The websocket metric's label grows without bound

已关闭
  • 哪里不对？
#337 2 条评论 0 个 reaction 已指派 1 人 在 GitHub 查看

维护者通常 4 天内回复

@Jossy-ossy 已经在做这个了。

开始于 2026年10月16日。

评估

这个 Issue 还没有评估数据。

描述

area:observability difficulty:intermediate help wanted points:150 Stellar Wave
Background

src/ws/fanout.ts increments:

ws_messages_sent_total.labels(client.pairKey).inc()

pairKey is user-supplied through the subscription — it is whatever pair a
client asked for. Prometheus creates a new time series for every distinct label
value and keeps it for the lifetime of the process, so the number of series
grows with the number of distinct pairs ever subscribed to, not with the number
of pairs that exist.

A client subscribing to randomly generated pair keys can grow the /metrics
response without limit. Even without malice, a screener that walks every pair
produces thousands of permanent series. This is the standard Prometheus
cardinality failure, and it ends with /metrics too slow to scrape.

What to build
  • Only label with a pair that is known — validate pairKey against the pairs
    the service actually tracks before using it as a label value.
  • Unknown pairs should either be rejected at subscribe time or folded into a
    single other label.
  • Audit the rest of src/metrics.ts and src/middleware/httpMetrics.ts for the
    same pattern; any label value derived from user input is a candidate.
  • Document the rule in src/metrics.ts so the next metric does not reintroduce it.
Acceptance criteria
  • No metric label can take an unbounded set of values from user input
  • A test subscribing to many unknown pairs does not create a series per pair
  • The existing * aggregate label keeps working
  • The rule is written down next to the registry definition

Drips Wave · Complexity: Intermediate · 150 points

Repo: Divinity-Labs/Lens. Node + TypeScript, Fastify, Prisma on Postgres,
Redis for queues, Prometheus via prom-client.

Run the tests: npm test. The tests are vitest, not jest. Running jest
against this repo reports "66 suites, 0 tests" and means nothing. A single file
is npx vitest run src/__tests__/yourfile.test.ts.

Load test (prices) is already failing on main and has been for some
time. It is not your PR that broke it. Every other check should be green.

This service prices real assets on Stellar mainnet. A number that is wrong
is worse than a number that is missing, because a caller cannot tell the
difference. Prefer failing loudly over returning a plausible value.

Never elide an address, a contract id or a key in code, a comment or a test
fixture. A previous batch wrote a shortened contract id in an issue text and
five contributors invented plausible-looking ids to fill the gap. Write the
whole thing or do not write it at all.

The house pattern for an outbound deadline already exists — see
src/alerts/delivery.ts, which passes signal: AbortSignal.timeout(10_000) to
fetch. Where an issue below asks for a timeout, copy that, do not invent a
second convention.

Required: Before submitting, join the contributor Telegram so your work can be tracked and counted toward the Stellar Wave: https://t.me/+fxHXq8f1SwlkZDBk

主要语言
TypeScript
星标
4
派生
71
平均合并
3 天 22 小时
30 天内合并 PR
142

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。

Divinity-Labs/Lens 的其他 Issue

查看 Divinity-Labs/Lens 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。