AWS Cloud Operations Blog
This Month in AWS Observability: June 2026
Introduction
Welcome to the latest edition of This Month in AWS Observability, featuring what’s new across Amazon CloudWatch and AI-driven operations this June! Native OpenTelemetry metrics with PromQL querying is now generally available in CloudWatch, 23 new Logs Insights commands launched for deeper statistical and structured analysis, Session Replay now in CloudWatch RUM, and AWS DevOps Agent shipped custom SRE agents with MCP and Agent-to-Agent protocol support. Whether you’re attributing GPU costs across EKS clusters, debugging frontend issues with real session playback, or encoding your team’s runbooks into autonomous SRE agents, you’ll find something useful. And if you want to see a demo of these new capabilities, register for “I Didn’t Know Amazon CloudWatch Could Do That” webinar on August 11.
OpenTelemetry and Prometheus
Amazon CloudWatch reaches a major milestone with the general availability of native OpenTelemetry metrics and PromQL querying, while Amazon Managed Service for Prometheus adds capabilities that reduce cost and complexity for high-cardinality workloads.
Native OpenTelemetry Metrics with PromQL Querying (GA)
Amazon CloudWatch now natively supports OpenTelemetry metrics in general availability. Teams can send metrics directly via the OpenTelemetry Protocol (OTLP) and query them using Prometheus Query Language (PromQL), with per-GB ingestion pricing and 15 months of storage included. This eliminates the need for separate Prometheus-compatible backends alongside CloudWatch. High-cardinality metrics with dozens of labels (namespace, pod, container, node, deployment) flow directly into CloudWatch and are immediately queryable using the PromQL syntax teams already know.
CloudWatch pipelines supports processing and enriching OpenTelemetry metrics
CloudWatch Pipelines now lets you process and enrich OpenTelemetry metrics during ingestion without custom infrastructure or changes to application instrumentation. For example, adding business context, stripping high-cardinality labels, or enforcing naming conventions.
Native Histogram Support in Amazon Managed Service for Prometheus
Native histograms replace classic Prometheus histograms that emit one time series per bucket boundary. Instead of predefining 20+ bucket boundaries and paying for 20+ time series per metric, native histograms store the full distribution in a single time series with dynamic resolution. This reduces storage costs and cardinality while preserving accurate percentile calculations for latency, request duration, and value distribution tracking.
GPU Cost Attribution with OTel and Managed Prometheus
For teams scaling AI/ML workloads on Amazon Elastic Kubernetes Service, a new reference architecture demonstrates GPU cost attribution using OpenTelemetry collectors, Amazon Managed Service for Prometheus, and Amazon Managed Grafana. Teams can attribute GPU costs per namespace, team, and workload type, answering the question of who is consuming the most expensive compute resources in a shared cluster.
Analyzing Claude Code Usage with CloudWatch and OpenTelemetry
As AI coding agents like Claude Code grow across engineering organizations, tracking token consumption, cost per team, and developer productivity becomes critical. This pattern uses the CloudWatch OTLP endpoint to ingest custom OpenTelemetry metrics from Claude Code sessions, enabling dashboards that answer questions existing tooling cannot: which teams consume the most tokens, what is the cost per developer, and where is AI-assisted development delivering the most value.
Log Analytics Evolution
New query and ingestion capabilities in Amazon CloudWatch Logs makes it easier to analyze, search, and route log data at scale.
23 New CloudWatch Logs Insights Query Commands
- CloudWatch Logs Insights adds 23 new query commands spanning multiple categories:
- Parse commands: parse_json, parse_kv, parse_csv, and parse_xml for structured extraction from diverse log formats
- Analytics functions: median, percentile_cont, mode, variance, and stddev for deeper statistical analysis
- Hash and IP functions: md5, sha256, ip_to_int, and cidr_match for security and network investigation workflows
- Conditional logic: case, coalesce, nullif, and if for more expressive query construction
These additions reduce the need to export logs to external analytics platforms for complex investigations.
The new Log Analytics console provides a unified interface for querying across log groups, exploring log data with natural language, and saving reusable query patterns. This consolidates what previously required navigating between multiple console pages into a single workflow.
Operators can now query log groups by tag rather than specifying individual log group names. For organizations with hundreds or thousands of log groups, this simplifies query scoping; tagging log groups by team, environment, or service tier allows analysts to query “all production logs for the payments team” without maintaining explicit lists.
CloudWatch Logs now supports managed syslog ingestion via VPC endpoints using TCP, TLS, and UDP. The feature handles RFC 5424, RFC 3164, and vendor-specific formats including Cisco FTD and ASA. Syslog messages are automatically parsed into structured fields, making them immediately queryable in Logs Insights without manual transformation.
Elastic Beanstalk Log Integration
AWS Elastic Beanstalk environments now stream application and platform logs directly to CloudWatch Logs with native integration, eliminating the need for custom .ebextensions configuration or log streaming agents on Beanstalk-managed instances.
Metrics and Monitoring at Scale
New query and Metrics capabilities expand to support larger, more complex multi-account architectures with longer retention and deeper service-specific visibility.
Cross-Account Metrics Centralization
CloudWatch now supports centralized cross-account metric collection with streamlined configuration. Organizations can designate a monitoring account that automatically aggregates metrics from member accounts, enabling fleet-wide dashboards and alarms without manual per-account setup.
Metrics Insights Extended Retention
CloudWatch Metrics Insights queries now support extended retention periods, allowing teams to run analytical queries across longer historical windows. This enables trend analysis, capacity planning, and year-over-year comparisons directly within CloudWatch rather than requiring data export to external analytics systems.
Amazon ElastiCache: 13 New CloudWatch Metrics
Amazon ElastiCache publishes 13 new CloudWatch metrics covering memory fragmentation, connection churn, key eviction patterns, replication lag granularity, and command-level latency distributions. These metrics surface early warning indicators for cache degradation that previously required connecting directly to Redis or Memcached instances to observe.
End-User and Synthetic Monitoring
Monitoring extends closer to the end user with enhanced real-user visibility and more flexible synthetic testing topologies.
CloudWatch Real User Monitoring (RUM) now includes Session Replay, which records and plays back actual user sessions as they occurred. When investigating customer-reported issues, engineers can watch the exact sequence of interactions, page loads, errors, and latency the user experienced, eliminating guesswork from frontend debugging.
Synthetics Multilocation Canaries
CloudWatch Synthetics introduces multilocation canaries that execute the same synthetic test from multiple AWS Regions simultaneously under a single management configuration. Each location operates independently; if one location reports a failure while others succeed, CloudWatch correlates the signals to distinguish localized issues from global outages, reducing false-positive alarm noise.
Intelligent Operations with AWS DevOps Agent
AWS DevOps Agent continues to evolve as the AI-powered operational assistant for cloud environments, with new capabilities improving both intelligence and reach of automated investigations.
Teams can now build custom SRE agents tailored to their specific operational runbooks and infrastructure patterns. Custom agents encode institutional knowledge about service architectures, common failure modes, and preferred remediation steps, enabling faster time-to-resolution for recurring issues.
AWS DevOps Agent now supports the Model Context Protocol (MCP) and Agent-to-Agent (A2A) communication. This enables DevOps Agent to orchestrate with other AI agents in a multi-agent architecture, calling external tools, querying knowledge bases, and coordinating remediation workflows across organizational boundaries.
Webhook Triggers and Disambiguation Cards
New webhook-based triggers allow external systems (PagerDuty, Datadog, custom alerting pipelines) to initiate DevOps Agent investigations automatically. When an investigation encounters ambiguity, disambiguation cards present operators with structured choices rather than halting execution, keeping the investigation loop moving forward with human guidance.
AWS DevOps Agent is now available in five additional AWS Regions, bringing the total to broad global coverage. Teams operating in Europe, Asia Pacific, and additional North American regions can now run investigations locally, reducing data residency concerns and improving latency for real-time operational workflows.
Developer Experience Improvements
AWS DevOps Agent receives targeted quality-of-life improvements that reduce friction in daily operational workflows.
Release Management Capability (Preview)
DevOps Agent now offers release management in preview, reviewing code changes for production readiness and checking for standards drift, dependency impacts, and Well-Architected alignment, while autonomously generating and running test plans to catch regressions before deployment. This helps teams ship faster and reduce MTTR across AWS, multicloud, and on-prem environments.
Agent spaces can be configured with static IP addresses for outbound traffic, enabling integration with external systems that require IP-based allowlisting (on-premises ITSM platforms, partner APIs, legacy firewalls).
Auto-Growing Chat and Full Context Retention
The DevOps Agent investigation interface now features auto-growing chat input and full context retention across investigation sessions. Longer, more detailed prompts are supported without truncation, and returning to a prior investigation preserves the complete conversation history and findings.
Related Blog Posts from AWS Cloud Operations Blog (June 2026)
The following posts were published on the AWS Cloud Operations Blog during May and June 2026:
June 2026
Log analysis with facets, correlation, enrichment, and automation in Amazon CloudWatch Log Analytics (19 Jun) – Salman Ahmed, Ravi Kumar
Analyzing Claude Code usage with CloudWatch and OpenTelemetry (17 Jun) – Rodrigue Koffi, Gianluca Cacace, Vadim Omeltchenko
GPU Cost Attribution in Amazon EKS Using Amazon Managed Service for Prometheus, Amazon Managed Grafana, and OpenTelemetry (12 Jun) – Siva Guruvareddiar
Introducing native histogram support in Amazon Managed Service for Prometheus (12 Jun) – Vinod Kisanagaram, Priyanka Verma, Rohit Sharma
Conclusion
Observability launches in the past two months reflect a clear trajectory: AI-assisted operations continuing to move to production-grade, log analytics becoming more powerful without leaving the CloudWatch console, metrics scaling to match multi-account enterprise architectures, and monitoring reaching further into both synthetic and real-user experience layers.
To get started with these capabilities:
- Send OpenTelemetry metrics via OTLP directly to CloudWatch and query them with PromQL with no separate Prometheus backend needed
- Try the new Logs Insights query commands on your existing log groups
- Enable Session Replay in CloudWatch RUM to see exactly what your users experience
- Configure multilocation canaries to improve synthetic test coverage and reduce false alarms
- Explore AWS DevOps Agent custom agents to encode your team’s operational expertise
For the full list of recent launches, visit the AWS What’s New page filtered to Amazon CloudWatch. filtered to Amazon CloudWatch.
Looking for more?
Join our “I Didn’t Know Amazon CloudWatch Could Do That” webinar on August 11 to see these new features in action and use them to accelerate troubleshooting.
Authors