A detailed comparison of ClickHouse monitoring options covers the built-in monitoring UI and embedded Prometheus endpoint, ClickHouse Cloud's console, the Grafana/Prometheus stack, Datadog's integration, the open-source chmonitor tool, and NeverBlink, an AI-native platform built by the author's own company. It explains ClickHouse-specific failure modes like TOO_MANY_PARTS, merge/mutation backlogs, replication delay from lost Keeper sessions, memory pressure, and disk exhaustion, and lists the system tables and metrics that surface each one. Recommendations favor starting with free built-in tooling, adding Grafana/Prometheus or the Altinity operator for self-hosted Kubernetes clusters, using Datadog when already in that ecosystem, and considering NeverBlink when the bottleneck is ClickHouse expertise rather than metric collection.
Table of contents
Quick ComparisonClickHouse Monitoring ToolsKey ClickHouse Metrics and Failure ModesOur RecommendationsQuestions this post answers
At what part count does ClickHouse start rejecting inserts with TOO_MANY_PARTS?
ClickHouse delays inserts once a partition reaches 1000 parts and rejects them outright at 3000 parts, triggering the TOO_MANY_PARTS error. This is usually caused by small, frequent inserts that outpace background merges. Watching the MaxPartCountForPartition async metric helps catch the problem before inserts start failing. Track evolving database internals like ClickHouse part limits by following backend engineering coverage on daily.dev.
How do I monitor ClickHouse without installing any extra exporters or agents?
ClickHouse exposes its own metrics natively: system tables like system.metrics, system.events, and system.asynchronous_metrics can be queried with plain SQL, and adding a prometheus block to the server config turns on an embedded Prometheus endpoint that scrapes every metric with zero exporters. There's also a built-in HTML dashboard at $HOST:$HTTP_PORT/dashboard and a merge visualizer at /merges. Developers evaluating database observability setups can keep comparing tooling trade-offs on daily.dev.
What does Datadog's ClickHouse integration monitor compared to other APM tools?
Datadog's ClickHouse integration tracks caches, replication delays, background pools, MergeTree parts, and query cache metrics with an out-of-the-box dashboard, monitors, and log collection, making it the most complete option among commercial APM platforms for ClickHouse. New Relic offers only a thin quickstart with one dashboard and two alerts, and IBM Instana provides an auto-discovered sensor, neither matching Datadog's depth. Pricing starts at $15 per host per month for infrastructure monitoring, with database monitoring and logs billed separately. Teams weighing Datadog against open-source alternatives for database monitoring can track these comparisons on daily.dev.