Skip to content

v1.3.2

v1.3.2: Causal anomaly detection, protobuf metric ingestion, and observability
@mfreeman451 mfreeman451 tagged this 15 Jun 00:24 
performance hardening.

**1.3.2**
- Added the causal anomaly and capacity forecasting spine: JetStream-backed
  metric ingestion, DeepCausality-backed anomaly evaluation, checkpointed
  context recovery, per-series configuration, clearing signals, capacity
  forecast rows, and the observability health/capacity UI surfaces.
- Cut metric producers and consumers over to the canonical
  `serviceradar.metric.v1` protobuf envelope, including sysmon, SNMP, OTEL,
  plugin, Wasm, and native add-on publishing paths, with hard rejection for
  stale JSON metric payloads and mis-bucketed OCSF/metric events.
- Hardened metric semantics for SNMP/plugin counters with raw cumulative
  counter preservation, reset-default delta normalization, configurable SNMP
  interface metric allow-lists, inferred metric-type telemetry, and
  NATS-Msg-Id de-duplication on the metrics stream.
- Improved anomaly engine performance and safety with shard-ready native
  batches, compact event input, reduced wrapper churn, bounded event-token
  pruning, checkpoint restore, eviction protection for open anomalies, and
  production benchmark instrumentation.
- Fixed observability and security dashboard load pressure by making health,
  attributed-flow, and security-finding views load asynchronously, trimming
  large security payloads, preserving dashboard frames on refresh errors, and
  reducing service-state query pressure.
- Tightened demo/runtime reliability with managed-CNPG network policy access,
  Falco sidekick ingress, stale DIRE device cleanup, Proxmox NIC identity
  preservation, causal anomaly demo enablement, and trace summary refresh
  worker timeout/index hardening.
Assets 2
Loading