_usage` → 现有 `/api/usage/tokens`、对话用量接口 → 首页与聊天摘要。
|
||||
|
||||
新增 `model_token_usage.cache_write_tokens INTEGER NOT NULL DEFAULT 0`。启动时现有声明式 SQLite schema reconciliation 自动补列；旧数据库无需手工执行迁移。现有事件 ID 幂等写入、外键删除关联和历史回填继续生效。日分组按服务本地时区显示，与首页「今日」边界一致。
|
||||
|
||||
面板采用文本节点渲染供应商模型名称，表格在手机上局部横向滚动，提供中文、英文、俄文标签。聊天轮次 tooltip、主对话与 WebShell 时间线均增加缓存写入字段。
|
||||
|
||||
## 当前边界与后续方向
|
||||
|
||||
统计基于已记录的 Agent 输出事件，包括已有采集链路中的子 Agent，**不是全站 API 请求账本**。摘要压缩、检索 embedding、隐藏的模型调用、没有返回 usage 的请求不能由当前链路完整覆盖；因请求取消丢失最终 usage 的消耗也无法还原。
|
||||
|
||||
模型排行按运行配置的模型名称归类，发生 failover 的运行可能混合多个模型，不能据此核算逐模型账单。近期列表每项是运行汇总，不是单次 HTTP 请求；模型调用次数也仅统计返回了非空 usage 的输出。
|
||||
|
||||
本次不展示费用或套餐剩余额度。自定义中转渠道和订阅套餐的价格、计费方式不同，token 乘固定单价无法可靠代表费用。下一步如需要完整成本管理，应先建设按请求记录的账本：调用来源、真实供应商与模型、成功/错误/取消、耗时、TTFT、usage 是否返回、价格版本和币种；统一覆盖主/子 Agent、压缩、重试、embedding，单独对接供应商额度接口。
|
||||
|
||||
## 验证
|
||||
|
||||
- `go test ./...`：全仓通过。
|
||||
- `node --test web/static/js/*.test.cjs`：295 项通过。
|
||||
- 针对缓存写入的保存/分组/近期记录、旧 schema 补列、用户可见范围、流式去重和总量回退添加回归测试。
|
||||
- DeepSeek 普通 JSON/SSE 的缓存字段转换、标准字段优先、错误/终止块保持、超大数值精度验证通过。
|
||||
- 浏览器使用合成数据验证桌面、390px 手机、深色模式、模型名称转义、空数据与接口不可用状态；没有读取真实运行数据库，也没有调用付费模型。
|
||||
Reference in new issue
Block a user