This is part 3 in a series about how healthchecks.io can be used to solve various kinds of monitoring scenarios for Qlik Sense Enterprise on Windows (QSEoW) environments.
Most of the Butler family of DevOps tools for Qlik Sense can now be monitored using healthchecks.io or similar tool.
If one of the supported Butler tools go down, an alarm is raised and one or more alerts are sent via email, Slack, Teams, PagerDuty or other supported channels.
This means that both infrastructure (servers etc), applications/services and data sources can be monitored using the same tool and on the same dashboards.
Lots of buzzwords in that title, but yes – things have now come together and the main members of the Butler family of tools for Qlik Sense now run as Docker containers on Arm64 and even Arm7.
In practice this means it is possible to run Butler SOS, Butler Cache Warming etc as Docker containers on for example a Raspberry Pi 4. If you are worried about that poor little Raspi failing you can add a bunch of them into a Kubernetes cluster, and then run the Butler tools there.
This actually works pretty well, here is a Raspberry Pi based Kubernetes (a.k.a “k8s”) cluster built using the excellent k3s distribution from Rancher, running the Butler tools:
Butler tools running as Docker containers in Raspberry Pi based Kubernetes cluster.
And here is the Raspi cluster itself, with 4 of the 7 nodes installed:
Kubernetes cluster built on Raspberry Pis
Ok, ok – I hear you – is it really a good idea to run close to mission critical software on Raspberry Pi’s? No of course not, but the cloud providers are starting to offer Arm based virtual machines (Amazon does this already with their EC2 A1 instances), with roughly twice the performance per dollar compared to Intel based VMs.
So running Butler on Arm VMs might be a way to save some money, if the cloud provider you use happens to offer Arm VMs.
Or just surrender to your inner geek and build your own Raspi k8s cluster… Lots of fun and a great way to learn about k8s!
This release both adds some nice new features as well as enhancing existing ones and fixing some bugs. Let’s take a look at the highlights.
Track in detail whatapps are loaded into each Sense server.
Regular apps and session apps are handled separately, making app metrics easier to understand and more relevant
Sample dashboards are now built using the brand new, shiny and all together awesome Grafana 7. Did I mention that Grafana 7 is awesome? Awesome.
Ever wondered how long Butler SOS has been running or how much memory it uses? The new uptime messages have you covered.
You are properly impressed with the uptime messages – good. Why not store them to Influxdb, so you can also visualize Butler SOS’ own memory use? It’s just a couple of changes in the config file away.
Better control over what features are enabled. Don’t need Docker health checks? Now you can turn that (and other) feature off.
Ah, you are a serious Sense user and have separate DEV and PROD environments? Good – now Butler SOS supports multiple instances running on a single server.
Who will monitor the monitor? Butler SOS can now send heartbeats to customisable URLs at desired intervals. Perfect if you want to monitor Butler SOS using for example healthchecks.io. Very, very cool actually.
Bugs, bugs and bugs. The known ones have been fixed. Keep reporting new ones!
Update all dependencies to latest versions, to ensure security concerns are adressed.
Curious what it looks like in practice? Seeing is believing:
Usually, yes. Still, things do happen and it certainly would be nice to get an push alert when a server hasn’t checked in according to schedule.
The most common monitoring – for Windows/Linux/…. servers, databases, Qlik Sense etc is based on the tool keeping an eye on some measurement and then alert when the measurement goes beyond some threshold. This is fine, and this is a very important monitoring use case. But in cases where a server just hangs the last measurement received might be fine, and no alerts are sent.
Black box monitoring kind of reverses the roles:
The monitored system has to prove that it’s doing fine. Failing to do so within some predefined schedule will trigger an alarm, with an optional alert being sent.
The previous article showed how this concept can be used to ensure that some Qlik Sense app has reloaded as intended, before a specific time each day. A concrete, common use case would be that yesterday’s data should be processed and loaded into Sense before 7 am next day. Alert if not)
Now, let’s use the same tool and concept to monitor also the Windows servers that Qlik Sense Enterprise on Windows (QSEoW) runs on.
When implementing Qlik Sense solutions for enterprise clients they usually have various requirements regarding alerts for failed reloads, support SLAs etc. Those are all interesting challenges to implement, but the most common request is probably:
“The system should alert me if data doesn’t arrive on time”
It’s a reasonable request. If some source system is delayed and doesn’t delivery data to Sense on time, they as a system or application owner should be notified.
Don’t bother me unless it’s broken
The second thing almost all clients say is:
“Oh, I only want alerts when data is delayed. No messages when data is on time.”
Again, this is very reasonable.
Let’s say the extract app in Sense sends a notification email once data has been loaded from the source system. Great – we now know that data has arrived, and when it did. Except that we will drown in such notifications from all our dozens (or hundreds..) of extract apps.
It’s kind of hard to implement this in a good way though, at least I have never found a really good, generic solution for this request. Sure, you can have a Sense app who’s only job is to execute every 15 minutes and check whether data has arrived, and alert if data is delayed. That app will however put load on your Sense environment and use one of the reload slots. Even if the app reloads quickly it’s still pretty bad system design, IMHO.
I have a suggestion for a better option:
Set up a monitoring tool that views the Sense app (or rather its load script) as a black box with unknown function. The only thing the monitoring tool cares about is whether that black box has checked in within some configured interval.
I stumbled upon this while looking at ways to monitor servers in general, but quickly realised it could be used also with Sense. By the way: while I have only used the described concept on Qlik Sense, it should work equally well with QlikView.
Don’t care for reading? This video outlines the concept, otherwise keep reading below.
During the past weeks I have developed what has turned into a fairly complex Qlik Sense solution, where reloading the main application takes on the order of 10 minutes.
I know – Qlik apps should be kept small, with intermediate data sets stored as QVDs. But sometimes that’s just not practical.
Qlik apps works the same as most apps – if you made a typo somewhere the reload (or compilation for regular computer applications) will fail. Sometimes after a few seconds, more often right towards the end, just before those 10 minutes have passed… Fix the typo and start another 10-minute reload. Sigh.
The other day I realised that I had been using an old trick without even thinking about it, to cut down the reload times from 10 minutes to a few seconds. Talking to some colleagues, it turned out they didn’t use this trick – so here it is: The best time saver ever for Qlik developers.
The idea is very simple: You want to run the entire load script to make sure it works, but without processing 100s of megabytes of data or more.
In Qlik scripting, this is achieved by using the First prefix, added before a Load statement. Two things needed:
Add a variable at the beginning of the script, and make it easy to change it to a few different values (for loading 1000 rows of source datasets, or something larger).
Add the variable before all load statements where large datasets are loaded into the app.
It can look something like this:
Sense’s load script editor does not quite like using variables in this way, and complains about the syntax being incorrect. Don’t worry about that, the script loads fine.
To do a quick load (to ensure the script is solid from a syntax perspective), just enable one of the commented out lines at the top. This will limit the numer of lines loaded in the following load(s). The load time will be nice and short.
Once the script works, comment out that first variable again, setting it to an empty string. All data will be loaded from sources.
Easy trick – I know. But it has literally saved me days during past months!
Hack the Crisis was a government sponsored hackathon with 5000+ people taking part – pretty impressive.
Friday night our hackathon team kicked off our project. The goal was to develop the core concepts of a service that makes it easier to help those who need help in these challenging times.
The verdict is still to come regarding official hackathon results, but the general feeling is pretty good. There is a distinct need out there, we demoed the core concepts of a solution, and there are clear paths onward from here.
And yes – a 48 hour online hackathon is a pretty effective way of achieving social distancing from your family living in the same house…
Qlik Sense is very much my go-to tool for many data related questions. It’s very powerful to be able to throw together a proof of concept or test a hypothesis in half an hour. Seeing is however believing, so those prototypes does need some polishing before they can be seen as production grade.
I have always pushed my peers in the Qlikosphere to become better at using icons for their Qlik Sense apps in general, and also for the sheets within those apps. Having an icon for a sheet within an app makes it way easier for the user to find a particular sheet – humans are visual by nature.
Butler SOS has matured quite a bit during past couple of years, with latest additions being fine-grained monitoring of what users are connected to what servers. Opens up for various interesting use cases, including notification to users before server restarts etc.
A brand new doc site (butler-sos.ptarmiganlabs.com) also goes live today (built with Hugo and Docsy, vastly better than the previous one-pager at GitHub.
Starting today that site is the goto-place for everything Butler SOS related. The site contains both installation and configuration instructions, as well as a (hopefully growing) set of examples on how Butler SOS can be used in various way.
Make sure to check out the blog section of that site too!
Real-time monitoring of user sessions using Butler SOS
Butler SOS 4.0 is out, adding features that make it easier than ever to monitor large Qlik Sense environments. We’ll return to this topic of course, but let’s first take a few steps back.
There are many variants of that quote: “If you can’t measure it, you can’t improve it”, “Measure what matters”, “Measure what is measurable, and make measurable what is not so.” and others. The last one supposedly originates with Galileo Galilei. Smart guy.
The development of Butler SOS continues in that spirit. Qlik Sense provides an awesome platform on top of which all kinds of data analysis, visualisation and presentation solutions can be built. A key word there is platform. Sense does not offer solutions to all software development challenges, nor should it. Instead, use the tools and best practices that millions of developers around the world have refined over the years.
Qlik Sense does on the other hand offer a very comprehensive set of APIs that give developers access to its internals – and this is part of why it’s such a powerful platform. Butler SOS taps into some of these APIs, exposing their data in the form of real-time dashboards, charts and alerts. Suddenly sysadmins know can get both an overview of how all servers are doing, as well as look at detailed server metrics when so needed. All made possible using the Sense APIs, but in general powered by various open source tools.
We’re basically back to Galileo – let’s make sure the important parts of Qlik Sense are measurable. It is then possible to improve the parts that don’t work well.
