Analytics & Healing

ContextView includes an analytics and self-healing subsystem that monitors service health, detects error patterns, and can automatically attempt recovery actions.

Analytics Agent

The AnalyticsAgent analyzes system metrics across three dimensions:

Trend Analysis

Metric	What’s Tracked
Request Rate	Volume changes, traffic spikes, drop-offs
Error Rate	Error frequency trends, new error types
Latency	P50/P95/P99 latency distributions, regression detection

Anomaly Detection

The analytics agent detects anomalies by comparing current metrics against historical baselines:

Sudden error spikes — new error types or rate increases > 2σ
Latency regressions — P95 latency exceeding historical 95th percentile
Traffic anomalies — unusual request patterns (load spikes, drop-offs)

Error Detection

The ErrorDetector classifies errors from trace data:

from contextunity.view.analytics.error_detector import ErrorDetector

detector = ErrorDetector()
errors = detector.detect(traces=recent_traces)

for error in errors:
    print(f"[{error.severity}] {error.service}: {error.pattern}")
    print(f"  Occurrences: {error.count}")
    print(f"  First seen: {error.first_seen}")

Service Healing

The ServiceHealer attempts automated recovery when errors are detected:

from contextunity.view.healing.service_healer import ServiceHealer

healer = ServiceHealer(admin_client)
result = await healer.heal(
    service_endpoint="router:50052",
    error_type="UNAVAILABLE",
)
# result = { "actions": ["restart_attempted"], "success": True }

Healing Actions

Error Type	Recovery Action
`UNAVAILABLE`	Service restart, connection pool refresh
`DEADLINE_EXCEEDED`	Timeout adjustment, load balancing
`RESOURCE_EXHAUSTED`	Cache eviction, memory pressure relief
`INTERNAL`	Log collection, diagnostic trace creation

Code Fixer

The CodeFixer module analyzes recurring error patterns and suggests code-level fixes:

Identifies common anti-patterns in graph configurations
Suggests timeout and retry parameter adjustments
Recommends model fallback chain modifications

gRPC RPCs

All analytics and healing operations are exposed via the AdminService gRPC:

RPC	Permission	Description
`GetSystemAnalytics`	`admin:read`	System-wide analytics
`GetErrorAnalytics`	`admin:read`	Error pattern analysis
`DetectSystemErrors`	`admin:write`	Run error detection scan
`TriggerSelfHealing`	`admin:write`	Initiate automated healing
`GetHealingStatus`	`admin:read`	Check healing operation status

Integration with Dashboard

Analytics data powers the Dashboard overview charts:

KPI cards — real-time counts and rates
Time-series charts — Chart.js graphs of request/error trends
Alert badges — anomaly indicators on service health panel