ServiceRadar v1.5.3
This release fixes publishing a dashboard into a slot whose previous
package was deleted. It moves the edge-record streams to work-queue
retention, stops AWX hosts from being merged into the wrong Proxmox
guest and adds a repair step for records that already were, and caps
SNMP targets per profile. Rolling upgrades no longer fail cross-node
calls to peers that are still on the previous release.
Upgrade Notes
- The edge-record source streams are recreated once.
TELEMETRY_EDGE_RECORD_V1_BULKand its audit stream now keep only
records EventWriter has not yet written (WorkQueue retention, one
owning consumer, no age limit). On the first core start, the 5-minute
reconcile seals the old stream, waits for it to drain, then deletes
and recreates it. While it is sealed, agents keep records in their
spool and publish them afterwards, so nothing is lost. A new alert,
ServiceRadarEdgeRecordBacklogPressure, fires when either stream
stays above 80% of its byte limit for 5 minutes (#6501, #6552). - Each SNMP profile is limited to 10,000 targets per agent. The
limit is thesnmp_max_targetsruntime setting (agent, partition or
global). A profile over the limit is refused, not truncated: it
contributes no targets, the rest of the agent's configuration still
compiles, and core logs a warning. An enabled profile over the limit
cannot be saved (#6523). - SNMPv3 credentials accept only the listed security levels and
protocols. The form, rotation, the REST API and the SNMP profile
promote path reject any other value with a field error (#6522). - Queued exposure snapshot jobs move to a scopes table. The
migration turns pendingExposureSnapshotWorkerjobs into scope rows
that one publisher drains (#6547, #6562). - AWX/Proxmox merge repair. Installs where AWX hosts were merged
into the wrong Proxmox guest can run theawx-proxmox-unmerge
remediation step, thenawx-proxmox-verify. Execution is off by
default; seedocs/awx-proxmox-unmerge-runbook.md(#6558, #6565,
#6572). - New migrations:
20261110233000,20261111030000,
20261111040000,20261111090000,20261111160000,
20261111190000,20261111230000,20261111234313,
20261111235445,20261112193000.
Whats New
Dashboards
- Publishing a dashboard no longer crashes when the slug's existing
instance points at a deleted package. The slot is treated as free
(#6570).
Durable edge pipeline
- Edge-record source streams hold only unwritten records (#6501), with
one cutover lock per core and no shape reconcile on sealed streams
(#6552). - Default alerts fire for stuck or full agent spools and for held dead
letters (#6508). - EventWriter waits for its streams and consumers at startup instead
of logging errors (#6515). - Misrouted flow drain consumers are removed (#6571).
Device identity and Ansible
- AWX hosts bind to a Proxmox guest only on machine evidence. Node name
plus VM ID is no longer treated as identity, and Proxmox v3 node
references are parsed (#6551, #6555). - Automatic merges of records that hold different hypervisor guests are
refused (#6557). - Restoring a merged record no longer takes an address another device
holds (#6565). - Settings > Ansible reviews AWX memberships and template bindings, with
bulk approval (#6539, #6544). Quarantined memberships show their
reason and can be released (#6546, #6549).
SNMP
- Target devices are read a page at a time and credential lookups are
batched per page (#6520, #6523). - SNMPv3 security level and protocols are chosen from lists (#6522).
- The Cisco AES-192-C and AES-256-C privacy variants work end to end
(#6531).
Core reliability
- Cross-node calls between agent-gateway, core and web-ng survive a
rolling upgrade. When a peer still running the previous release lacks
the called function, the caller tries the other peers and waits
briefly for a capable one (#6498, #6500). - Exposure snapshots publish from one worker over a scopes table, and
stale scopes are skipped (#6547, #6562, #6567). - AshOban schedules run at priority 0 so they no longer go stale, and
superseded AWX observations are acknowledged (#6505). - Sweep lease refresh and re-issue no longer stall the maintenance
queue or deadlock (#6504, #6527, #6534). - Topology rebuild and graph reads stay under the gRPC receive limit
(#6528, #6533). - Withdrawn Proxmox results are acknowledged and duplicate cron ticks
stop (#6519). - StarRocks Stream Load separates DNS lookup from the connect timer
(#6518). - Release consoles print allowlisted Logger metadata, and three core
warnings now include their reason (#6532, #6535).
Integrations and plugins
- Armis retries transient request failures within a sync run (#6550).
- Add-on rollouts re-check agent compatibility before launch (#6513).
- v2 API endpoints accept the JSON:API media type (#6503).
Helm and release
- Optional compact JetStream profile for 10Gi volumes (#6543).
- Release packages ship signed SHA256 checksums (#6507).
NVR (disabled by default)