v2.4 adds log clustering, alert silence and escalation, and DM8 support; trace hydration is 45% faster, with fixes for timezone drift and duplicate webhook alerts.
No more paging through a wall of errors by hand. The system groups similar logs into clusters, annotating each with its occurrence count, first seen, and last seen times — so you can spot the high-frequency error at a glance. Clustering ignores fields that change on every line, such as timestamps and request IDs, and groups by the actual template difference, so one class of error isn't split into hundreds of entries just because the parameters differ. You can filter before clustering by service, time range, or keyword, e.g. cluster only the ERROR logs of one service within a given hour.
Example: after a release, the system surfaced two high-frequency clusters — one null-pointer, one connection-timeout. Clicking into either drills down to the specific logs and associated traces, and the debugging direction is clear on the spot. The practical value shows up in triage: instead of reading a thousand near-identical stack traces, the on-call engineer opens two clusters, sees one is a known null-pointer regression and the other is a new connection-timeout pattern, and starts fixing rather than searching. Clustering runs incrementally, so you can diff today's clusters against yesterday's baseline and spot new failure modes as soon as they appear. Clusters can be exported, which is handy for pasting into a post-mortem document or a ticket.
Silence and suppression rules are both editable in the UI and take effect immediately — no service restart required. Before you save a rule, it shows a live preview of how many active alerts it would currently suppress, so you can sanity-check it first.
Official support for DM8 in MySQL-compat mode means sovereign environments can deploy directly, covering Kylin V10, UOS, and Kunpeng/Phytium platforms. We also added a database connectivity self-check so you can see the connection test result during deployment, instead of discovering a dead connection only after the service fails to start. Deployments on DM8 run the same functional test suite as MySQL, so behavior parity is verified rather than assumed.
Log clustering alone changes how on-call engineers spend their nights: instead of grepping through noise, they open the top cluster and start on the actual problem. Alert silence and escalation close the loop on the second most common complaint after noise — alerts that nobody acknowledges. And if you are mid-migration to a domestic stack, DM8 support removes a blocker that has kept observability out of sovereign environments. The tracing improvements aren't vanity numbers either: a 45% faster hydration is the difference between a trace you actually open during an incident and one you skip because it takes too long. Put together, this release is less about new dashboards and more about making the existing ones actually get used.
This release is backward compatible — ingestion tokens, alert rules, and query syntax are unaffected. Private-deployment customers should contact ops for the upgrade package; cloud tenants are updated automatically. We recommend exporting your alert rules as a backup before upgrading, then running the built-in self-check afterward to confirm logs, metrics, and traces are all flowing normally.