Generic webhook template doesn't identify which metric attribute/label combination (e.g. trace name) triggered the alert

未关闭
#2,670 0 条评论 2 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
5/5
预计耗时
一周以上
新手友好度
45/100
Issue 类型
功能
描述清晰度
基本清楚
活跃度
冷清
技术栈
typescript
领域
observability

调研方向

该 issue 未指出文件或测试。首先定位通用 webhook 模板处理、指标告警评估以及 External API v2 的告警历史路径;跟踪隐式和显式序列标签如何一路传递到 firing。完成的标准是:触发越界的属性组合可在 webhook 模板和告警历史响应中使用,并且已定义一次或多次 webhook 投递时的行为。

由索引模型根据 Issue 内容生成。

描述

external

My alert is on a dashboard tile with a single series: a metrics data source, a gauge aggregation, and a where clause — no groupBy is configured anywhere in the tile. Despite that, the chart renders multiple lines, because the underlying OTel metric carries multiple distinct attribute/label combinations (e.g. per trace/span name), and HyperDX implicitly plots one line per unique combination even though I never set an explicit group-by.

When the alert fires, the generic webhook body currently supports {{title}}, {{body}}, {{link}}, {{state}}, {{startTime}}, {{endTime}}, and {{eventId}}. None of these identify which underlying attribute combination (e.g. which trace name) actually crossed the threshold — the payload describes the alert's overall state and timing, but has no reference to the specific label values that produced the offending line.

This makes it impossible to route or triage generic webhook alerts without manually opening the dashboard and visually picking out which of the several implicit lines crossed the threshold. For downstream automation (custom incident routing, deduping, auto-annotating tickets, posting to the right team channel, etc.) this is a hard blocker — the alert is not attributable to a specific cause without a human eyeballing the chart.

Current behavior
  • Dashboard tile: single series, metrics data source, gauge value, where clause. No groupBy configured.
  • The metric itself has multiple attribute/label combinations (e.g. per-trace or per-span name), so the tile implicitly renders several lines even without an explicit group-by.
  • Alert is attached to this tile/series.
  • On firing, the generic webhook body supports {{title}}, {{body}}, {{link}}, {{state}}, {{startTime}}, {{endTime}}, and {{eventId}}.
  • None of these resolve to the specific attribute/label combination (e.g. trace name) whose line crossed the threshold — {{link}} sends you to the tile, and the rest describe the alert's state/timing/ID, but none carry the identity of the offending line. You still have to visually inspect the chart to identify it.
Expected behavior

The webhook payload/template should include the specific attribute/label combination that breached the threshold for that firing — e.g. a {{seriesLabels}} or {{attributeValue}} (name TBD) template variable, populated with the trace name / span name / other metric attribute value that produced the crossing line.

This is closely related to (and possibly the same underlying gap as) groupBy-based alerts not surfacing the specific group value that fired — in both cases, the tile can implicitly or explicitly contain multiple lines, and the webhook has no way to say which one triggered.

Why this matters

Without this, generic webhook consumers (custom bots, incident routing via generic webhook, internal Slack bots, etc.) can only say "something on this metric tile crossed a threshold" — not "trace X crossed the threshold." For any metric with multiple attribute combinations feeding one tile, this makes the alert nearly useless for automated triage; a human has to click through and visually inspect the chart every time just to know what actually needs attention.

Suggested implementation
  • Add a template variable (e.g. {{seriesLabels}}) available in the generic webhook body template, populated per-firing with the specific attribute/label combination (as key-value pairs) of the line that breached.
  • Since a metric-source tile with no explicit groupBy can still implicitly fan out into many lines by attribute combination, consider either:
    • firing one webhook call per breaching attribute combination (each with its own labels), or
    • including an array of breaching attribute combinations in the payload if a single call is preferred.
  • Apply the equivalent fix for explicit groupBy-based alerts (surface the specific group value that fired, not just the field being grouped on) — same underlying gap, different mechanism for producing multiple lines.
  • Expose the same field on the External API v2 alert-history/evaluation response, if separate from the webhook payload, so it's fetchable after the fact too.
Environment
  • Alert type: Dashboard tile alert (chart), single series, metrics data source, gauge aggregation, where clause, no groupBy configured
  • Multiple alerts appear implicitly due to distinct attribute/label combinations in the underlying metric
  • Notification channel: Generic Webhook
主要语言
TypeScript
星标
10.1k
派生
527
平均合并
1 天 20 小时
30 天内合并 PR
186

环境准备

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

ClickHouse/ClickStack 的其他 Issue

查看 ClickHouse/ClickStack 的全部 Issue

相似的 Issue

更多 TypeScript Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。