← Back to blog
Changelog 4 min read 炬鲸团队

v1.5.0 Release: Alert Grouping & Noise Reduction, Log Context, and Metric Drill-Down

Observe v1.5.0 ships alert grouping, inhibition, and silencing; log search gains context jumps and trace_id reverse lookup; metric panels drill into logs and traces; and tracing gets configurable sampling—plus upgrade notes and compatibility caveats to plan your rollout.

This release targets two long-standing complaints: alerts are too noisy, and troubleshooting means hopping between the log, metric, and trace pages. The rewritten alert engine merges identical alerts by default and lets root-cause alerts suppress downstream noise; log context and metric drill-down shrink the distance from "something's wrong" to "here's why" to a single click. Everything below is built around those two.

Alert engine: grouping, inhibition, silencing

v1.5.0 rewrites the alert engine's notification pipeline with three new capabilities:

  • Grouping: merge same-source alerts into one notification using group_by dimensions, with a count attached—within a 30-second window, identical alerts fire only once;
  • Inhibition: declare source-to-target relationships so a root-cause alert automatically suppresses downstream cascade alerts, with chaining supported;
  • Silencing: silence by time window or by label, with an expiry, so planned maintenance no longer produces false alarms.

Existing alert rules from earlier versions need no changes—grouping logic activates automatically after the upgrade. Enable inhibition and silencing only after you've updated collectors to the latest version, or the configs won't propagate to edge nodes. The notification payload now carries the grouped count and a link to the affected service, so the person on call can judge severity at a glance.

Log context jump

Troubleshooting often means finding the lines around a suspicious log entry. v1.5.0 adds a "context" button on the log detail page, expanding the 50 entries before and after with one click, plus reverse lookup of same-trace logs by trace_id and same-host logs by host. Combined with SQL-style search, once you locate the anomalous line you can reconstruct the scene immediately instead of paging through timestamps by hand. The context window is configurable, and the trace_id and host lookups reuse the same query syntax as the rest of search, so there's nothing new to learn.

Metric drill-down

Charts on the monitoring dashboard are now drillable: select an anomalous time range on a metric curve and jump straight to the log list for that window, or open the trace view for that period. What used to be "metric anomaly → switch to logs → scrub through time" is now a single click. Drill-down keeps the exact time range you selected, so the logs you land on are the logs from the anomaly—not a fresh query you'd have to re-narrow.

Trace sampling

To control storage costs under high traffic, tracing now supports configurable sampling: 10% probability by default, with forced retention by service or by error—failed traces are kept in full, normal ones are sampled proportionally. Sampling rule changes take effect immediately without a restart. Operators can also exclude health-check endpoints from sampling so synthetic traffic never pollutes the trace view.

Other fixes and improvements

  • Fixed high memory usage in log search over very large time ranges, noticeably improving query stability;
  • Optimized collector reconnect handling, so data is no longer dropped during network blips, with a configurable local cache ceiling;
  • Added a batch of out-of-the-box alert rule templates (CPU, memory, disk, slow queries) that new tenants can apply in one click;
  • Fixed incorrect field mapping for the DM slow-SQL view on certain versions.
  • Optimized console home-page load time, so large tenants open faster;
  • Alert notifications can now embed jump links that go straight to the relevant chart or log page.

Next up

v1.6 focuses on alert root-cause analysis: attaching the related logs and trace entry points directly to the alert notification, merging "what happened" with "why it happened" in a single message. We'll also ship the cross-midnight timezone fix for silences and add metrics collection for more domestic databases, including KingbaseES and GBase.

Upgrade and compatibility

  • Upgrade path: swap the single binary and restart—the data format is backward compatible, no migration needed;
  • Dependency change: collectors should upgrade to v1.5.0 in lockstep, otherwise inhibition/silencing/sampling configs won't propagate to edge nodes;
  • Known issue: silence windows crossing midnight need a manual offset for tenants on UTC; fixed in the next release;
  • Deprecation: the old alert.group field is removed in v1.7—migrate to group_by.

The full changelog is in the docs. If you hit problems during the upgrade, open a ticket in the console with your version number and trace_id attached.