Files
aevolution/deploy/kubernetes/monitoring/config/evo-region.rules.yml
T
dingoandClaude Opus 5.5 b4493b3d0f Add M8.5 observability: pinned Prometheus, tested alerts, capacity metrics
An AEvolution-only Prometheus 3.13.3 and Alertmanager 0.34.1, pinned by
digest and delivered by Argo (ADR-0072), scrape each region's operations
Service through static targets (no API access, no RBAC); the region
namespaces admit port 9100 from Prometheus only. Sixteen alert rules
cover authority, lease, commits, backlog and object capacity and are
proven by promtool injection tests in CI with tools copied from the same
pinned images. Two new gauges expose object volume free space and the
reserve. Alerts route to a null receiver until a destination is chosen;
no broker is adopted and the database outbox stays.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-30 06:16:36 -04:00

122 lines
6.6 KiB
YAML

# M8.5 region host alerts (ADR-0072). Thresholds follow the measured behaviour:
# the production lease is 20 s renewed every 5 s (15-20 s remaining when healthy);
# batches are at most 64 ticks with a 1 s flush; an idle region (no inputs, no
# gateway yet) always accrues tick debt, so debt and staleness only alert while
# inputs are waiting. Unit tests: evo-region.rules.test.yml (promtool).
groups:
- name: evo-region-authority
rules:
- alert: EvoRegionDown
expr: up{job="evo-region"} == 0
for: 2m
labels: {severity: critical}
annotations:
summary: "Region {{ $labels.environment }}/{{ $labels.world }}/{{ $labels.region }} is not scrapeable"
description: "No operations endpoint answers. The process or its pod is gone; a forced deletion recovers in about 20 s, so 2 minutes is a real outage."
- alert: EvoRegionNotReady
expr: evo_region_ready == 0
for: 2m
labels: {severity: critical}
annotations:
summary: "Region {{ $labels.environment }} has no authoritative host"
description: "The process is alive but holds no lease or has not verified its head. Check the phase metric and the host log."
- alert: EvoRegionQuarantined
expr: evo_region_phase{phase="quarantined"} == 1
for: 1m
labels: {severity: critical}
annotations:
summary: "Region {{ $labels.environment }} is quarantined"
description: "Committed state or history failed verification. The region stays unavailable; never re-provision it. Audit with evo-world-clone audit."
- alert: EvoRegionLeaseMargin
expr: evo_region_ready == 1 and evo_region_lease_remaining_milliseconds < 10000
for: 1m
labels: {severity: warning}
annotations:
summary: "Region {{ $labels.environment }} lease is not being renewed"
description: "Healthy renewals keep 15-20 s of lease. Under 10 s for a minute means renewals are failing; the host will fence itself."
- alert: EvoRegionFenced
expr: increase(evo_region_fences_total[10m]) > 0
labels: {severity: warning}
annotations:
summary: "Region {{ $labels.environment }} host lost authority"
description: "A host was fenced within the last 10 minutes. Expected once after a forced deletion's predecessor resumes; repeated fences point to database or clock trouble."
- name: evo-region-progress
rules:
- alert: EvoRegionStoreFailures
expr: increase(evo_region_store_failures_total[5m]) > 3
labels: {severity: warning}
annotations:
summary: "Region {{ $labels.environment }} storage or database failures"
description: "Publication or database calls are failing; progress pauses and nothing is acknowledged until they succeed."
- alert: EvoRegionUnknownOutcomes
expr: increase(evo_region_unknown_outcomes_total[15m]) > 0
labels: {severity: warning}
annotations:
summary: "Region {{ $labels.environment }} had commits with unknown outcomes"
description: "Commits timed out or lost their connection and were resolved by exact bytes. Occasional is safe; frequent indicates database latency near the 5 s commit deadline."
- alert: EvoRegionCommitLag
expr: evo_region_commit_lag_ticks > 256
for: 2m
labels: {severity: warning}
annotations:
summary: "Region {{ $labels.environment }} is not committing what it simulates"
description: "More than four full batches are simulated but not durable; acknowledgements are waiting."
- alert: EvoRegionTickDebt
expr: evo_region_tick_debt > 8 and evo_region_queued_inputs > 0
for: 2m
labels: {severity: warning}
annotations:
summary: "Region {{ $labels.environment }} is behind real time with inputs waiting"
description: "Due ticks are not being simulated although inputs are queued (catch-up is 8 ticks per step)."
- alert: EvoRegionCheckpointStale
expr: evo_region_checkpoint_age_milliseconds > 300000 and (evo_region_queued_inputs > 0 or evo_region_commit_lag_ticks > 0)
for: 1m
labels: {severity: warning}
annotations:
summary: "Region {{ $labels.environment }} has not committed for 5 minutes with work pending"
description: "An idle region does not commit; this fires only while inputs or simulated ticks wait for durability."
- alert: EvoRegionQueueSaturated
expr: evo_region_queued_inputs >= 0.9 * evo_region_queue_capacity_inputs
for: 2m
labels: {severity: warning}
annotations:
summary: "Region {{ $labels.environment }} input backlog near its bound"
description: "New inputs are being backpressured or rejected."
- alert: EvoRegionInputsDropped
expr: increase(evo_region_dropped_total{reason!="refused"}[10m]) > 0
labels: {severity: warning}
annotations:
summary: "Region {{ $labels.environment }} accepted inputs that will never be acknowledged"
description: "Inputs were drained, fenced or quarantined ({{ $labels.reason }}); clients must resubmit to the current owner."
- name: evo-region-storage
rules:
- alert: EvoRegionObjectsLow
expr: evo_region_objects_available_bytes < 2 * evo_region_objects_reserve_bytes
for: 5m
labels: {severity: warning}
annotations:
summary: "Region {{ $labels.environment }} object volume is below twice its reserve"
description: "local-path volumes do not enforce their size; the node's disk is running out."
- alert: EvoRegionObjectsFull
expr: evo_region_objects_available_bytes < evo_region_objects_reserve_bytes
for: 1m
labels: {severity: critical}
annotations:
summary: "Region {{ $labels.environment }} object volume is below its reserve"
description: "New checkpoints are refused; the region cannot commit or acknowledge until space is freed."
- name: aevolution-monitoring
rules:
- alert: Watchdog
expr: vector(1)
labels: {severity: none}
annotations:
summary: "Always firing: proves the alert pipeline end to end"
description: "Route to a heartbeat receiver once one exists; its absence then means monitoring is down."
- alert: AlertmanagerDown
expr: up{job="alertmanager"} == 0
for: 5m
labels: {severity: critical}
annotations:
summary: "Alertmanager is not scrapeable"
description: "Alerts are evaluated but cannot be routed."