App performance monitoring

Knowing is better than guessing

Monitoring a Qlik Sense environment - or any system - comes down to knowing what’s going on.
If you don’t know what’s going on, you’re guessing.

And if you’re guessing, you’re not in control.

If you’re not in control, well - that’s a bad place to be..

Qlik’s performance analysis apps

Qlik has a couple of apps that can be used to analyse performance data from Qlik Sense Enterprise on Windows.

First, the Operations Monitor app ships with every instance of Qlik Sense Enterprise on Windows.
This app is the go-to tool for retrospective analysis of what’s going on in a Sense environment.

The app puts all the great features of Qlik Sense to good use on top of the Sense logs it has access to.
What apps were reloaded when, which reload tasks failed most, which users are most active, etc.

Then there’s the Telemetry Dashboard app.

It uses the same data as Butler SOS, but as it’s a Sense app it must be reloaded in order to get updated data.
Butler SOS on the other hand uses real-time delivery of the events from Sense’s Log4Net logging framework.

If you want to do retrospective analysis of detailed performance data, the Telemtry Dashboard app is an excellent choice.

This page provides additional info about the telemetry logging available in Qlik Sense.

Lots of use cases

Categorisation of events can be done in different ways depending on what you want to achieve.
Here are some examples of what can be done using Butler SOS:

• You want high level insights into how long various app actions take for all apps.
  • Turn on engine performance monitoring in general, but no need for app details
  • Enable tracking of rejected performance log events. This will capture high-level metrics (such as time used and number of events) for all methods/commands used in the engine.
  • Create a Grafana dashboard that shows the average time taken for each method/command.
  • 🎉 You get immediate, visual insight into how fast (or slow) apps open.
• Users complain about an app being slow to open.
  • Turn on performance logging for app open events from that particular app. Store data in InfluxDB.
  • Create a Grafana dashboard that shows a histogram of app open times for that app.
  • 🎉 You get hard numbers on how long the app usually takes to often, and what the outliers are.
• A certain chart is slow to update.
  • Turn on performance logging for that specific app and chart. Store data in InfluxDB.
  • Create a Grafana dashboard that shows a histogram of chart update times for that chart. Or add a new chart to an existing dashboard.
  • Define alerts in Grafana so you get notified if rendering time for the chart exceeds some threshold.
  • Use Butler SOS’ user events features to keep track of what users access the app when the chart is slow.
  • 🎉 You know in real-time how many and which users were impacted, and get a notification when it happens.
• You suspect that an app has a poorly designed data model.
  • Enable detailed performance monitoring for that specific app
  • Create a Grafana dashboard that
    • tracks the “traverse_time” metric. This metric shows how long it takes to traverse the data model. A poorly designed data model tend to have longer traverse times. Break it down by app object and proxy session.
    • tracks the render time for each chart and table in the app. Slow charts and tables may be caused by a bad data model, but not always. Lots of (complex) data may also be a reason.
  • 🎉 You get real-time insight into which objects are slow for what users, due to a bad data model in that specific app.

Performance log events

A “performance log event” is a log event that originates from the Qlik Sense engine, more specifically from the “QixPerformance.Engine.Engine” subsystem.
They are sent to Butler SOS as UDP messages, just like regular log events (using the same port).

The performance log events have INFO log level in the Sense logs but still provide very detailed information about what the engine is doing.

Accepted vs rejected performance log events

Due to the potential high volume of performance log events sent by Qlik Sense, Butler SOS has a filtering feature that can be used to only store the events that are of interest.

Events that meet the filtering criteria are called “accepted” events, and events that don’t are “rejected” events.

Accepted events are stored in InfluxDB, including detailed performance metrics for the event.
Details on what is stored for accepted events can be found in the reference section.

Rejected events are not stored in InfluxDB, but are still counted by Butler SOS.
This feature can be enabled/disabled in the Butler SOS config file. More info here.

Data available for accepted and rejected events

The full set of data stored for accepted events is described in the reference section.

Suggested workflows for different use cases

For some use high-level cases, such as identifying apps that are slow to load, the data for rejected events may be sufficient.

Other uses cases, such as monitoring the behaviour of specfic app objects (charts, tables etc), require the detailed data stored for accepted events.

The recommended, general principle is:

Start with the data captured via rejected events (counter and process_time), and then add detailed monitoring.
The detailed monitoring involves setting up filters to capture specific events for specific apps and/or app objects and/or engine methods.

Also: make sure to have a reasonably short retention period (10-14 days are usually enough) for the InfluxDB database, as the performance log events can generate a lot of data that can easily affect the performance of the InfluxDB instance.

Grafana dashboard examples