ops-tracker: Architecture and Self-Reference

Component: ops-tracker Tags: ops-tracker,self-reference,architecture,events,flow,issues Author: — Updated: 5/20/2026, 9:28:34 PM

ops-tracker is the institutional memory and observability hub for the Prysm platform. Port: 3563. Section: section7-system-ops.

COMPONENTS:

Issues: ops_issues table. Auto-created by error-log-poller, event handlers, and manual filing.
Flow Tracer: fingerprint-builder + flow-analyzer. Compares live traces against known route profiles. Research pipeline section shows web.fetch.* chain completeness.
Signal Catalog: ops_signal_catalog. Log patterns that trigger escalation.
Knowledge Base: ops_knowledge_articles. Per-component known failure modes and runbooks.
Diagnostic Inventory: ops_diagnostic_inventory. Runnable checks via docker exec or API.

EVENT HANDLERS (/events/*):

/events/diagnostic-alert: from log-collector signal matches
/events/request-stalled: from request-queue-manager stall detector
/events/request-failed: from request pipeline failures
/events/web-fetch-failed: from web.fetch.*.failed events; looks up parent_event_id to get actual URL
/events/ingest-failed: from docs.ingest.failed
/events/research-pipeline-failed: from research.topic.started.failed
/events/zil-extraction-failed: from ZIL extraction failures
/events/acknowledge: no-op for informational events (research.feed.new_post, research.feeds.discovered, research.discovery.complete)

COMMON FAILURES:

1. DB connection at startup: If MariaDB is not ready when container starts, it will crash. Ensure mariadb is healthy before ops-tracker starts.

2. LLM synthesis unavailable: agent-shared callLLMFromTemplate not mounted. Check docker-compose volume mounts.