Releases Index

BetaWeekly digests are a beta — we're trying something new. Feedback welcome.

Profiling goes mainstream, from eBPF daemonsets to OpenTelemetry explorers

November 9–15, 2026

Dash0DatadogSentryAxiom
Dash0, Datadog, Sentry, Axiom

Dash0 shipped a complete profiling story — an eBPF Parca profiler managed by its operator plus a Profiling Explorer for OpenTelemetry profiles — while Sentry's Python SDK pre-release made span streaming the default tracing model and Datadog's tracer majors moved toward Python 3.15.

The week at a glance

Replay the week
FeaturesEnhancementsFixesMajorLandmarkTile size reflects impact

Profiling becomes a first-class workload

Dash0 had the week's most coherent story: profiling that you can turn on, collect, and actually explore. An operator-managed Parca eBPF profiler now runs as a DaemonSet and collects CPU profiles from monitored workloads with no code changes, enriching them with Kubernetes metadata and resolving native frames through a debug-info proxy — opt-in, and requiring privileged access, so treat it as a deliberate cluster decision. On the reading end, the new Profiling Explorer for OpenTelemetry profiles offers flame graph, table, flame chart and sandwich views, with Sum by splits and a differential flame graph for baseline-versus-comparison work. Profiles can come from the operator's integration or a separately deployed profiler, so the pair works for teams already instrumenting themselves.

GitOps-friendly instrumentation and a smaller CRD

The rest of the Dash0 week leaned toward making the operator behave in real delivery pipelines. A new pods instrumentation mode leaves deployments, stateful sets and cron jobs untouched and instruments each pod at creation — which is exactly what Argo CD, Flux and Terraform users need to stop seeing perpetual drift — and the same mode calls for running the operator manager highly available. The release also deprecates the container.image.tag resource attribute in favor of container.image.tags, so queries and dashboards referencing the old attribute should be updated before it disappears. Separately, the Dash0OperatorConfiguration CRD shrank from roughly 256 KiB to 200 KiB, enough that client-side kubectl apply no longer trips the annotation size limit, alongside a new Helm value for prefixing operator-managed resources. On the CLI side, Dash0 added signal-to-metrics rules with a dry-run test and an experimental diff preview, plus commands for managing generative AI evals.

Tracers modernize: new Python, new defaults

Datadog's APM week was mostly about compatibility headroom and defaults. Support for Python 3.15 lands in dd-trace-py, though some integrations aren't validated on it yet — and the same release removes the legacy CI Visibility pytest plugin, so DD_PYTEST_USE_NEW_PLUGIN and its beta sibling are now inert and should be cleaned out of your configuration. Two smaller patches cleaned up real-world pain: uWSGI startup hangs and profiler crashes when ddtrace is imported in the master process, and a memory leak in @tracer.wrap contextvars copies that pinned finished calls in memory. On the JavaScript side, dd-trace-js turned allocation profiling off by default, made OTel APM RED metrics an explicit opt-in, and added tRPC and async-mutex tracing.

The Agent got two meaningful hardening releases. Health Platform is now disabled by default on Kubernetes, because reporting Autodiscovery errors could stall Autodiscovery for seconds and cause workloadmeta to drop event bundles; other environments keep it on, and Kubernetes users can restore it via configuration. The trace-agent also validated payload lengths before allocating for declared array and map headers, capping non-string-table headers at 500,000 elements.

SDK majors and pipelines

Sentry's Python 3.0.0b4 is the one to read carefully: span streaming replaces transactions as the default tracing model, and a fine-grained data_collection setting supersedes the flat send_default_pii — a migration guide exists, and this is a pre-release, so plan the upgrade rather than absorb it. The 2.x line still got Bedrock Converse and ConverseStream tracing with gen_ai model, token and tool attributes, plus credential-free GraphQL transport URLs. Sentry JavaScript enables hidden source maps in the bundler plugins when the bundler has none and deletes them after upload, and on the UI side Sentry made fingerprint rules easier to write and debug.

Elsewhere, Datadog's Cloud SIEM now resolves signals to a named person via identity sync from Google Workspace, Entra ID and Okta; Kubernetes inference workload monitoring entered Preview for vLLM on KubeRay, Dynamo and KServe; and Axiom made the Console usable on phones, foldables and tablets.

AI-generated digests may contain mistakes.
Releases covered18