A quarter of shipping, condensed: what went GA, what got faster, and what we killed along the way.

We don't do a big annual keynote. Features ship when they're ready and we write them up as they land. Every few months it's worth stepping back and looking at the quarter as a whole, because individual changelog entries hide how much a query planner improvement compounds with everything shipped around it.

What went generally available

Tail-based sampling and continuous profiling both graduated from beta this quarter, along with dashboards-as-code and HelixQL-based alert routing. Combined, design partners running all four report a median 45% reduction in signal storage cost with no loss in incident detection time.

The query planner got quietly faster

Less visible: we rewrote the join planner's cost estimator, which cut p95 query latency for cross-signal joins by 38% without anyone changing a query. You can see the improvement directly:

from traces
| where duration > 1s
| join logs on trace_id
| summarize count() by bin(1h)
// p95 planning time: 340ms -> 210ms this quarter

What we killed

The legacy "classic dashboards" UI editor is deprecated as of this quarter in favor of dashboards-as-code; it stays read-only until Q3 for teams mid-migration. We also removed the old fixed-rate-only sampling mode — every workspace now defaults to tail-based sampling, with fixed-rate still available as an explicit opt-in for teams that genuinely want it.

What's queued for Q2

Multi-region ingest and retention tiering are both in active development and expected to land within the next two quarters, along with a second pass on the query planner focused on aggregation-heavy dashboard queries rather than joins, since that's now the largest remaining share of p95 query latency across the fleet. Expect a similarly blunt writeup once those ship, including what didn't survive first contact with real traffic.

Support ticket volume, as a sanity check

We track support ticket volume per feature area as a rough proxy for whether a launch actually reduced friction rather than just adding surface area. Tickets tagged sampling and alert routing both dropped in the four weeks after this quarter's launches, even as adoption climbed, which is the pattern we want to see — a feature that's genuinely solving the underlying problem should show up as fewer questions, not more, once the initial onboarding wave passes.

  • GA: tail-based sampling, continuous profiling, dashboards-as-code, HelixQL alert routing
  • 38% faster p95 planning time on cross-signal joins
  • Deprecated: classic dashboard editor (read-only through Q3)
  • Removed: fixed-rate-only sampling as the default