KQL: find the accounts hit hardest by latency in Application Insights
KQL queries on the requests table and performanceBucket to spot slow requests per account.
KQL queries for Application Insights (the requests table) to find the accounts or users suffering from latency. The performanceBucket column classifies each request by duration (<250ms, 250ms-500ms, 500ms-1sec, 1sec-3sec, 3sec-7sec, ...). Open the Application Insights resource, then Logs.
Slow requests over time, by account
requests
| where performanceBucket in ("3sec-7sec","1sec-3sec")
| summarize count() by bin(timestamp,5m),user_AccountId
| render timechart
Slow request breakdown for one account
requests
| where user_AccountId == "account01"
| where performanceBucket !in ("<250ms","250ms-500ms","500ms-1sec")
| summarize count() by bin(timestamp,5m),performanceBucket
| render timechart
Top 5 accounts with the most slow requests
requests
| where performanceBucket !in ("<250ms", "250ms-500ms", "500ms-1sec")
| summarize BadPerformanceCount = count() by user_AccountId
| top 5 by BadPerformanceCount
| project user_AccountId;
Slow requests over time for those 5 accounts
let Top5Users = requests
| where isnotempty(user_AccountId)
| where performanceBucket !in ("<250ms", "250ms-500ms", "500ms-1sec")
| summarize BadPerformanceCount = count() by user_AccountId
| top 5 by BadPerformanceCount
| project user_AccountId;
requests
| where isnotempty(user_AccountId)
| where user_AccountId in (Top5Users)
| where performanceBucket !in ("<250ms", "250ms-500ms", "500ms-1sec")
| summarize Count = count() by bin(timestamp, 10m), user_AccountId, performanceBucket
| render timechart
Isolate a time window
let StartTime = datetime(2025-01-20T08:00:00Z);
let EndTime = datetime(2025-01-20T10:00:59Z);
requests
| where performanceBucket == "1sec-3sec"
| where user_AccountId == "account01"
| where timestamp >= StartTime and timestamp <= EndTime
To analyse the requests themselves, then use Transaction search in the resource menu.