Datadog vs Uptrace: Choosing Who Operates Your OTel Backend

Several services already emit OpenTelemetry, and a Collector is running reliably. Uptrace naturally enters the backend shortlist: it can receive existing telemetry, support investigations across traces, logs, and metrics, and run in an environment the team controls. Datadog may already provide host monitoring, alerts, and other operational workflows across the company.

The choice is less about the fewest components needed to draw a chart than the service the team wants to own. Uptrace Cloud and a self-hosted Uptrace installation are different arrangements, with different costs and responsibilities. Treating them as one option obscures the decision.

Reviewed against official documentation on October 21, 2026. No comparative performance testing was performed. Sizing examples are illustrative.

Receiving OTel data is the start of a migration

Uptrace uses OpenTelemetry as a primary ingestion path for traces, metrics, and logs. Existing SDKs and Collectors can preserve useful investment, but endpoints, credentials, resource attributes, and signal configuration still need checking. A received span is not yet proof that service names, errors, and timestamps support a real investigation. Uptrace getting started.

Datadog also supports OpenTelemetry, with differences in compatibility across ingestion paths and product capabilities. It is inaccurate to assume that every deployment requires replacing OTel with proprietary SDKs. List the features the team actually uses, then evaluate the chosen path. Datadog OTel compatibility.

Naming can be a larger obstacle than transport. If a URL containing an order identifier becomes an operation name, an endpoint's traffic may fragment into many groups. If test and production share service identifiers, imported data may require additional separation. Standard transport does not solve those conventions.

Test a real request that crosses a queue. Missing producer-to-consumer context can leave two individually valid traces that still require manual stitching. Start with that difficult path rather than treating a synchronous HTTP demonstration as complete migration evidence.

Verify the workflow, not just the feature label

Uptrace includes logs, metrics, error analysis, and alerting as well as tracing. Datadog has a broader product catalog, but breadth matters commercially only where those capabilities are needed. A claim that another tool covers “80% of requirements” is not useful without the requirements themselves. Uptrace documentation.

Synthetic monitoring illustrates the distinction. Uptrace documents HTTP endpoint checks using the OTel Collector's httpcheck receiver. It therefore has a supported path for active checks. A customer-operated endpoint probe is nevertheless a different service from a package containing managed probe locations, browser journeys, and test operations. Uptrace HTTP checks.

Recreate an existing alert with its evaluation window, threshold, missing-data behavior, and notification owner. Then follow it into the evidence used by the on-call engineer. Discovering a similarly named feature is less important than preserving a useful sequence of actions during an incident.

Self-hosting involves more than the binary

The current self-hosting guide lists Uptrace plus ClickHouse, PostgreSQL, and Redis. Telemetry storage, project configuration, and caching or background coordination have distinct roles. Deployment automation simplifies installation, but the dependencies still need capacity, backups, and upgrades. Self-hosting architecture.

A team experienced with ClickHouse can reuse that knowledge. A team without it should not equate a working container with a maintained service. Establish who configures retention, backs up configuration data, recovers failed upgrades, and investigates the observability platform when it becomes unhealthy.

A small recovery exercise is revealing. Save a known trace, back up the necessary data and configuration, rebuild the environment, then repeat the query and check access. This exposes cases where telemetry survives but project settings do not. Control becomes valuable when someone can perform the operations it enables.

Ingestion peaks expose operational differences

Using an analytical database does not guarantee fast queries under every workload. Uptrace's scaling documentation covers processing buffers, insertion batches, sharding, and Kafka. Full in-memory buffers can cause incoming records to be dropped. More buffering absorbs some short stalls, but requires memory and does not fix a sustained throughput deficit. Scaling Uptrace.

For illustration, an additional 10 MB per second waiting to be written creates 600 MB of raw backlog during a sixty-second storage stall. That is not the complete process-memory requirement: data structures, other signals, and concurrent queries also consume resources. Quiet-period memory usage is consequently a poor basis for a production limit.

Kafka can decouple ingestion and provide buffering, but the current documentation limits its integration to Enterprise. It also introduces another system to operate. The decision depends on tolerated outages and available expertise. More components are not automatically better, but ignoring backlog behavior is not a cost-free simplification either.

Uptrace Cloud or an explicitly managed deployment can transfer more backend work to the provider. The customer still owns collection, business attributes, and delivery monitoring. Confirm recovery and support scope rather than inferring it from the number of commands on an installation page.

Separate free software, self-hosted plans, and Cloud

At the review date, the public Uptrace repository's main license is AGPLv3. Check the selected release and commercial terms rather than describing every edition as Apache- or BSL-licensed. Current self-hosted plans distinguish Community from paid tiers; retention, access management, scaling, and support are not uniformly included. Repository license, self-hosted editions.

Paid self-hosted plans combine edition subscriptions with applicable usage tiers. Cloud uses its own telemetry meters. Its current pricing page separates traces, logs, and metrics, with metrics described in ingested data points. Series count alone is insufficient: measuring the same series every thirty seconds produces more observations than measuring it every sixty. Additional retention also needs a separate check. Cloud pricing.

Measure bytes and data points from representative traffic, then allow for growth, peaks, and retention. A vendor's illustrative span size is not the size of your records. Business attributes and exception stacks can change it considerably. Network egress and self-hosted infrastructure also sit outside the software subscription.

Datadog requires an equally scoped calculation. Moving APM while retaining logs, infrastructure, or browser monitoring does not eliminate the entire existing bill. Include temporary dual shipping, historical-data access, and alert reconstruction in the migration budget. A lower steady-state subscription may still require a substantial transition effort.

Plan the end of the trial too: exportable configuration, retirement of old alerts, and access to historical incident links all need owners. The first trace in the new backend is not a reason to close the old service immediately. For a small team, database upgrades may also compete with business releases for the same engineer. Put that work on a calendar before assigning it a cost of zero.

Make operating capacity part of the decision

Self-hosted Uptrace merits evaluation when OTel is established, the required investigations are focused, and the team can maintain the database services. Cloud or a managed arrangement deserves a separate assessment when the workflow is attractive but backend ownership is not. An established Datadog deployment should be replaced to solve an actual limitation, rather than simply because another product starts for free.

Guance offers another route to service and trace investigation through supported collection methods. SaaS reduces backend construction, while private deployment requires an explicit maintenance agreement. Application context, sampling, and propagation still need owners. Follow the same business request through alerting, diagnosis, and verification in each candidate, then compare the budget and responsibilities of the service you would actually operate. Guance APM.