Root cause analysis

How QUARK traces incidents back to their origin.

When an incident spans multiple hosts and symptoms, the hard question is always the same: what actually started this? QUARK's root cause analysis (RCA) answers it by walking the evidence backwards.

How RCA works

  1. Build the incident scope. Starting from the triggering symptom, QUARK collects every anomaly and event in the surrounding window that involves related infrastructure.
  2. Order by causality, not time. Using the dependency graph (network topology, host groups, shared storage, upstream services), QUARK determines which symptoms could have caused which others.
  3. Find the origin. The earliest anomaly that can explain the downstream symptoms is proposed as the root cause, with a confidence level.
  4. Explain the chain. The result is presented as a causal chain you can audit link by link.

Example

Finding: Application response times degraded on 6 web servers at 14:03.

Causal chain:

  1. 13:58 — Disk latency rises on storage node san-02 (firmware task started)
  2. 14:01 — Datastore latency alarms on the VM cluster backed by san-02
  3. 14:03 — Response-time degradation on the 6 web VMs hosted on that cluster

Root cause (high confidence): scheduled firmware maintenance on san-02 was not excluded from production hours.

Confidence levels

LevelMeaning
HighThe full causal chain is supported by direct evidence
MediumThe chain is plausible but has an inferred link
LowMultiple competing explanations — QUARK lists all candidates

QUARK never presents a guess as a certainty. When confidence is low, it says so and shows every candidate cause with its supporting evidence.

Learning from your incidents

When you resolve a finding, you can record the confirmed cause. QUARK uses these confirmations to improve future RCA in your environment — recurring patterns get recognized faster and with higher confidence.

On this page