JUJING OBSERVE v2.3: Alert Deduplication, Deeper OTel Support, Faster Queries

JUJING OBSERVE v2.3 is out today. This release focuses on three things: stop alert spam, make OTel data actually usable, and make queries faster. Full changelog below.

Alerts: grouping and deduplication

Previously, ten disk alerts on one host produced ten notifications. v2.3 introduces alert grouping: multiple events from the same rule, grouped by the same dimensions (host, service, endpoint), collapse into a single notification that summarizes first-trigger time, the list of affected objects, and latest status.

group_by: [host, alertname]
group_wait: 30s
repeat_interval: 4h

We also added alert fingerprints. Repeated triggers with the same fingerprint update state without re-notifying; only a state change (firing → resolved, or an escalation) sends a new message. For teams using tiers, P2 alerts now default to a single hourly digest.

OpenTelemetry: native protocol support

Previously OTLP data went through a conversion layer before ingestion. v2.3 parses OTLP/JSON and OTLP/Protobuf directly, cutting the latency from Collector to queryable by roughly 40%. Also new:

• OTLP Logs support, with logs and traces correlated automatically via a shared trace_id;
• automatic indexing of span attributes, so you can search by any business attribute (order_id, user_id, etc.);
• a built-in probabilistic sampling policy for rate-based sampling, so high traffic doesn't blow up storage.

Queries and storage

Log queries now use SIMD acceleration on aarch64, cutting aggregate-query P95 latency by about 35%. The archive tier gains hot/cold tiering on object storage: data older than 30 days moves to cold storage automatically, meaningfully lowering cost, while queries remain transparent (they span tiers automatically).

Fixes: the DM cursor-leak bug (pre-8.1.1) that caused occasional query stalls; excessive memory usage for dashboards over large time ranges; and several frontend display issues.

Upgrade steps and compatibility

Upgrading from v2.2 is in-place; the database schema migrates automatically:

./upgrade.sh --from 2.2 --to 2.3

Notes:

1. Back up the database and your alert-rule configs before upgrading.
2. Alert grouping is the new default. Existing notification rules migrate to grouping automatically — verify the notification cadence in a staging environment first.
3. If you run a third-party Collector, the OTLP protocol is unchanged and no app-side config changes are needed, but upgrade to Collector 0.90+ to get OTLP Logs support.

Full docs and offline installers are updated on the download page; the Xinchuang (aarch64) build ships in lockstep.