v2.5.0 is out with metric pre-aggregation and dynamic sampling, regex pushdown in log search, faster trace hydration, and fixes for timezone and alert deduplication issues. Upgrade notes included.
v2.5.0 is released today. This version targets two problems that came up repeatedly in feedback: slow metric queries at scale, and fixed sampling ratios that either waste data or drop it when traffic swings. The full change list is below.
Metric pre-aggregation: frequently queried metrics are now pre-aggregated over time windows, so dashboards and alert queries return in seconds instead of scanning raw data. Pre-aggregation rules are configurable in the UI, support grouping by label dimensions, and also work on cold data.
Dynamic sampling: trace sampling ratios are no longer hard-coded. The platform adjusts automatically based on traffic and error rate — normal sampling in steady traffic, then automatically raising the ratio when it detects a traffic surge or rising error rate, keeping 100% of abnormal traces so you never lose a critical trace at the wrong moment.
Regex pushdown in log search: regex matching is now pushed down to the storage layer, noticeably speeding up regex search over large indexes, and combines with histogram() bucketing for finer distribution analysis.
Alert rule import/export: rules can now be imported and exported as YAML, making it easy to migrate between test and production environments and to version them.
This is a backward-compatible upgrade: ingestion tokens, query syntax and alert rules are unaffected. Self-hosted customers should contact support for the upgrade package; cloud tenants are updated automatically. Upgrades from v2.3.x must go through v2.4.0 first. Before upgrading, export a backup of your alert rules, then use the self-check script after the upgrade to confirm logs, metrics and traces are all reporting normally.