Uptrace

Overview

Uptrace is an OpenTelemetry backend covering traces, metrics and logs, built around ClickHouse. It competes with Datadog and New Relic mainly on how it charges: there are no per-seat or per-host fees, so adding engineers or scaling out a fleet does not change the bill.

The pricing model is the product

Ingestion bills per GB and the rate falls as volume rises — published rates start at $0.10/GB and reach $0.016/GB at scale, with metrics from $0.025 per million datapoints. Extended cold storage runs $0.01 per GB-month for spans. Data you sample out still costs $0.01/GB.

The more interesting control is a monthly budget cap. You set a ceiling and the bill does not pass it. That addresses a specific complaint about usage-priced observability, where a bad deploy that floods logs produces an invoice nobody approved. Datadog's high-watermark billing, which charges against the 99th percentile of monthly usage, makes short spikes expensive in a way a hard cap does not.

The free tier is 50 GB and 5,000 timeseries a month, enough for a small production service rather than just a demo.

Self-hosting

A community edition runs from GitHub with no volume limits. There is also an on-premises enterprise edition, which needs a licence; the pricing page does not publish a figure and points at a free evaluation period instead.

Where it fits

Uptrace suits teams who have already instrumented with OpenTelemetry and want a backend that bills on ingest rather than infrastructure footprint. That describes a lot of Kubernetes shops where host counts move constantly and per-host pricing punishes autoscaling.

It is a smaller company than the platforms it is compared against, and the integration catalogue reflects that. Its compression figures — 10-20x from ClickHouse storage — come from Uptrace's own documentation and have not been independently benchmarked.

Migrating onto it

Because Uptrace speaks OpenTelemetry, migration is a collector change rather than a re-instrumentation. If services already emit OTel, pointing the exporter at an Uptrace endpoint is the bulk of the work, and running two exporters side by side during evaluation lets you compare query results on identical data before cutting over.

That is worth doing deliberately. Backend behaviour differs on high-cardinality metrics and on how aggressively spans are sampled, and a dashboard that looks right in one backend can be subtly wrong in another. A fortnight of dual-writing costs little and surfaces those differences before they matter.

Who it suits

The clearest fit is a Kubernetes team whose host count moves constantly. Per-host pricing penalises autoscaling directly — a cluster that doubles overnight doubles the bill, even if request volume did not change much. Billing on ingested data decouples cost from infrastructure shape, which is the behaviour most teams actually want.