Find out what is loading the service and the database

When the data service slows down or the database is working hard, you want to know: Is there more traffic than usual? At what time of day? Which queries cost the most, and what can be done about them? The Auslastung (Load) area answers these questions from the service's load log, without any tools on the server.

Solution overview

The service groups requests that differ only in their values into one query shape: the same data source, the same object, the same compared fields, with every value shown as ?. For each shape, it counts requests, time in the service, time in the database, rows, and errors. A shape contains neither persons nor values.

The Auslastung area (under Überwachen, Monitoring) has five tabs:

Tab	Contents
Verlauf (History)	Period (24 hours, 7 days, 30 days, 1 year), four key figures, and a column chart of the time spent in the service, split into database and the rest of the service
Wochenmuster (Weekly pattern)	Requests by weekday and hour, as a color scale from light (none) to blue (most)
Abfrageformen (Query shapes)	Every shape from the last seven days with requests, total time, average, 95th percentile, maximum, database time, rows, errors, and refusals; expand a shape to see its slow requests and its hints
Hinweise (Hints)	Shapes that match a known pattern, with the obvious fix and the time the hint was recorded; the service stores them once an hour, and older ones stay in the list
Personen (People)	Load by day, shape, and person; shown only with the `load-persons` right

The key figures on the Verlauf tab:

Key figure	Meaning
Anfragen (Requests)	Requests in the period; below it, how many ended with a service error (5xx)
Zeit im Dienst (Time in service)	Average time per request, in milliseconds
Anteil Datenbank (Database share)	Share of the time spent in the database (from opening the connection to the last row)
Hinweise (Hints)	Number of shapes with a known fix in the last seven days

Steps

Review the load for a period

Open the Auslastung area.
Choose the period at the top of the Verlauf tab.

Result: The key figures and the chart cover the period. The chart shows one column per hour (24 hours), per day (7 and 30 days), or per week (1 year). A large database share points to expensive queries; if most of the time is spent in the rest of the service, it goes to transfer or to the service itself.

Find expensive queries

Select the Abfrageformen tab. Sort by Gesamtzeit (ms) (Total time) or p95 (ms) by clicking the column header.
Expand a shape. Langsame Anfragen (Slow requests) lists its individual requests above the service's threshold, and Hinweise lists the patterns it matches.
The Hinweise tab lists every shape with a fix, for example reads without a filter that return many rows, a filter on a column without an index, or an automatic refresh at short intervals. A shape that stays expensive appears there once, with the time (UTC) the hint was recorded, so you also see what was expensive last week.

Result: You know the shape, its data source, and its object. The Vorlage (Template) column shows the statement the service generates for it, without values. A database administrator can use the shape's Kennung (ID) to find the same statement in SQL Server (plan cache, Query Store), because the service writes it as a /* wkl:… */ comment into every statement it generates.

Send the load to the vendor

Select Weitere (More) > Ausfuhr für den Hersteller … (Export for the vendor).
Enter the number of days (30 is suggested) and choose the file.

Result: A JSON file with shapes, hours, and hints. It contains the names of data sources, objects, and columns, but no persons, keys, addresses, or values. The export is recorded in the change log.

Notes

Up to a minute behind. The service writes its counts once a minute, so a request shows up only after that. Aktualisieren (Refresh) reads the data again.
Times. The chart and the lists show UTC. The Wochenmuster tab counts by the server's clock, because a working day follows local time; its columns are the hours 00 to 23.
Retention. Shapes and hints cover the last seven days. How long hours, slow requests, and daily figures per person are kept is set by whoever runs the service; a longer period shows only what is still there.
Load per person. This shows who worked how much and when. The service therefore requires the load-persons right, which a rule must name explicitly; admin alone is not enough. Without it, the Personen tab is hidden and slow requests show no person.