ServiceRadar v1.4.64
The agent now hands every native add-on a persistent state
directory, so the anomaly detector keeps its baselines across
add-on upgrades and agent restarts instead of cold-starting;
drift findings draw their episode on the chart for real event
rows; and Create alert rule is back on the event details page.
Whats New
1.4.64
- Agent: every native sidecar add-on receives a persistent
state directory,<runtime root>/addons/<id>/state(mode
0700, owned by the user the add-on runs as), created before
each spawn and exported asSERVICERADAR_ADDON_STATE_DIR. It
sits beside the versions tree, so an artifact upgrade or
rollback never touches it (PR #431). - anomaly add-on 0.3.10: the restart checkpoint no longer
depends on an operator settingcheckpoint_path. Resolution
is explicitcheckpoint_path, elsecheckpoint.jsonin the
agent-provided state directory, else off. A missing parent
directory is created on the first write, which previously
left an operator-chosen path silently unwritten. Without a
checkpoint, every restart cold-started the detector and its
open episodes were stale-closed by core instead of recovered
(PR #431). A checkpoint path applied to a running add-on by
a config-only update now takes effect on the open metric
feed; before, the feed loop kept the settings it had at
stream open and nothing was written until the next restart
(PR #433). - Web: the drift episode band, baseline and sustained-level
lines added in 1.4.63 now render for real event rows, whose
verdict payload lives underunmapped.anomaly; the first
version only matched the projectedmetadata.anomalyshape
(PR #429). - Web: Create alert rule is back on the event details page for
operators and admins. The modal is pre-filled from the event
(provider, subject, severity, message conditions, a slugified
name anddevice.uidgrouping), validates the threshold and
window, and saves a stateful alert rule before opening the
Alerts tab (PR #425, fixes #345).
Upgrade Notes
- The state directory is created by the agent in this release;
add-ons older than 0.3.10 ignore the variable and behave as
before. The first checkpoint is written after 100 feed
frames, so the first restart after this upgrade still
cold-starts once; later restarts re-warm. - A profile or assignment that sets
checkpoint_path
explicitly keeps that path. Remove it to use the agent
default.