Solutions
Follow a customer-app slowdown to the affected API.
A network operator's customer app feels sluggish when people check their balance. The team moves from the application symptom to the API path behind it.
The situation
A network operator's app is how many customers check their balance, buy a data bundle or top up. At the end of the month, when many people do all three on the same day, the balance screen starts to take several seconds to load.
The customer sees a spinner. The engineer sees an alert on the API that serves the balance screen, where latency has moved well above its usual range.
What the team sees first
Service health shows the balance API slow and the rest of the app's APIs normal, so the problem is specific to that screen and not to the app or the network in general. The alert's start time tells the engineer when the month-end rush began to bite.
How the investigation goes
Slow traces from the window all look alike: the balance API makes a call to a billing service, and the billing service spends its time in a database query. The traces from earlier in the week show the same query finishing quickly.
Logs from the billing service for the same requests show the same query appearing again and again with larger result sets. The dependency map shows nothing else waiting on it, which is a useful thing to know before choosing a fix.
The cause fits the calendar: more accounts have more records to read at month-end, and the query was written for lighter days.
What gets shared
The service, the time range and two traces go into one incident record. The billing team can reproduce the slow query from the evidence, and the app team can decide whether to show a cached balance in the meantime.
What this shows
A customer screen is a chain of APIs, and customers feel the slowest link. Starting from the screen and following its requests is how a vague complaint about the app becomes a named query in a named service.
These stories are illustrative examples. They describe invented scenarios in an industry, not a real customer, a real incident or any company's systems.
Start with your own services.
A trial workspace is free for 14 days. Send your telemetry and follow your own requests.