Python tracing goes span-native as Sentry preps its next major

Two Python tracers turn a corner

The biggest shift of the week is structural. Sentry's Python SDK moved span streaming to the default tracing model, retiring transactions and replacing the flat send_default_pii flag with a fine-grained data_collection setting — the first public pre-release of the November 3.0 major, complete with a 2.x-to-3.x migration guide. On the stable line, Sentry Python records Anthropic reasoning parts in both streaming and non-streaming output, and fixed aiohttp clobbering an existing baggage header, sampling that ignored the incoming sample rate, and WSGI integrations overwriting server.address.

Datadog's dd-trace-py kept pace across a handful of updates. The headline is Python 3.15 support plus aiohttp threat detection and blocking, SQL injection protection for byte queries, and MCP 2.x tracing — alongside the removal of the legacy pytest plugin, so the DD_PYTEST_USE_NEW_PLUGIN variables are now inert and should be cleaned out of CI configs. Two fixes round out the week: repeated Git history fetches no longer slow successive pytest runs in a shallow checkout, and remote config handling no longer discards sibling settings when one configuration is updated or deleted, with Falcon ASGI startup repaired.

Datadog's agent and browser surface settle down

The Datadog Agent reverted config streaming to off by default after 7.85.0 and 7.85.1 could leave trace-agent and process-agent ignoring DD_* environment variables in their own containers on Kubernetes — a rollout worth checking if you enabled it. The same window brought FIPS-mode CSPM evaluation fixes and a 32-bit ARM IoT startup crash repair, including cleaner oscap-io temp files. On the browser side, the RUM SDK lets you set user and account at init, survives an inaccessible document.cookie, splits oversized page-exit batches across beacons, and instruments only the console APIs it collects.

Instrumentation reaches further across the stack

Sentry JavaScript added cron monitoring for NestJS and cron-package jobs, replacing the deprecated Sentry.cron.instrumentCron, and unified Cloudflare Workers on the shared request handler so client.address comes from CF-Connecting-IP by default. Its Anthropic integration now tracks @anthropic-ai/sdk 1.x without duplicate spans. Axiom's Go SDK added monitor threshold levels and closed an ingest goroutine leak — note that per-level notifier routing awaits a server change — while Axiom Console came to phones with SLOW/FAST span chips in traces and Jira Service Management alerting.

Dash0's week was mostly about reach. The Operator gained automatic .NET 6 and 7 instrumentation through a new Dash0 OpenTelemetry distribution, a timely move given upstream's planned .NET 10/11-only focus as 8 and 9 hit EOL. The Operator also began deprecating container.image.tag in favor of container.image.tags, with a v0.165.0 cutoff, and Agent0 can read Kubernetes cluster state during investigations via the operator, including namespaces with no telemetry. Elsewhere in Dash0, AutoMerge scores PRs with AI evaluators in a fail-closed dry-run mode, custom Agent0 skills became code via Terraform and the CLI, 28 eval templates prefill span filters and judge prompts, and the CLI added auth-token, AWS scrape-config, and Agent0 skills commands.

Query migration and agent evaluation

Datadog's blog carried the strategic news: Translate and Import are generally available at no extra cost, converting SPL into log searches and translating Splunk alerts and dashboards into drafts. Agent Observability added Online Experiments for comparing agent variants on live traffic and Fireworks AI as a model provider. For Kubernetes teams, DatadogInstrumentation consolidates checks, logs, and APM into one CRD that applies without app rollouts, and Datadog detailed FETCh turning TLA+ specs into production runtime monitors, which caught a year-old routing bug on Husky. Seer Agent also connects to Grafana to correlate Sentry telemetry.