github carverauto/serviceradar v1.4.102

56 minutes ago

ServiceRadar v1.4.102

This release makes the durable edge-record pipeline the only way agent
data reaches the platform, on every install, with no switch. Agents
spool everything they collect to disk and deliver it as signed records.
The legacy status path no longer carries any durable data. Alert
evaluation now lags under overload instead of dropping work, and a
self-hosted install on default values stores telemetry and accepts its
first upgrade.

Upgrade Notes

  • Upgrade your agents along with the platform. Agents older than
    this release send their data on the legacy status path, which the
    gateway now refuses. Until an agent is upgraded, its metrics, sweep
    results and other data are counted and shown on the agent as
    agent_upgrade_pending in the edge pipeline readiness report and on
    the agent page, but they are not stored (#6340, #6341).
  • Remove edgeRecords from your values. The pipeline is always on,
    and the chart now refuses to render while any edgeRecords key is
    set, including an empty one. Delete the block, along with any
    edge_record_sender opt-ins (#6338).
  • The edge issuer key moves into CNPG. On first start, core imports
    the existing edge-issuer-seed from the generated Secret once, and
    keeps that key afterwards. Installs that set
    secrets.autoGenerate: false and supplied their own seed get a new
    key instead (#6320, #6330, #6338).
  • Every NATS hub session reconnects once. On the first upgrade,
    core moves each session, including edge leaf connections, to the
    current certificate grant. That happens in batches of 64 per scan per
    core replica (#6336).
  • Apache AGE is removed. Migration 20261104004947 drops the AGE
    graph storage and the age extension. Topology lives only in Dgraph,
    and the chart refuses to render without Dgraph (#6306, #6310).
  • New migrations run through 20261105100000; the chart's expected
    migration version moves with them.
  • Self-hosted installs on default values now start storing
    telemetry: EventWriter uses core's NATS connection when
    EVENT_WRITER_NATS_URL is unset. They can also run their first
    helm upgrade, which the CNPG Secrets used to block (#6348).

Whats New

One durable path for agent data

  • Agents always run the edge-record spool and sender. Each producer
    emits records only while it holds core's grant, and holds its data,
    rather than falling back, when it does not. The spool starts on a
    1 GiB budget that core can change at runtime (#6325, #6340).
  • Sysmon, SNMP, ICMP, MTR, rperf and sweep metrics, unleased sweep
    results, plugin and add-on telemetry, Wasm plugin results, discovery,
    sync, workload identity, Kubernetes endpoints, endpoint inventory,
    bumblebee scans, flow attribution and OTLP relay frames all travel as
    typed edge records (#6255, #6264, #6280, #6284, #6289, #6305, #6311,
    #6312, #6313, #6314, #6315, #6316).
  • The gateway publishes no durable data from the status RPCs. Only
    status summaries and control traffic remain there (#6341).
  • Leased sweep slots run the signed plan and its check sets, and core
    decides completion from the chunks it received (#6257, #6259, #6260,
    #6272, #6275, #6302).
  • The spool pauses admission at its high-water mark and shares bytes
    fairly between producers. A reconnecting agent renews its grants and
    drains its backlog (#6269, #6274, #6296).
  • Edge records whose projection keeps failing are dead-lettered and can
    be replayed. A read-only readiness report shows each agent's pipeline
    condition (#6276, #6295, #6318).

Signed contract registry

  • Core derives the contract registry from the release catalog and
    signs each registry epoch. The gateway admits a record only against
    the signed epoch it names (#6319, #6320, #6330, #6331).

Alerting and EventWriter

  • Under overload, alert evaluation lags instead of losing work.
    Refused admissions are kept in a durable backlog and admitted in
    order as capacity frees. A backlog health check reports when the
    backlog grows (#6342).
  • Each consumer group runs its own pipeline, so one stalled batcher no
    longer starves the others. A consumer that loses messages before
    delivery raises an alert (#6332).
  • Alert admission is chunked and claimed before dispatch, and
    anomaly-open rules admit only events on their own subject (#6304,
    #6317, #6321).

Topology: Dgraph only

  • Apache AGE is retired, and mapper topology is projected to the graph
    outside the ingestion lane (#6299, #6300, #6306, #6310).

NATS and certificates

  • Built-in identities get their chart permissions under the auth
    callout, and hub sessions move to the current certificate grant
    instead of keeping a stale one (#6326, #6336).
  • Hub certificate grants use a jittered lease, and managed certificates
    are issued on the authenticated session node (#6252, #6254, #6301).

Availability

  • When the agent that decides a device's availability goes silent, the
    device is re-resolved within one sweep interval instead of waiting up
    to two intervals plus five minutes for its ledger entry to expire
    (#6346, #6352).

Fixes

  • SSO toggles persist and the Entra preset is restored (#6265).
  • Sysmon metrics are attributed to the attested agent device (#6283).
  • Edge Sites and Data Collectors show managed installation state, and
    service certificates are paginated (#6290, #6307).
  • The log collector owns its health listener and keeps mTLS (#6303).
  • macOS and Windows installs create the agent runtime directory
    (#6323).
  • Image builds fetch pinned Ubuntu packages from Launchpad when the
    snapshot archive is unavailable, and retry Bazelisk downloads for
    longer (#6350).

Don't miss a new serviceradar release

NewReleases is sending notifications on new releases.