Problem Management

How recurring issues get promoted from symptom-chasing to a documented known error.

A Problem represents the underlying cause behind one or more incidents — the difference between patching a symptom repeatedly and fixing it for good. Its lifecycle runs new, under investigation, known error, resolved, closed.

Unlike Incidents, Problems have no duplicate-linking or merge feature — if two problems turn out to be the same underlying issue, consolidate them manually by moving the incidents onto one and closing the other.

How problems get created

Problems can be opened manually, but most are detected automatically: three or more incidents against the same CI/category within 7 days, five or more firing alerts against the same CI/alertname within 60 minutes, a scheduled infrastructure review, or a post-change monitor triggered by a high/critical-risk change closing. A vulnerability-scan finding is another path in — either detected automatically or pushed in directly by a security team, with the CVE, severity, and affected CI attached.

Rather than starting from a blank form, Problems offers a set of built-in templates for common patterns — a recurring network outage, database performance degradation, an authentication/SSO failure, a security vulnerability, and others — that pre-fill category, priority, impact, and tags. Admins can also add custom templates, and problems can be bulk-imported from a spreadsheet-style file when migrating a backlog in.

Known Error Database

Marking a problem as a known error transitions it out of investigation and can auto-draft a knowledge-base article from the AI RCA output plus the workaround — landing in draft state until a human publishes it. Any known-error problem's root cause and workaround are then directly searchable through the KEDB.

Resolving

Moving a problem to resolved requires a documented root cause and a confirmed, verified fix — usually a linked Change that has closed successfully. Resolving cascades automatically: every incident still linked to the problem is closed along with it, with an auto-generated note explaining the permanent fix, and the people who reported those incidents are notified. That's the actual payoff of problem management — one resolution clears the whole backlog of related tickets instead of someone closing them one by one.

Closing

Closing a resolved problem is a separate step, and for a major problem it requires a completed major-problem review first — closing a major problem automatically opens a Post-Incident Review if one doesn't already exist, so that review requirement is never a dead end you have to chase down manually.

Running AI RCA on a problem analyzes its linked incidents together and returns a root cause, a pattern description, a confidence score, a permanent-fix suggestion, and affected services — but the automatic detection paths above (recurring incidents, alert patterns) are statistical, not AI-driven.

Opening a problem and linking incidents

From Problems, click + New Problem — optionally starting from one of the built-in templates — and fill in a title, description, and priority. In the Linked Incidents tab, search for the recurring incidents you suspect share a root cause (same CI or symptoms is usually the pattern) and link them — three to five is typical before running an analysis. If a new incident related to the same underlying issue shows up later, it can be spawned directly from the problem instead of created separately and linked after the fact.

Running a cross-incident RCA

In the Problem RCA tab, click Run AI Problem RCA. Unlike a single incident's RCA, this analyzes every linked incident together and looks for the common thread — for example, several unrelated-looking incidents that all trace back to the same misconfigured NTP server across sites.

Marking a known error

Once the root cause is established, click Mark as Known Error and enter a workaround — optionally linking a knowledge-base article. From that point on, the workaround is visible to any agent working a related incident, so the same fix doesn't get rediscovered from scratch each time.

Example

Three unrelated-looking incidents — a login failure in one data centre, a certificate warning in another, a scheduling error in a third — all recur within the same week. A manager opens a problem, links all three, and runs AI Problem RCA. The result: all three trace back to the same misconfigured NTP server drifting time across three data centres. The problem is marked a known error with the workaround "restart the NTP service on affected servers," and a change is raised for the permanent fix. Once that change closes successfully, the problem resolves automatically — closing all three linked incidents with a note pointing back to the fix, and notifying everyone who originally reported them.