What's new this week

This section summarizes what was released this week.

Fast lane GA features

These features are generally available for all customers. Slow lane customers and tenants will have these features automatically enabled on November 25, 2026.

service-explorer: The service map now shows uninstrumented peers such as third-party APIs and services without OpenTelemetry instrumentation. Graph IDs carry the service type, and the Service Inspector can group by port. Engineers can see a service's complete set of dependencies, including the ones that are not traced. Cmd-click or Ctrl-click on a service in the catalog now opens it in a new tab.
service-explorer: Service monitoring is generally available. Monitors can bind to a service, and the service map surfaces the monitors covering each service. Teams can see which services are monitored and set up coverage for the ones that are not, without leaving the map.
dashboards: The improved dashboard editing experience is generally available. Editing a dashboard is faster and more consistent for all customers.
dashboards: Dashboards can now be exported as PDF. Lets teams share a point-in-time snapshot of a dashboard with people outside Observe or attach it to reports and incident reviews.
data-links: Data link URLs can now be dynamic (parameterized), and drilldowns are easier to discover from the places where data links are defined.

Public preview and private preview features

features are available by invitation only. features are available by inquiry.

ai-sre: AI-SRE is available in the Service Inspector side rail for selected customers, so engineers can investigate a service without leaving the inspector.

Bug fixes

The following issues are fixed in this week's release:

ingest: Token verification now returns a retryable 503, instead of a 401, when the customer or user lookup fails because of a temporary backend problem. Ingest clients retry instead of dropping data. (OB-70744)
ingest: Ingest tokens are now extended when they are used successfully, so tokens in active use no longer stop authenticating three years after creation. (OB-70305)
service-explorer: Services that have no health entry are shown as unmonitored instead of uninstrumented. (OB-70189)
service-explorer: A namespace-filtered service map now shows health for services reached through cross-namespace edges. (OB-70188)
apm: Services health reflects monitor and alarm changes on the next request instead of after the cache expires. (OB-70190)
reference-tables: Object tags saved on a reference table's properties page are now kept after a refresh. (OB-67873)
datasets: The dataset delete confirmation no longer lists auto-created metric SMA datasets as dependents. (OB-68956)
compiler: Queries on datasets with a grouping key, alias, content type, or field mappings now report that metadata correctly. (OB-69397)
ai-sre: Alert investigation replies no longer repeat credentials or other confidential values found in logs, spans, or query results.

REST API changes

REST API surface changes for this release are published separately in the REST API Changelog, which covers version changes, breaking changes, deprecations, and migration.

Terraform provider

The following updates are available in the Observe Terraform provider.

v0.14.74 (2026-10-21)

New features

Pin OPAL versions on workspace objects.

v0.14.73 (2026-10-21)

Bug fixes

Build with Go 1.26.10 to distrust compromised eMudhra CAs.

Coming soon

The following GA features are already available for fast lane customers, and will be enabled for slow lane customers and tenants next week.

ai-sre: Claude Opus 5.5 with high reasoning effort for AI-SRE investigation and monitor-authoring workflows.
visualization: Optimized Vega legend rendering for charts with large or complex legends.

APM Tracing Content Rollout (Week of 2026-10-26)

Latest in source: 2.6.0
Reference version (most-deployed): 2.5.5 on the clusters with an upgrade window this week

Non-breaking changes rolling out this week

External calls are emitted as first-class edges, and External Call Metrics is unioned onto the invocation graph. ProduceExternalCalls and UseExternalCallMetricsDataset will be flipped on by default in the next scheduled breaking upgrade.
Tracing/Trace gains a search-optimization equality index on trace_id.

Breaking changes flipping this week

ProduceExternalCalls (false → true) and UseExternalCallMetricsDataset (false → true) are flipped for some clusters. External calls move to a new Tracing/External Call Metrics dataset and are retired from Tracing/Service Metrics. An external peer is now counted once, at the innermost span of the call, instead of twice. Service edge metric counts and sums are scaled by canonical-trace sampling, and those edges carry parent.span.name and parent.span.type tags. Tenants with content upgrades paused are not upgraded.