AXS knocks it out of the park with modern data analytics

Learn how Etleap and Looker helped AXS reduce manual ETL work and reporting, allowing them to focus on growth for themselves and their clients.

AXS is a ticketing company for live entertainment

AXS powers the ticket buying experience for over 350 world-wide partners

AXS is a leading ticketing, data, and marketing solutions provider in the US, UK and Europe. The company and its solutions empower more than 200 clients— teams, arenas, theaters, clubs and colleges— to turn data into action, maximize the value of all their events and create joy for fans. It is an enterprise event technology platform that services venues, promoters and sports teams; providing fans the opportunity to purchase tickets directly from their favorite venues via a user-friendly ticketing interface. While customers know them as a destination for tickets, clients recognize AXS for their data services, including transforming, reporting, analyzing, and more.

The data services offered by AXS have always been incredibly helpful for clients. But to make them so valuable, a significant amount of ETL work and reporting was required, which resulted in some challenges for the team. To learn more about those challenges, and how they eventually found a solution in both Etleap and Looker, we spoke with Ben Fischer, the Sr. Director of Business Intelligence and Strategy.

THE CHALLENGES

Ben oversees the Business Intelligence and Strategy team, which manages everything from integrations and building data models to powering the data warehouse and products across AXS. The team’s main objectives are to power AXS’s internal data services, while also delivering data services for clients.

Before Etleap and Looker, the Data Engineering team was spending more and more of their time working on internal and external requests for one-off ingestions and custom data sources. Each data source would take anywhere from half a day to weeks (or even months) to implement, which meant the team was spending most of their time on ETL work, and not enough time on making the data useful.

“With Etleap, we’re able to do the ETL end-to-end and get it directly into the hands of whoever’s trying to use it right away.”

– Ben Fischer, Sr. Director of Business Intelligence

Ben told us, “The whole team was just getting sucked into ETL work constantly, which was not the best use of their time. We wanted to be working on modeling and on the products.”

SEARCHING FOR A BETTER DATA SOLUTION

In order to find the right solutions to fit their needs, the AXS team compared several modern ETL solutions.

When comparing the options, they found that most of the tools were good solutions for getting ETL out of the engineers’ hands, allowing less technical people to consistently bring in the data, while also offering support and monitoring. However, Etleap stood out in two main areas: transformation and transparency.

For AXS, transformation was important because they wanted the ability to not only bring in a new data source, but automatically transform it into something useful for analysts. “Stitch and Fivetran are really focused on the “extract” and “load,” so they’ll bring data in from an outside source and put it into your data warehouse, but they don’t offer much in the way of transformation. You still have to transform it afterwards into something that’s usable, which means you’re still relying on engineering to access the data. With Etleap, we’re able to do the ETL end-to-end and get it directly into the hands of whoever’s trying to use it right away.” said Ben.

Etleap’s data wrangler makes parsing and structuring data take minutes instead of months.

Beyond the transformation aspect, the AXS team was also impressed by Etleap’s level of transparency around reliability. “A lot of the competitors emphasize this idea of 100% reliability. They would say that they would never miss any data, everything would come through perfectly, and you would never have any issues. But we knew that wasn’t the case. No tool is 100% perfect, and when talking with Etleap, they were much more open about what we could expect. They acknowledged that 100% reliability is the objective, but that it’s challenging to achieve in practice, so it’s something they’re continuously working to achieve. Some of the competitors wouldn’t even acknowledge that reliability could possibly be an issue, which makes you feel like they may not support you if anything goes wrong.” said Ben.

To top things off, Etleap was also very straightforward to use. It required very little training and offered reliable support, which meant AXS could get up and running immediately. When the team first started evaluating ETL solutions, they encountered complexity with managing the tools and building integrations. “But with Etleap, it’s pretty straightforward. There’s always somebody available if you need to reach out. That meant we could start using Etleap in just a matter of days, rather than undergoing weeks of training,” said Ben.

Once the team found an ETL solution, it was time to help out the analysts. They looked at a variety of products for business intelligence, and even tried a few different solutions, but Looker stood out in part because it could get report building out of the hands of the analysts. Ben told us, “With Etleap, it was about getting ETL out of engineering. With Looker, it was about getting report building out of analytics, so analysts can spend their time actually forming opinions, defining strategy, doing analyses, and digging into the data, rather than just building reports day in and day out.”

Consistency and confidence are critical to democratizing data, and Looker’s data modeling layer allows people across AXS to pull their own insights and reports very quickly without having to worry about whether the numbers match. This means the Business Intelligence & Strategy team can now stay focused on building models and driving insights, instead of just building reports.

WITH ETLEAP AND LOOKER, THE ENTIRE AXS TEAM IS ABLE TO FOCUS ON HIGH-VALUE TASKS THAT DRIVE THEIR BUSINESS FORWARD.

Since implementing both Etleap and Looker, the impact has positively affected the entire AXS team, as well as their clients.

First, the Data Engineering and Business Intelligence & Strategy teams are spending far less time on manual ETL and reporting work, and much more time on high-value tasks that contribute to internal growth, as well as client successes.

“These tools make our various teams more impactful across the business. For example, if our engineers were just doing all the ETL work manually, we would not be able to do even half of the work that we’re doing to drive the business forward. And the same applies with Looker. Right now, we’ve got people all over the organization looking at reports every day in Looker and answering their own questions about what’s going on with the business.

“Our lives have become much less about pulling reports or bringing in data, and more about really driving value for the company beyond the mundane day to day.” Beyond making life easier and their work more impactful, reporting has also become much faster. Previously, you may have had to wait weeks, or potentially months, to get access to a new data source so you could do your analyses. Now, you can solve that yourself in a couple of hours, without having to wait for other people.

Looker gives companies a single source of truth for all their data.

“With Looker, it was about getting report building out of analytics, so analysts can spend their time actually forming opinions, defining strategy, doing analyses, and digging into the data, rather than just building reports day in and day out.”

– Ben Fischer, Sr. Director of Business Intelligence

“It also makes iteration much faster. We can define something, put it into production, and report off of it. If three days later we realize we forgot something, it’s a two-minute fix rather than going back to engineering, having someone spend half a day on it.” said Ben.

Finally, having access to Etleap allows the team to easily look at data from different angles, making the analysis and insights for clients even more valuable. “Etleap has a function for modeling data, which is useful for reporting, as it allows you to build the aggregations you need to power impactful reports. We can have processes that run everyday and get a quick summary of the data from all different perspectives. Before, it would have taken an engineer a couple of days to build that.” said Ben.

With Etleap and Looker, the AXS team finally has the time and resources to focus on bigger initiatives, including GDPR, internationalization, increasing accessibility across the organization, and providing even more data services to clients. With these tools in their arsenal, the sky is truly the limit.

What is the “length” of a string?

Finding the length of a string in JavaScript is simple, you use the .length property and that’s it, right?

Not so fast. The “length” of a string may not be exactly what you expect. It turns out that the string length property is the number of code units in the string, and not the number of characters (or more specifically graphemes) as we might expect. For example; “😃” has a length of 2, and “👱‍♂️” has a length of 5!

Screenshot from Etleap’s data wrangler where the column width depends on the column contents.

In our application we have a data wrangler that lets you view a sample of your data in a tabular format. Since this table supports infinite scrolling, both rows and columns are rendered on demand as you scroll vertically or horizontally. We can’t render all the rows and columns at once since a table could easily include more than a hundred thousand cells, which would bring the browser to its knees.

“The ‘length’ of a string may not be exactly what you expect.”

Imagine if most rows of a column contains a small amount of data, such as a single word, but a single row contains more data, such as a sentence. If this row is outside of the currently viewed area we don’t want the column to expand as you scroll down, and we definitely don’t want to cram the sentence into the same small space that’s required by the word. This means that we need to find the widest cell in the column before rendering all the cells. It’s fast and straightforward to find the length of the content in each cell, however what if the cell contains emojis or other content where we can’t rely on the length property to give us an accurate value?

Code units vs. code points

Let’s do a quick Unicode recap. Each character in Unicode is identified by a unique code point represented by a number between 0 and 10FFFF.  Unfortunately, 10FFFF is a large number and requires 4 bytes to represent. To prevent having to allocate 4 bytes for each character, Unicode also specifies different encoding standards that can be used to interpret it, including UTF-16 which is the internal string encoding used by JavaScript.

UTF-16 is a variable length encoding, which means that it uses either 2 or 4 bytes for each code point depending on what is required. To differentiate, we say that UTF-16 uses one or two code units to represent one Unicode code point. The most used characters all fit into one code unit, however some of the more exotic characters, such as emojis, require two code units.

“It turns out that code points are not the only caveat regarding string lengths in JavaScript.”

This is where a problem arises. Since the .length property returns the number of code units, and not the number of code points, it does not directly map to what you may expect. As an example, the emoji “☺️” has a length of 2, even though it looks like only one character.

How can we work around this? ES2015 introduced ways of splitting a string into its respective code points by providing a string iterator. Both Array.from and the spread operator […string] uses this internally so both can be used to get the length of a string in code points.

Combining Characters

It turns out that code points are not the only caveat regarding string lengths in JavaScript. Another is combining characters. A combining character is a character that doesn’t stand on its own, but rather modifies the other characters around it. This is supported in Unicode, meaning that characters such as “è” is actually made up of two code points, “e” and  “\u0300”. This is widely used to combine emojis to get a new representation, such as “👱‍♂️” which is a combination of ” 👱” and ” ♂” with a zero width joiner (\uDC71) in between.

Working around this is more complicated. Currently there is no built in way of reliably counting graphemes in JavaScript. A current stage 2 proposal suggests adding Intl.Segmenter which will return the number of graphemes in a string, however there’s no guarantee that it will make it into the spec (there’s a polyfill for the proposal if you’re desperate.)

Environment Specific Differences

Did you know there’s a ninja cat emoji? Neither did we, because it’s a Windows-only emoji! It’s represented by a combination of “🐱” and “👤”. This means that Windows users will see this combination as one character, while other users will see it as two characters. Depending on the users choice of fonts, they could even see something completely different. You could try to prevent this issue by choosing a specific font for your web app, however that won’t be sufficient as the browser will still search through other fonts on your system if a character is not available in your chosen font.

“The various environment specific differences means that there’s generally no way of measuring the rendered width of a string mathematically. “

Checkmate?

The various environment specific differences means that there’s generally no way of measuring the rendered width of a string mathematically. Therefore, the only way to determine the pixel length is to render it and measure. For our use case in the wrangler, this is exactly what we wanted to avoid in the first place. However there are some optimizations that we can make. 

Instead of rendering all the strings in each column, we can split the strings into their corresponding graphemes and render them individually. This allows us to cache the pixel length of each grapheme we encounter. Since there are substantially fewer graphemes than unique strings in a table, this results in a significant reduction in total rendering. This way we can easily determine the correct width of a column, all while keeping the scrolling snappy and your browser happy.

High Pipeline Latency Incident Post-Mortem

Between 15:30 UTC on 8/27 and 14:00 UTC on 8/29 we experienced periods of higher-than-usual pipeline latencies. Between 04:00 and 10:00 UTC on 8/29 most pipelines were completely stopped. At Etleap we want to be transparent about system issues that affect customers, and this post summarizes the timeline of the incident and our team’s response, and what we are doing to prevent a similar incident from happening again.

Number of users with at least one pipeline with higher-than-normal latency.

What happened and what was the impact?

At around 11:30 UTC on 8/27 our ops team was alerted about spikes in two different metrics: CPU of a Zookeeper node and stop-the-world garbage collection (STW GC) time in a Java process responsible for orchestrating certain ETL activities. The two processes were running in different Docker containers on the same host. From this point onwards we saw intermittent spikes in both metrics and periods of downtime of the orchestration process, until the final fix was put in place at 14:00 UTC on 8/29. Additionally, at 15:30 UTC on 8/27 we received the first alert regarding high pipeline latencies. There were intermittent periods of high latency until 10:00 UTC on 8/29.

Incident Response

When our ops team received the first alert they followed our incident response playbook in order to diagnose the problem. It includes checking on potential causes such as spikes in usage, recently deployed changes, and infrastructure component health. The team determined that the issue had to do with the component that sets up source extraction activities, but found no other correlations. Suspecting an external change related to a pipeline source was leading to the increased garbage collection activity, they went on to attempt to narrow down the problem in terms of dimensions such as source, source type, and customer. Etleap uses a Zookeeper cluster for things like interprocess locking and rate limiting, and the theory was that a misbehaving pipeline source was causing the extraction logic to put a significant amount of additional load on the Zookeeper process, while at the same time causing memory pressure within the process itself. However, after an exhaustive search it was determined that the problem could not be attributed to a single source or customer. Also, memory analysis of the Java process with garbage collection issues showed nothing out of the ordinary.

The Culprit

Next, the team looked at the memory situation for the host itself. While each process was running within its defined memory bounds, we found that in aggregate the processes’ memory usage exceeded the amount of physical memory available on the host. The host was configured with a swap space, and while this is often a good practice, it is not so for Zookeeper: by being forced to swap to disk, Zookeeper’s response times went up, leading to queued requests.

Stats show Zookeeper node in an unhealthy state.

In other words, the fact that we had incrementally crossed an overall physical memory limit on this host caused a dramatic degradation of the performance of Zookeeper, which in turn resulted in garbage collection time in a client process. The immediate solution was to increase the physical memory on this host, which had the effect of bringing Zookeeper stats back to normal levels (along with the CPU and STW GC metrics mentioned before).

Zookeeper back in a healthy state after memory increase.

Next steps

We are taking several steps to prevent a similar issue in the future. First, we are configuring Zookeeper not to use swap space. Second, we’re adding monitoring of the key Zookeeper stats, such as latency and outstanding connections. Third, we are adding monitoring of available host physical memory to make sure we know when pressure is getting high. Any of the three configuration and monitoring improvements in isolation would have led us to find the issue sooner, and all three will help prevent issues like this from happening in the first place.

While it’s impossible to guarantee there will never be high latencies for some pipelines, periods of high latencies across the board are unacceptable. What made this incident particularly egregious was the fact that it went on for over 40 hours, and the whole Etleap team is sorry that this happened. The long resolution time was in large part because we didn’t have the appropriate monitoring to lead us towards the root cause, and we have learned from this and are putting more monitoring of key components in place going forward.