How we grouped 209 Magento log occurrences into 81 incidents

On one of the Magento hosts we operate, a single PHP runtime error was recorded 104 times. The MNTNET Observability system recorded those occurrences under one incident, with a suspected cause and a recommended next step for the operator. That was the problem we set out to solve. When an error keeps repeating, the operator needs to know whether it is a new failure, how often it has returned, and what to check first.

We built Observability to support Magento stores we operate. This case study uses retained records from two Magento hosts. It shows how the system grouped Magento errors, prepared an AI assessment, and escalated selected incident groups through Slack. The figures below describe the records we verified across both hosts.

Collect the signals that lead to a decision

The storefront agent reads new entries from Magento’s system and exception logs. It carries its reading position forward, then sends matching errors to Observability. Magento’s debug log can provide more detail during a targeted investigation. The deployed collector we audited covers the system and exception logs, which are also the sources for the results in this case. Adobe Commerce logging documentation

Observability normalizes the message and computes a signature so repeats can update the existing incident. It retains the first and last times an error was seen and increments the occurrence count. The incident then reflects recurrence as well as the original error. That makes it possible to distinguish a single burst from activity that keeps reappearing over several days. The system also maintains trend counts across shorter and longer periods to give the operator context during triage. OpenTelemetry guidance on logs

For the operations team, this changes the first question during an alert. Start with the source log, the first and last occurrence, and the count. Then check whether the affected operation still fails. A group can be old and noisy while the current storefront is healthy. A newly recurring group deserves a fresh look, even if its signature is familiar.

Give the operator a diagnosis to check

For a critical incident, an AI worker takes the message and recurrence data and stores a suspected root cause, a recommended action, and a model confidence score. The recommendation gives the operator somewhere to start. A database failure, for example, may need a look at the affected operation and recent changes before anyone touches production. The model’s score describes its own assessment and the operator checks the evidence before applying a change.

A notification worker can send the incident and its trend context to Slack after an assessment is ready. The dashboard and notification both use the incident as the unit of work. In the 104-occurrence example, the retained record shows a single runtime error group on one host. The operator could investigate the recurring signature in one place and still see that it had happened repeatedly. Slack documentation on webhook messaging

What the retained records show

Magento source Recorded occurrences Incident groups Groups with AI assessment
Exception log 136 12 12
System log 73 69 69
Total 209 81 81

Seven of those incident groups had a Slack delivery recorded. We counted groups with a delivery so repeated notifications for the same incident would not inflate the figure. The retained log entries span February through October 24, 2026. The two hosts contributed 44 and 165 occurrences. These are operational records and can include deployment checks. We used the stored source log, occurrence totals, assessment records, and delivery status for the review. These figures measure recorded activity, grouping, and assessment coverage. Repair time and avoided outages were outside that measurement.

Accelerate Your Engineering Velocity

Looking to scale your infrastructure, implement enterprise AI automation, or secure critical cloud dependencies? Partner with MyNew Technologies LLC to engineer robust technical systems built for enterprise efficiency.

Request Technical Consultation

Where recovery automation fits

Magento operators receive a suggested action and decide how to address it. Observability can close a quiet incident in its queue based on the recorded trend. That status represents the state of the incident record. Recovery of the application needs a service or storefront check that confirms normal behavior after the fix.

We also tested recovery for a separate infrastructure service within MNTNET. A local rule initiates a narrowly defined recovery action after repeated health failures. It then checks whether the service became active and limits repeated attempts. The AI assessment remains advisory. The test used simulated service failures. That is the approach we would bring to Magento recovery work, beginning with a failure the team can reproduce, a specific allowed action, and a health check that can verify the result.

The dashboard is available to authorized operators through MNTNET. The published evidence above comes from the two hosts’ retained aggregate records.

Related

How to evaluate accessibility before buying business software

Add practical accessibility checks to software procurement so the team can evaluate the workflows employees and customers will actually use.

How to prepare a B2B storefront for a punchout catalog

Explore the buyer, supplier, and order workflows to test before connecting a B2B storefront to a purchasing system.

How to test cloud portability before it matters

Run an exit rehearsal to see which parts of a cloud application move easily and where your team needs a documented migration plan.

How to decide where to run AI inference

Compare hosted and self-managed AI inference using representative requests, accepted outputs, peak demand, and ownership costs.

How to design a safer file upload workflow

Design uploads with clear limits, private storage, validation, and a review process that keeps incoming files away from production users until they are ready.