Stop a Tableau Extract Refresh without stopping Server

Oh no! I kicked off this Tableau extract and I need to take it back!! How do I stop it?

As long as I’ve been using Tableau, the ‘run-away’ extract has always eluded me. Sure, there’s netstat and procexp that can sort of get you there in a pretty raw way. If it’s late at night and there’s just one job running, that’s pretty easy to find the port and process ID (netstat -ano in PowerShell 3+).

bye-bye-ts-2
Oops. We didn’t want to run that one.

The challenge is when you have multiple backgrounders (on multiple machines) and they’re all active; you then start asking: which one is it? Hopefully, you guess correctly and, voila, it’s gone.

Until now.

bye-bye-ts-1
The process id for the currently running extract

Here’s what we’ll do:

1.) grab the log from this directory: C:\ProgramData\Tableau\Tableau Server\data\tabsvc\vizqlserver\Logs and follow (or parse) this log: backgrounder*.txt

NOTE: If you have a Log Analytics strategy, you can easily follow this log file and leverage some of the vendor’s command line tools to make this process entirely automatic.

2.) Wrap your code in a PowerShell function which allows you to enter your TWB/TDS name.

Stop-TsExtract -ExtractName '<some name>'
bye-bye-ts-3
The result of stopping the process.

3.) The file in #1 is JSON so it’s super easy to parse and dig out the ‘PID’ which, believe it or not, corresponds to the process on your Tableau Server box. We’re going to look for this key/value pair:

k="ds-parser-connect-extract" AND v.caption=<name of your data source/workbook>

NOTE: You could also look for the session via “k=lock-session” (which you then have to correlate from the other backgrounder log file) but this next value gives you the ability to grab (and enter) the data source/workbook name.

4.) Now, if you’ve set up PowerShell remoting or SSH, you remote into your Tableau Server (via Invoke-Command) and enter (where the variable procID is the backgrounder process):

gps -Id $procID | kill -Verbose

 

BONUS

If you remember, I posted about refreshing your Tableau extracts via Slack here. Well, the next step (if you want to open this up outside the admin group) is the let users drop the name of a data source/workbook in a Slack channel. Those background jobs they enter will be stopped if they’re accidentally started.

Oh, and, I should mention: test, test, test.  This isn’t supported by Tableau, but people believe it should be:) Make your vote count.

Tableau Server Performance fun with Slack and Logentries

The Beginning

It started as a joke. Then it became a pretty good idea.

You see, often times Tableau Server admins get a lot of flak. We *constantly* get comments stating:

  • Tableau is slow
  • This workbook is slow
  • Tableau can’t do this because someone I know said as much
  • I like <insert silly dashboard tool here> more because it does <something I can’t remember right now but I’m just trying to sound smart>
  • You’re just an analyst so you don’t know how to optimize data
  • and more!

Let’s be honest, the above comments can, 99% of the time, be tied to someone who designed something incorrectly but wasn’t aware of the implications of said design. Until now 🙂

And in a natural way that a Slack conversation can allow, the comment dropped: ‘It’s like a dumpster fire’

Inspiration!

It goes like this:

  • Slow TWBs trigger alerts all the time on our platform (Server admins should know these bottlenecks already)
  • We pull log data (yes, you can also pull from Postgres but logs are so rich) for those queries via Logentries
  • We parse the log data and convert to the unruly string data into something usable (thank you PowerShell and, specifically, ConvertFrom-String)
  • At an interval of our choosing, we drop the results in Slack (only our team) with a mixture of funny GIFs (because levity is a good thing)
  • We analyze and reach out to the workbook owners for learning opportunities & improvement

Details

ts-slow-content-2
Monitoring the 90th percentile of workbook load time

 

This is the trigger and the cause of much confusion around what Tableau can actually do. You see, if the performance concerns aren’t addressed, every Server admin is going to get the ‘Tableau is slow’ argument. At that point, you’re left defending the platform you set up. But, the question and concerns should all be about what is *causing* Tableau to appear slow.

We address the performance concern with a solid Log Analytics strategy. The above image is one of many examples of alerts we’ll get. This time, we’re going to leverage the Logenties CLI to automatically pull this info out. Yes, automatically.

Here’s what we’ll use:

lecli query -n vizql -q slowwb -f $start -t $end | Out-File -FilePath $workpath\slowwb.txt -Force

The start and end variables are timestamps; we usually do a rolling 24 hours.

ts-slow-content-0
The output of the Logentries query. Not optimal for parsing/sorting, etc.

If you haven’t explored PowerShell’s ConvertFrom-String cmdlet, you’re missing out. It’s pretty remarkable what it can do with a template file and some string data. And it’s based off of some solid (and profound) research.

ts-slow-content-1
Example template file for ConvertFrom-String. We’re implicitly defining some structure and letting the cmdlet work its magic

After you have (1) pulled the log data and (2) set up your template file, run this:

ConvertFrom-String -InputObject $((gc $workpath\slowwbclean.txt)|Out-String) -TemplateFile $workpath\template_slowwb.txt | select Name,Value

Once you do that, you get a beautiful PowerShell object for which the possibilities are endless (well, as much as you want them to be).

So that string data from above is now easily manageable and prepared for a Slack channel drop.

ts-slow-content-5
A real PowerShell object

Enter Levity

Here’s what the daily GIF might look like in our Slack channel. No, no one sees this but us and the ‘movement’ of it really forces us to pay attention. We’re (as Server admins) responsible for maintaining and teaching others how to best use the platform. If we ignore it, then the tool (Tableau) gets misused and decision makers may start to see if the grass is greener.

Again, and I can’t stress this enough, levity is a good thing. I can’t tell you how many times I’ve seen C-level people view workbooks that take 3+ minutes to load. Three minutes of awkward silence. That. Scares. Me.

So we monitor and when it gets bad and we’re being questioned about the slowness, we need to laugh. That’s healthy and I can’t be convinced otherwise. If you’re intentions are wrong, you’re not being human.

This:

slowWB
It moves.

Becomes this:

 

ts-slow-content-4
Daily performance metrics

That’s it! Pretty easy, right? So what else can we do with this?

Well, here are a few ideas, based off of the magic of GIFs.

And remember, be responsible and kind. There are too many rude people on this planet. One more isn’t a good thing.

(1) Slow extracts become:

wrap_it_up

(2) Slow login times on Apache become:

tumblr_lil0h0CeAx1qavy44o1_500

(3) Large Server downloads become:

file.gif

Tableau Conference 2016: Server Admins

Make sure you come to the Tableau Server Admin User Group meeting at the conference. I’ll be speaking!

http://tc16.tableau.com/learn/sessions/3640

Here is the abstract:

While there are numerous and exceptional benefits in administering Tableau Server via the GUI, the hidden gem is its capability for automation and integration. In simple terms, automating as much of the administration and monitoring makes for a very happy Tableau user base. In this session, you’ll learn how having at least some automation can make your environment faster and leaner.

We’ll automate everything from user provisioning (and removal), auditing views, securing content, and ‘Garbage Collection’ (or just removing old content). Want integration too?! We’ll show you how to reach pretty much anything via the REST API and trigger extracts via tools like Slack.

Oh, one more thing. We’ll show a new platform called Tableau Working Wax: or the ability to automatically generate reports and deliver them to the people.

In the end, you’ll be on your way to a fully automated and healthy Tableau Server infrastructure.

See you at #data16

-Mike

 

Tableau Server: Get Bytes

Continuing the Log Analytics theme for Tableau Server, specifically the ‘Monitor’ pillar of A.I.M., it’s time to show another quick tip for analyzing your Apache logs. Understanding how much data is moving through the wire is another technique one could use to decide whether dashboards, CSVs or general crosstab download needs to be optimized. I shouldn’t need to mention it’s also a great way to monitor the security and integrity of your data.

apache-bytes-0
The ‘Alert’portion. If something is outside the norm, we’re alerted. 

We’ll use the module I loaded here to do the heavy lifting. Once you export a csv, load into Tableau for some analysis (or perhaps mash it up with the geo dashboard I demoed a while back).

You’ll want to pass this query for the ‘leFilter’ section: ‘/HTTP\/1\.1″ “-” \d{3}\s(?P<size>\d*)/ AND size>1000000’

The idea is that, over time, you’ll be able to see what is being requested with the most amount of bytes and, potentially, a better way to optimize your content. Further, it’s a fantastic way to be alerted (‘Alert’ portion of A.I.M.) on large (consistent) GET requests from questionable sources.

In the end, this all fits together with time-to-serve data and other performance and security related inputs.