Your observability bill doubles every quarter, and half of the logs are never read again. Before you change tools, ask a better question: where should telemetry live? A growing answer is ClickHouse telemetry storage, a column-oriented analytical database released under Apache 2.0.

Two open source projects build on it: SigNoz and ClickStack (formerly HyperDX). This article follows the operating decision rather than the documentation: what do you gain, what do you lose, and when is it wiser to stay where you are? The facts come from the GitHub repositories and the official documentation, read on 11 November 2026.

A note on honesty: span.observer has not deployed these stacks. The excerpts reproduce the cited sources, and we say what is left for you to check. To place the topic against other approaches, see OpenTelemetry and Prometheus and Dynatrace and Grafana.

ClickHouse telemetry storage: a database built for analysis, not for search

Logs and spans look alike: many rows, almost no updates, filters on time and on a few fields. That profile suits a column store. It reads only the columns you ask for and compresses repetitive values, such as a service name, very well.

SigNoz puts it in its own words: ClickHouse is the central data store, and its compression is presented as excellent for cutting storage cost. We have not measured that ratio on your data, and it depends on its shape. Test a real sample before promising any saving.

The downside is well known. This is not a classic full-text search engine. What used to be simple becomes a matter of schema and sort keys.

From the Collector to ClickHouse: the shortest path

The neutral entry point is the clickhouse exporter of the OpenTelemetry Collector (contrib repository). Its README marks logs and traces as beta and metrics as alpha. That is a decision signal: do not trust critical metrics to an alpha component without a fallback.

File otel-collector.yaml: clickhouse exporter with endpoint, database otel, async\_insert, ttl 72h, compress lz4 and create\_schema true

Source : github.com

Three options deserve attention. ttl defaults to 0, meaning no expiry. create_schema defaults to true, yet the README recommends false in production, where you then own the DDL. async_insert stays on by default to batch writes.

To go further on Collector components, read our tour of the contrib components.

The schema decides your query bill

The default sort keys of the otel_logs and otel_traces tables start with ServiceName. The ClickHouse documentation warns that they probably do not match your access patterns, and advises picking 4 or 5 columns from your most frequent filters.

SQL excerpts: default ORDER BY keys of the otel\_logs and otel\_traces tables, and a RequestType column materialised from the LogAttributes Map

Source : clickhouse.com

  • Attributes in a Map: reading one key loads the whole parent column, so it is slower than a normal column. The documentation advises promoting frequent keys to root columns.
  • Materialised columns: computed at insert time, they speed up filters but add storage.
  • Secondary indexes: the page judges the default Map-related ones largely ineffective. Do not copy them without measuring.

That is the hidden cost. A search engine indexes everything and hides the question. Here you choose, and you pay for that choice in engineering time.

SigNoz: ClickHouse behind a complete platform

SigNoz bundles logs, metrics, traces, alerts and dashboards. Its architecture places a Collector that writes to ClickHouse and an API server that queries it. According to the README, queries can use a Query Builder, PromQL or ClickHouse SQL.

On licensing, the repository is mixed. The LICENSE file says content outside the ee/ and cmd/enterprise/ directories is under the MIT licence, while those directories fall under an enterprise licence. Read it before any production use: it sets contractual conditions for the code concerned.

Retention is the concrete point. By default, the documentation gives 15 days for logs and traces and 30 days for metrics. A change applies only to new data, and a longer period recovers nothing that has already expired. Self-hosted, you set it under Settings then Retention Controls; on SigNoz Cloud it goes through support and changes the bill. For a real use case, see our article on AI agents with SigNoz.

ClickStack: a UI on top of your ClickHouse

ClickStack, formerly HyperDX, is published under the MIT licence according to the repository LICENSE file. The README describes it as a search interface for logs and traces on top of any ClickHouse cluster, with a simple search syntax and optional SQL.

The approach differs from SigNoz. The all-in-one image bundles ClickHouse, the UI, an OpenTelemetry Collector and MongoDB. The documentation also says the UI can connect to an existing ClickHouse, and that collectors, database and UI scale independently.

Terminal: docker run command for the clickstack-all-in-one image with ports 8080, 4317 and 4318 and the list of its components

Source : github.com

Careful: the overview page we read describes neither a TTL nor numeric limits. Retention is therefore a matter of your own ClickHouse configuration. The README recommends at least 4 GB of RAM and 2 cores for testing, which is not production sizing.

What you gain, what you lose

To sum up the decision, here are the trade-offs, in order of importance for an SRE team.

  • Cost and retention: you control the disk and the duration. In return, you run the database: backups, upgrades, capacity.
  • Queries: SQL gives power over high-cardinality fields. But a poor sort key is paid for on every query.
  • Maturity: beta or alpha components in the Collector, and table changes that are sometimes not backward compatible according to the README. Read the changelog before each upgrade.
  • Correlation and automation: a commercial platform adds causal analysis and alerting out of the box. Here you build them.

At Dynatrace, retention is set per platform: for Managed, see our article on retention in Managed. For SaaS, rely only on docs.dynatrace.com. We do not compare those features with ClickHouse, for lack of measurements.

How to check your choice before migrating

  1. Export a week of real telemetry to a test ClickHouse with create_schema: true.
  2. Note the size on disk per signal, then compare it with your current bill.
  3. Replay your five most frequent queries and time them.
  4. Adjust the sort key, promote two attributes to columns, then repeat.
  5. Switch create_schema to false and put your DDL under version control.

To confirm it works: the otel_logs and otel_traces tables should fill up and your typical query should answer within an acceptable time. If not, look first at insert errors in the Collector logs.

Best practices

  • Set an explicit ttl: its default is 0, meaning no expiry.
  • Disable create_schema in production and keep the DDL in Git.
  • Promote frequently filtered attributes to columns.
  • Test compression on your data before quoting a saving.
  • Plan for restores: self-hosted storage means backups.

Frequently asked questions

Are ClickStack and SigNoz the same thing?

No. Both use ClickHouse, but SigNoz is a complete platform with its own schema, whereas ClickStack is first a UI on a ClickHouse, with a schema supplied by its Collector.

What is the default retention with SigNoz?

According to its documentation, 15 days for logs and traces and 30 days for metrics. Check the page before relying on it, as these values may change.

Can I send Collector data to ClickHouse without SigNoz?

Yes, with the clickhouse exporter from the contrib repository. Its status is beta for logs and traces and alpha for metrics.

Conclusion

ClickHouse is a good candidate when you want to control cost and retention, and accept running a database. Start with a one-week test on real data, then decide. Sources: ClickHouse documentation, SigNoz/signoz and ClickHouse/ClickStack.

span.observer is an independent community blog, not affiliated with Dynatrace LLC, ClickHouse, SigNoz or ClickStack. Brands mentioned belong to their owners. Tools change quickly: check commands and options in the official documentation.

Originally published on span.observer. French version: https://span.observer/fr/clickhouse-stockage-telemetrie/.