v2.6.0 is live: slow-query analysis with execution plans, alert silencing windows, and OTel improvements including gzip transport, auto service mapping, and logs-to-traces linking. Includes the upgrade path and caveats.
v2.6.0 is live. This release targets three long-standing complaints: slow queries showed the result but not the process, there was no way to silence alerts temporarily, and some edges of the OpenTelemetry ecosystem weren't covered. Full changelog below, plus upgrade guidance.
Slow databases are the root cause of many incidents. Previously you could only see that a query took 3 seconds — never the execution plan. v2.6.0 adds an "Application Performance → Slow Queries" page:
The ranking matters more than the plan itself: sorting by rows-scanned surfaces queries that quietly chew through the database even when each one is individually "fast." Clicking a slow query opens the calling service and the trace that triggered it, so you can answer "which code path issues this query" without asking the dev team. Plans drift as data volume grows, so capturing them in-band beats asking a DBA to re-run EXPLAIN after the fact.
During routine releases, load tests, and data migrations, alerts pile up. You can now silence them in bulk by time window, alert rule, or service under "Alert Center → Silence":
silence:
matchers:
- name: service
value: "payment-*"
startsAt: "2026-10-20T01:00:00Z"
endsAt: "2026-10-20T03:00:00Z"
comment: "release window"
Silenced alerts don't send notifications, but they are still recorded, so after the release you can review "how many would have fired during this window" — useful for tightening thresholds that would have been noisy. Silences support wildcard matchers and a comment field, and they expire automatically at endsAt, so you can't forget to turn alerts back on.
service.name to project/service. New services appear automatically; you can still override the mapping manually.status field reported by the Python SDK was misclassified as an error.field IN (...) and NOT operatorsUpgrade: use "System Settings → Version Upgrade" in the console for a one-click online upgrade; storage nodes restart in a rolling fashion and data collection is not interrupted. Customers on offline (Xinchuang) environments should download the observe-2.6.0-arm64.tar.gz bundle, run docker load, then follow the same upgrade flow.
Versions ≤2.4.x must upgrade to 2.5.0 first — you cannot jump directly to 2.6.0. Back up metadata before upgrading; the entry point is "System Settings → Backup". Full release notes and checksums are on the download page.
All three headline features are enabled by default on new installs and available to existing tenants after upgrade — no license changes. The slow-query page needs the execution-plan switch turned on per data source; the alert-silence API accepts the YAML shown above as well as a UI form. If you're on gRPC OTLP today, the gzip improvement only applies to the HTTP path, so switch exporters to otlphttp to take advantage of it.
Alert rules and silence records from 2.5.x carry forward unchanged. One behavioral change to note: auto service mapping now runs on first sight of a new service.name, so services that previously needed a manual mapping entry will appear on their own — delete any leftover manual entries to avoid duplicate projects. Full release notes, including the complete bug-fix list and SHA-256 checksums for the online and offline bundles, are on the download page. As always, report issues through the console's feedback channel or the support portal.