Series limit isn't applied in SQL for metric charts: every series is fetched, then capped in the browser

オープン
#3,502 コメント 0 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
4/5
見積もり時間
3〜5日
初心者へのやさしさ
48/100
issue の種類
バグ
明瞭さ
明確に書かれている
活発さ
活発
技術スタック
clickhouse, typescript

調査の方向性

Start by reading renderChartConfig.ts around renderSeriesLimitCte and the metric CTE construction, then compare with the client-side cap in ChartUtils.tsx. Reproduce the issue with an integration test in the style of renderChartConfig.int.test.ts using gauge, sum, and histogram metrics. Done when metric queries return only the requested top N series from SQL, as log and trace queries do.

索引モデルが issue の本文から書いたものです。

説明

external

What happened?

With seriesLimit: 2 and a group-by over 5 hosts, on main (2bfc630):

Source Groups the SQL returns
logs (control) 2
gauge metric (avg) 5
sum metric (sum) 5
histogram metric (count) 5

renderSeriesLimitCte skips metric sources ("Skip CTE/metric sources (no real table to re-scan)"):

The app relies on that CTE, and otherwise caps the series client-side:

So a top-N metric chart on a high-cardinality attribute reads, aggregates and transfers every series, then the browser draws N of them. On our deployment:

  • a top-20 chart over a histogram with ~280 groups read 76 GB;
  • a top-10 chart over a sum with ~2,700 groups hit our 8 GiB per-query memory limit.

Expected: metric charts return the top N series from SQL, as log and trace charts do. The ranking could run on the per-series values after the metric CTEs, before the final pivot.

Related: #3204 (how the top N is picked across chunks).

Steps to reproduce

Reproduces on play.hyperdx.io? Didn't try.

An integration test in the style of renderChartConfig.int.test.ts, on main:

  1. Insert a gauge, a cumulative sum and a cumulative histogram repro.many for hosts h1–h5 (ResourceAttributes['host']).
  2. renderChartConfig with groupBy: "ResourceAttributes['host']", seriesLimit: 2, granularity: '5 minute', and one select of each metric type in turn.
  3. Each query returns all 5 hosts. The same config on a log source with groupBy: 'ServiceName' returns 2.

How are you running HyperDX?

Helm (ClickStack chart 2.1.0) with the hyperdx:2-nightly image (2.42.0 plus the Node bump from #3456), our own ClickHouse 26.2.5.45. Reproduced on main 2bfc630.

Where does it show up?

Dashboard time charts on metric sources with "Top N series"; we saw the cost through the MCP tools and in system.query_log.

Debug info

N/A.

Logs

None.


Written with help from an AI assistant (Claude Code). The behaviour comes from our own deployment and an integration test on main.

主要言語
TypeScript
スター
10.1k
フォーク
527
平均マージ
1日 20時間
マージ済み PR(30日)
186

環境構築

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

ClickHouse/ClickStack のほかの issue

ClickHouse/ClickStack の issue をすべて見る

似ている issue

TypeScript の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。