Skip to content

Glossary

Common terms used throughout LogSys documentation.


Core Concepts

Term Definition
AIOps AI-driven IT operations: applying ML to automate incident detection and response
Alert A configured rule that fires when events match conditions; can escalate
Application A logical grouping of sources (e.g. payments-api), with environment + hosts
Canonical Event The normalized event schema shared by all pipeline services
Collector A connector that ingests events from an external system (Jira, Prometheus, …)
Deduplication (Dedup) Merging repeated identical events into one, incrementing occurrence_count
DLQ Dead-letter queue — events that failed processing, kept for replay
Drain3 Online algorithm for log-template mining
Event A single log/metric/trace/event record flowing through the system
Incident A tracked operational issue, potentially correlating many events
KPI Key performance indicator aggregated from events (error rate, events/min, …)
Materialized View Precomputed SQL aggregation for fast dashboard reads
Normalizer Pipeline stage that validates canonical events + strips PII
Pipeline The consumer pipeline: parse → normalize → dedup → features → ML → storage
Priority Operational severity ranking P1–P4 assigned per template
Source A named data origin (e.g. prometheus-prod)
Template A mined message pattern with placeholders (Drain3 cluster)
Time Series Bucketed event counts used for charts

Infrastructure

Term Definition
KRaft Kafka's built-in consensus protocol (no ZooKeeper)
Broker A Kafka server node
Topic A named Kafka stream; LogSys uses 6 topics
Partition A topic's parallel unit; limits consumer parallelism
Consumer Group Set of consumers sharing a topic's load
Consumer Lag How far consumers are behind the latest offset
Keyset Pagination Cursor-based pagination, O(limit)
Partition Pruning PostgreSQL skipping non-relevant table partitions
GIN Trigram PostgreSQL index enabling fast ILIKE search
PgBouncer Connection pooler for PostgreSQL

ML Terms

Term Definition
RoBERTa A transformer language model used for sequence classification
Fine-tuning Training a pretrained model on task-specific labeled data
Softmax Probability distribution over classes; confidence source
Template Cache L1/L2/L3 priority cache (dict → Redis → PostgreSQL)
Label Ground-truth priority for a template
Ground Truth Human-verified labels used for retraining
Fallback Deterministic severity→priority rule when ML is unavailable

Auth & Security

Term Definition
JWT JSON Web Token — signed access/refresh tokens
RBAC Role-Based Access Control (viewer → superadmin)
HttpOnly Cookie flag preventing JS access (XSS-safe)
SameSite Cookie attribute; Strict prevents CSRF
PII Personally Identifiable Information (stripped by normalizer)
TLS Transport Layer Security (terminated at load balancer)

Frontend

Term Definition
SPA Single-page application (React)
TanStack Router/Query/Table Routing, server-state, and table libraries
WebSocket Persistent connection for live KPI snapshots
Zustand Lightweight client state store