SigNoz vs Uptrace vs HyperDX (2026): OTel-Native APM Compared

If you are shortlisting an OpenTelemetry APM you can run yourself, the three-way that keeps coming up is SigNoz vs Uptrace vs HyperDX. Short answer: SigNoz is the safest default on project momentum and query flexibility, Uptrace is the pick if flat-rate self-hosted pricing matters more than ecosystem size, and HyperDX is the pick if you want session replay and are comfortable living inside ClickHouse Inc’s roadmap. All three store telemetry in ClickHouse and ingest OTLP natively, so this is not a standards argument. It is an argument about operations, retention defaults, and how you want the bill to arrive.

If your search started with the Datadog invoice rather than the tooling, our SigNoz vs Datadog comparison covers the open-source-versus-SaaS framing first.

The short answer

Pick SigNoz if:

• You want the most actively developed of the three, with weekly releases and by far the largest community.
• You need PromQL for a Prometheus migration, plus a visual builder and a raw ClickHouse SQL escape hatch.
• You want no artificial retention cap on the self-hosted edition.
• You are comfortable operating ClickHouse, ClickHouse Keeper, and PostgreSQL.

Pick Uptrace if:

• Predictable flat-rate cost is the priority and you will pay for an edition rather than run the free tier.
• You want the vendor to have published concrete hardware sizing so you can budget capacity before you deploy.
• You accept a single-maintainer project and a smaller ecosystem in exchange for a simpler mental model.

Pick HyperDX / ClickStack if:

• You want session replay stitched to backend traces, which neither of the others ships.
• You already run ClickHouse and want an observability UI that points at your existing cluster.
• You prefer Lucene-style search over building queries, and want a managed path through ClickHouse Cloud later.

Deciding factor to pick

What each one is

SigNoz is an all-in-one observability platform covering traces, metrics, and logs on a ClickHouse backend. Its core is MIT-licensed with a separate proprietary enterprise directory in the repository, an open-core split introduced back in 2022. It is the busiest of the three by a wide margin: roughly 32,000 GitHub stars and a release roughly every week, with v0.140.0 shipping in early September 2026. The company raised a 5.4 million dollar seed round in 2023 led by SignalFire; despite frequent claims to the contrary, there is no public record of a later round.

Uptrace is a Go-based APM on ClickHouse, licensed AGPL-3.0. It is genuinely smaller: around 4,300 GitHub stars, and effectively a single-maintainer project whose commit history is dominated by one contributor. Its v2.0 release in July 2025 rebuilt storage on ClickHouse’s native JSON type, with a vendor claim of 5 to 10 times faster queries. Worth knowing before you adopt: it has been on v2.1.0 beta releases since spring 2026, with the last stable release, v2.0.3, dating to May 2026.

HyperDX is the odd one out, because it is no longer an independent project. ClickHouse Inc announced its acquisition in March 2025 and folded it into ClickStack, which bundles the HyperDX UI with an opinionated OpenTelemetry Collector distribution and ClickHouse. HyperDX itself remains MIT-licensed and actively released at around 9,900 stars. Be careful with the naming: HyperDX was not renamed to ClickStack, it is the UI component inside it, and both brands still exist commercially.

Head-to-head

Self-hosting effort

The honest framing: you are not operating an APM, you are operating a ClickHouse cluster with an APM on top. That dominates everything else. What differs is how many additional stateful services come along for the ride.

SigNoz is the most conventional. Docker Compose brings up ClickHouse, ClickHouse Keeper, PostgreSQL, an OTel Collector, and the SigNoz backend and UI, with a stated minimum of 4 GB allocated to Docker. Production goes through the official Helm chart, and there is a genuine wrinkle to plan for: the Helm chart still ships Zookeeper while the Compose path uses ClickHouse Keeper. The chart also defaults to a single shard and single replica, so high availability is opt-in work, and its default resource requests are unrealistically small for production.

Uptrace has the shortest install path but the widest dependency list on paper: ClickHouse, PostgreSQL for metadata, and Redis, which the docs describe as a mandatory dependency. The strongest thing Uptrace does here is publish an actual hardware sizing table, mapping ingest volume to vCPU, RAM, SSD, and HDD, which is rare and genuinely useful when you are writing a capacity plan. The weakest: there is no published high-availability guide, so you are on your own for that design.

ClickStack offers the most deployment options, from an all-in-one container explicitly labelled as non-production, through Docker Compose and an official Helm chart, to a browser-only local mode and a fully managed ClickHouse Cloud version. The cost is a dependency nobody expects: MongoDB stores dashboards, users, and config in open-source deployments. That is a second stateful database with its own backup and HA obligations, and the documentation mostly points you at MongoDB’s own security checklist. On the plus side, ClickHouse publishes the best capacity-planning guidance of the three, with concrete vCPU-per-megabyte-per-second ratios for ingest and query.

Storage and retention cost

Because all three sit on ClickHouse, compression is broadly comparable and the real variable is retention policy and tiering effort.

Defaults are where teams get hurt. SigNoz defaults to 15 days for logs and traces and 30 days for metrics. Uptrace Cloud defaults to four weeks. ClickStack defaults to 3 days, and ClickHouse’s own production documentation says plainly that this default usually needs changing. If you deploy ClickStack for a proof of concept and come back a week later expecting to investigate an incident, the data is gone.

Tiering effort ranks like this:

• SigNoz supports S3 cold storage through Helm values (coldStorage.enabled, endpoint, credentials, IAM role annotations for EKS), though it is documented in the chart rather than on the retention page, and community reports suggest the cold-storage TTL controls are inconsistently surfaced in the UI.
• ClickStack gives you plain ClickHouse TTL, including TO VOLUME 'cold' for tiering to object storage. Maximum power, zero abstraction. ClickHouse’s own cost playbook recommends hot SSD for recent data, S3 after seven days, delete at a year.
• Uptrace handles it commercially instead: cloud spans can go to cold storage for up to 100 weeks at a published per-gigabyte-month rate, and per-project retention policies arrived in v2.0. But custom retention is a paid-edition feature, and the free Community edition is hard-capped at 14 days.

That last point deserves its own line, because it is the least-reported fact in this comparison and it reverses a lot of naive “just self-host Uptrace for free” advice. Uptrace’s self-hosted editions are priced by tier rather than by volume: a Starter tier around 39 dollars a month for up to 50 GB a day, a Team tier at 199 dollars for 200 GB a day, a Business tier at 499 dollars for 500 GB a day with SSO. If your telemetry is large and your budget hates surprises, that flat-rate model is the single best reason to choose Uptrace. If your plan was to run it free forever, the 14-day wall is the thing that ends the plan.

For the wider version of this cost argument against a commercial vendor, see reducing your Datadog bill with OpenTelemetry.

Query experience

SigNoz has the broadest query surface, and it is not close. A visual query builder with autocomplete and formulae, first-class PromQL in dashboard panels and alert rules, and raw ClickHouse SQL as the escape hatch across metrics, logs, and traces. Alerts can be built from any of the three. If you are migrating from Prometheus and Grafana, PromQL support alone can save you a dashboard rewrite.

Uptrace asks you to learn two custom languages: UQL, a SQL-like language for spans and logs with underscore-prefixed span fields, and MQL, a PromQL-like language for metrics with a documented compatibility page. “PromQL-like” is doing work in that sentence. There is also no documented raw ClickHouse SQL escape hatch, which means when the query language cannot express what you need, you are stuck.

HyperDX is the best “just find the thing” experience. Lucene-style full-text search with wildcards, booleans, quoted phrases, property filters like level:err, range operators, and natural-language time input such as “1 hour ago”. When that runs out, a SQL mode accepts any valid WHERE clause, with the documented caveat that aliases are not supported. It also does log pattern recognition, live tail, and anomaly detection on event deltas. One licensing catch to check before you build alerting on it: Slack API and PagerDuty integrations are ClickHouse Cloud exclusive, so open-source deployments are limited to generic webhooks.

What none of them do

Two gaps apply across the board and are worth stating plainly.

None of the three does continuous profiling. SigNoz’s profiling feature request has been open since August 2024 with no linked pull request; Uptrace and HyperDX do not list it either. If you want CPU and memory profiles correlated with your traces, you pair the APM with a dedicated profiler and join on trace ID. Our Pyroscope vs Parca comparison covers that layer.

None of the three matches a commercial suite on breadth. No synthetics, no CI visibility, no security modules, and a fraction of the turnkey integrations. That is fine if you scoped the comparison to core APM. It is a problem if you assumed feature parity. The OpenTelemetry versus Datadog framing is the honest way to draw that line before you commit.

Verdict by team size

• Under 10 engineers, no dedicated platform person: use a managed tier rather than self-hosting anything. HyperDX Cloud’s 20 dollar entry including 50 GB a month with 30-day retention and unlimited users is the cheapest credible starting point; SigNoz Cloud at 49 dollars including 49 dollars of usage is the better long-term shape if you expect to grow.
• 10 to 50 engineers with one platform engineer: self-hosted SigNoz. It has the momentum, the query flexibility, no retention cap, and enough community that your operational questions have already been asked by someone else.
• 50-plus engineers, high telemetry volume, cost-driven: model a paid Uptrace edition against per-gigabyte alternatives. The flat-rate self-hosted pricing is the only model here that stops the bill tracking your growth, and at 500 GB a day the arithmetic gets persuasive fast. Weigh it against the single-maintainer risk.
• Already running ClickHouse at scale: ClickStack, pointed at your existing cluster. You have already paid the operational price that dominates this comparison, and you get session replay for free on top.
• Frontend-heavy product where user-reported bugs are the pain: HyperDX, for session replay stitched to backend spans. Nothing else on this list does it.

Common pitfalls

• Deploying ClickStack and forgetting the 3-day TTL. Set retention on day one, not after the first incident you cannot investigate.
• Assuming Uptrace Community is unlimited. Free means 14 days, no RBAC, no SSO. Price the edition you actually need before you compare.
• Budgeting for the APM, not the datastore. ClickHouse operations, plus Redis or MongoDB depending on your pick, is where the engineer-hours land.
• Planning around profiling that does not exist. None of the three ships it. Pair a profiler in from the start.
• Comparing cloud prices in different units. SigNoz and HyperDX quote per gigabyte ingested; Managed ClickStack quotes infrastructure consumption per gigabyte per month. Those are not the same number and cannot be divided against each other.

Getting help

We run vendor-neutral observability selection sprints: structured trials against your real traffic, total-cost modelling for self-host versus managed across all three, and a recommendation matched to the operational appetite you actually have rather than the one the vendor assumes. If the stack is already chosen and the problem is that it is not finding your bottlenecks, our Performance Audit and Continuous Profiling engagements wire it into a working performance practice.

Book a free scope call.