AgentSkillsCN

altinity-expert-clickhouse-part-log

通过分析 system.part_log 表（分区创建、合并、变更、下载、移除、移动），诊断 ClickHouse 的各类问题。适用于处理分区数量过多、微批次插入、合并积压或合并缓慢、变更风暴（ALTER DELETE/UPDATE）、异常的复制下载波动、突发的 RemovePart 增长，或 ZooKeeper/Keeper ZNode 数量随分区活动而增长等场景。

SKILL.md
--- frontmatter
name: altinity-expert-clickhouse-part-log
description: Diagnose ClickHouse issues by analyzing system.part_log (part creation, merges, mutations, downloads, removals, moves). Use for too many parts / micro-batch inserts, merge backlog or slow merges, mutation storms (ALTER DELETE/UPDATE), unusual replication DownloadPart churn, unexpected RemovePart spikes, or ZooKeeper/Keeper znode growth correlated with part activity.

Part Log Based Diagnostics

Run all queries from checks.sql (cluster-wide) and interpret the top offenders by rate (events/min), volume (rows/bytes), and errors.

Notes:

  • •Default timeframes are relative (e.g., last 1h/6h/24h). Only switch to an explicit time range when the user provides one in the prompt.
  • •Replace {cluster} with your ClickHouse cluster name (DataGrip).
  • •Keep queries time-bounded (event_time > now() - INTERVAL ...) and use LIMIT.
  • •If a query fails due to schema differences, run DESCRIBE TABLE system.part_log and drop only missing fields.

Cross-module triggers:

  • •High NewPart rate / micro-batches → load altinity-expert-clickhouse-ingestion + altinity-expert-clickhouse-merges
  • •High MutatePart rate → load altinity-expert-clickhouse-mutations
  • •Many DownloadPart → load altinity-expert-clickhouse-replication
  • •Merge saturation / slow merges → load altinity-expert-clickhouse-merges + altinity-expert-clickhouse-storage