← All databases
CH

ClickHouse monitoring

ClusterEye monitors ClickHouse through an agent over the native protocol (port 9000) — the HTTP interface is not used. Query history, part and merge behaviour, replication queue and mutation progress are collected from the system database. The monitoring user can be created with readonly = 1. Versions: 21 — 24.

Agent-basedDefault port: 9000Versions: 21 — 24
ClickHouse — ClusterEye Performance Analyzer

What we read

system.query_log — query history, duration, rows read and memory usage
system.metrics, system.events, system.asynchronous_metrics, system.metric_log — live and historical server metrics
system.parts and system.detached_parts — part counts, size distribution and detached part accumulation
system.merges and system.mutations — in-flight merges and stuck mutations
system.replicas and system.replication_queue — replication lag and queue backlog
system.processes — running queries and their resource consumption
system.disks, system.clusters, system.zookeeper — storage, cluster topology and Keeper state
system.errors and system.settings — error counters and configuration state

What it does specifically for ClickHouse

Query regression detection

Queries from system.query_log are grouped by fingerprint and given a baseline from their own history; those whose duration or bytes read climb abnormally are flagged as anomalies.

MergeTree health

Part counts, merge backlog, stuck mutations and detached part accumulation are tracked — warning before it becomes a 'too many parts' error.

Replication and Keeper visibility

For ReplicatedMergeTree tables, replication queue, lag and Keeper connectivity are tracked, identifying which table is falling behind.

Minimum privileges required

The baseline is read-only and safe to grant in production.

sql
CREATE USER clustereye IDENTIFIED WITH sha256_password BY '<strong-password>'
  SETTINGS readonly = 1;

GRANT SELECT ON system.* TO clustereye;
GRANT SHOW DATABASES, SHOW TABLES, SHOW COLUMNS ON *.* TO clustereye;

Prerequisites

·The native protocol port (9000) is used — the HTTP interface (8123) is not supported.
·log_queries = 1 (default) — for query intelligence
·metric_log enabled — for per-second metric history
·Keeper configured — for system.zookeeper
Install and privilege details →

Frequently asked questions

No. The agent uses the native protocol (default port 9000); the HTTP interface (8123) is not supported. That port must be reachable from the host running the ClusterEye agent.

Start monitoring your ClickHouse estate in five minutes.

One command to install. Talk to us if you have questions.

Book a live demoInstall and privilege details