Searching DATEV's logs for a customer ID used to mean processing almost a terabyte in three minutes; now it scans about 100 GB in under three seconds. That's the example Grafana Labs used on October 20, 2026, at ObservabilityCON, to launch full-text search in Grafana Cloud Logs, generally available at no extra cost. It came with company: automatic metric pruning, two new trace sampling strategies and agents that watch your telemetry are all GA too.
What's new and what's actually ready
Grafana bundled production-ready features with previews in the same announcement. This is the split that matters before you plan anything:
| Feature | Status | What it does |
|---|---|---|
| Full-text search in Cloud Logs | GA, no extra cost | Finds specific lines (request or customer IDs) without scanning the whole volume |
| Adaptive Metrics auto-apply | GA | Applies metric aggregation recommendations automatically |
| Volumetric and diversity sampling (Adaptive Traces) | GA | Spread the sampling budget across services and routes |
| Query dropped traces | GA | Fetches any trace from the last 24 hours by ID |
| Assistant Watchers and Investigations | GA | Agents that watch signals and open investigations |
| Adaptive Logs Archive | Public preview | Sends dropped logs to a bucket you own |
| Fleet-managed Adaptive Logs | Public preview | Filters logs in your own collectors, before they leave your network |
| Archived log replay | Private preview | Brings archived logs back on demand |
| Managed Sandboxes | Public preview | The assistant prepares a pull request with a proposed fix |
Grafana says Adaptive Telemetry customers already cut their telemetry spend by 30% to 50% on average. The new features push the same idea further: your bill should track the value of your data, not how much of it you ingest.
Traces: sampling without running your own pipeline
Until now, tail sampling meant building and babysitting your own pipeline. Adaptive Traces does it inside Grafana Cloud and adds two strategies. Volumetric sampling takes a single percentage and spreads that budget across services and request types, adjusting as traffic shifts; in Grafana's testing it delivered about 25% higher information density than probabilistic sampling at the same volume. Diversity sampling keeps one trace every 15 minutes for each unique combination of attributes such as service, route and status code, so a brand-new code path shows up without hand-written rules.
The safety net is querying dropped traces: if a user hands you a trace ID from the last 24 hours, you can open it even if sampling threw it away.
Logs: search, archive and drop at the source
Full-text search is built on technology from Logline, which Grafana Labs acquired in April 2026, and cuts the volume scanned for newly ingested logs by 80% to 90%. For teams paying per gigabyte scanned, or hunting a request ID in the middle of an incident, this is the change with the most immediate payoff.
The other two pieces are still in preview, but they change how you design the pipeline: Archive routes what used to be dropped to object storage you control, and the Fleet Management version applies your drop rules in the collectors you already run, so those logs never even generate egress traffic.
Agents that watch without alert rules
Watchers are always-on agents: you describe what worries you, pick the data sources, and the assistant builds a baseline from historical telemetry to catch sustained errors or slowdowns that no predefined rule covered. When they find something critical, they can kick off an Investigation, which works through metrics, logs, traces and profiles, tests hypotheses and hands back the queries behind its findings. Watchers can also read BigQuery, so a drop in conversions can land in the same investigation as an error spike.
How to turn it on
- Auto-apply is enabled per Adaptive Metrics segment. Choose coverage-first, which follows real usage, or cost-first, which guarantees the active series count never grows, and list the metrics to exempt.
- Volumetric sampling: set the percentage of traces you want to keep and let Adaptive Traces distribute it.
- Archive and Fleet are switched on from Adaptive Logs; Replay requires signing up for the private preview.
Gotchas
The one that can really hurt is auto-apply: it applies new recommendations with no human review. Grafana says it won't break existing dashboards or alerts, but metrics that feed SLOs or external reports belong on the exemption list from day one. Dropped traces can only be fetched by ID: they don't show up in search or aggregations, and they're gone after 24 hours. And until Replay ships, an archived log is good for audits, not for debugging a live incident.
Sources
- ObservabilityCON 2026: all the announcements, Grafana Labs, October 20, 2026.
- Break the link between data volume and your observability bill with Adaptive Telemetry, Grafana Labs, October 20, 2026.
- Grafana Assistant and autonomous operations, Grafana Labs, October 20, 2026.