AWS has introduced the public preview of OpenTelemetry metrics support in Amazon CloudWatch. This update allows developers to send metrics directly to CloudWatch using the OpenTelemetry protocol and view them alongside existing AWS service metrics.
The preview introduces a high-cardinality metrics store that supports up to 150 labels per metric, allowing rich metrics to be sent directly to CloudWatch without conversion or truncation. With AWS automatically adding resource metadata to metrics, CloudWatch can serve as a single place to collect infrastructure, container, and application metrics.
In the "Introducing OpenTelemetry & PromQL Support in Amazon CloudWatch" article, Rodrigue Koffi, specialist solutions architect at AWS, writes:
With this launch, CloudWatch completes its support for OpenTelemetry across all three pillars of observability. CloudWatch already accepts traces and logs through its OTLP endpoints, adding native OTLP metrics ingestion means you can now send all your telemetry to CloudWatch using open standards, through a single protocol.
Source: AWS blog
Teams can then analyze them using PromQL, the query language used in Prometheus to retrieve and analyze monitoring metrics, in the CloudWatch console, and build dashboards and alarms to monitor applications across environments, including Kubernetes clusters and on-premises systems. This allows teams already using Prometheus to reuse the same query language in CloudWatch and Managed Grafana. Julius Volz, founder at PromLabs and co-founder of prometheus.io, comments:
Nice to see Amazon CloudWatch now also betting on PromQL as a metrics query language.
AWS already supported the opposite scenario, delivering CloudWatch metrics to OpenTelemetry with "near-real-time delivery and low latency" using metric streams. Koffi explains the advantage of the new feature:
This capability fundamentally changes how you query and filter metrics across your AWS infrastructure. CloudWatch enriches every ingested metric with AWS resource context: account ID, Region, cluster ARN, and resource tags from AWS Resource Explorer. This enrichment happens automatically, without additional instrumentation.
The community's reaction to the OpenTelemetry support in CloudWatch has been mostly positive, though some concerns have been raised about long-term costs and metric cardinality. Luc van Donkersgoed, principal engineer at PostNL and the author of aws-news.com, comments:
This is both very cool and potentially very expensive for high-cardinality metrics.
In his newsletter, Corey Quinn agrees:
"Free during preview" is doing a lot of heavy lifting here. CloudWatch finally accepting OTEL metrics natively means you can stop maintaining that janky conversion pipeline you pretend doesn't exist. Enjoy the honeymoon period before pricing drops and your CFO discovers you've been shipping metrics from every microservice known to humanity.
In a separate announcement, the cloud provider also introduced the preview of OpenTelemetry-based Container Insights for Amazon EKS in Amazon CloudWatch. The feature extends existing Container Insights capabilities by collecting additional metrics from open source and AWS collectors and sending them to CloudWatch via the OpenTelemetry Protocol.
OpenTelemetry metrics support is currently available in five regions, including Northern Virginia and Ireland, with no charge during the preview period.