The Commonwealth of Massachusetts.

Earlier in the year, Last Call Media worked with the Massachusetts Department of Unemployment Assistance on a project in which we implemented automated testing and other automation processes. Word of the successful outcomes traveled through departments. When it came time to enhance the DFML’s program with DevOps automation, they contacted LCM.

What the DFML had was a team of teams each optimized to their own workflows and working on individual pieces of one greater product. The component-based team model increases efficiency as the larger technical foundations of a product are built, yet the integrations between those components can become a blind spot needing special consideration: integration testing was a known need in the project strategy. Last Call Media was brought on to be the integration testing team, and we knew from experience that concentrating testing of all functionality to a separate group, and as a final “phase” all work must pass through, leads to surprise issues arising too late in the life of the project. This was important as the timeline was one of the most important factors of this project: constituents needed to be able to apply for PFML benefits on January 1, 2021, no matter what.

As we began to work with the existing teams, we saw exactly what we could bring to the table: a strong strategy, clear approach, and defined process for integrating all work across every team, and systematically testing that work, as early in the development process as possible, so that fully tested product releases could be done with confidence and ease.

There were four main aspects of this project that needed to be considered in order to achieve success:

• The claimant portal, built using React, where constituents would be able to submit PFML claims and receive updates about the status of those claims,
• The claim call center, where customer service representatives would take calls from claimants and enter their claim information into the claimant portal,
• The claims processing system, the tool in which customer service representatives can process PFML claims via a task queue (and which is fed information from the portal, call center, and other third-party tools), and
• The API that would bring all of these parts together to work seamlessly.

Then, of course, there’s the testing. LCM began our work by establishing three types of tests that all project work would need to pass in order to be considered complete:

• End-to-End (E2E) testing: automated continuous verification of the functionality and integration of all 4 systems.
• Load and Stress testing: verifying the E2E functionality and integration under substantial strain to see what the system can sustain, where it breaks, what breaks it, etc.
• Business Simulation testing: verifying if the people behind the scenes who will be doing this work on a daily basis can effectively perform said work with the systems and functionality that have been put into place, and whether this work can be performed when there is a substantial amount of it.

As we worked to set up the proper tests for the product, we found many opportunities to gain alignment across all of the development teams with our overall testing philosophy: it should be a part of each team’s workflow instead of a final phase removed from the team(s) performing the work. We helped coach each team on delivering value incrementally, and their eventual ownership of where the E2E testing suite impacts their work. LCM brought testing to the program and enabled the teams to absorb it as their own.

September, 2017

The new Mass.gov site launched after roughly a month in pilot mode, and we saw an increase in traffic which corresponded with a small response time bump. The site initially launched with a cache lifetime of over an hour for both the CDN and the Varnish cache. This meant that the site was stable (well insulated from traffic spikes), but that editors had to wait a relatively long time to see the content they were publishing.

November, 2017

We rolled out the Purge module, which we used to clear Varnish whenever content was updated. Editors now knew that it would take less than an hour for their content changes to go out, but at this point, we still weren’t clearing the CDN, which also had an hour lifetime. Site response time spiked up to about two and a quarter seconds as a result of this work; introducing “freshness” was slowing things down on the back end.

December, 2017

We realized that we had a cache-tagging problem. Authors were updating content and not seeing their changes reflected everywhere they expected. This was fixed by “linking up” all the site cache tags so that they were propagating to the pages that they should be. We continued to push in the direction of content freshness, at the expense of backend performance.

To address the growing performance problem, we increased the Drupal cache lifetime to three hours, meaning Varnish would hold onto things for up to three hours, so long as the content didn’t get purged out. As a result of our Purge work, any content updates would be pushed up to Varnish, so if a page was built and then immediately updated, Varnish would show that update right away. However, we saw very little performance improvement as a result of this.

January, 2018

Early in the month, we experienced a backend disruption due to some JavaScript changes that were deployed for a new emergency alert system. In development, we added a cache-busting query parameter to the end of our JSON API URL to get the emergency alerts. However, in the production environment, we were adding one additional, completely uncached request for every person that hit the site. As a result of this relatively minor change, the backend was struggling to keep up (although constituents saw almost no impact because of the layered caching). This illustrated the importance of considering the performance impact of every single PR.

Careful study of the cache data revealed that each time an editor touched a piece of content, the majority of the site’s pages were being cleared from Varnish. This explained the large spike in the response time when the Purge work was rolled out, and why raising the Drupal cache lifetime really didn’t affect our overall response time. We found the culprit to be the node_list cache tag, and so we replaced it with a system that does what we called “relationship clearing.” Relationship clearing means that when any piece of content on the site is updated, we reach out to any “related” content, and clear the “cache tag” for that content as well. This let us replace the overly-broad node_list cache tag with a more targeted and efficient system, while retaining the ability to show fresh content on “related” pages right away. The system was backed by a test suite that ensured that we did not have node_list usages creep back in the future. This earned us a massive performance boost, cutting our page load time in half.  

We found that the metatag module was generating tokens for the metatags on each page twice. The token generation on this site was very heavy, so we patched that issue and submitted the patch back to Drupal.org.

February, 2018

We had another backend disruption due to some heavy editor traffic hitting on admin view; our backend response time spiked up suddenly by about 12 seconds. A pre-existing admin view had been modified to add a highly desired new search feature. While the search feature didn’t actually change the performance of the view, it did make it much more usable for editors, and as a result, editors were using it much more heavily than before. This was a small change, but it took what we already knew was a performance bottleneck, and forced more traffic through it. It demonstrates the value of being proactive about fixing bottlenecks, even if they aren’t causing immediate stability issues. It also taught us a valuable lesson—that traffic profile changes (for example, as a result of a highly desired new feature) can have a large impact on overall performance.

We got a free performance win just by upgrading to PHP 7.1, bringing our backend response time from about 500 milliseconds down to around 300.

We used New Relic for monitoring, but the transaction table it gave us presented information in a relatively obtuse way. We renamed the transactions so that they made more sense to us, and had them broken down by the specific buckets that we wanted them in, which just required a little bit of custom PHP code on the backend. This gave us the ability to get more granular about what was costing us on the performance side, and changed how we started thinking about performance overall.

We added additional metadata to our New Relic transactions so we could begin answering questions like “What percentage of our anonymous page views are coming from the dynamic page cache?” This also gave us granular insight on the performance effects of changes to particular types of content.

We performed a deep analysis of the cache data in order to figure out how we could improve the site’s efficiency. We broke down all the cache bins that we had by the number of reads, the number of writes, and the size. We looked for ways to make the dynamic page cache table, cache entity table, and the render cache bin a little bit more efficient.

We replaced usages of the url.path “cache context” with “route” to make sure that we were generating data based on the Drupal route, not the URL path.

On the feedback form at the bottom of each page on the site, the form takes a node ID parameter, and that’s the only thing that changes when it’s generated from page to page. We were able to use “the lazy builder” to inject that node ID after it was already cached, and we were able to generate this once, cache it everywhere, and just inject the node ID in right as it was used.

We took a long hard look at the difference between the dynamic page cache and the static page cache. Without using the Drupal page caching, our average response time was 477 milliseconds. When we flipped on the dynamic page cache, we ended up with a 161 millisecond response, and with the addition of the static page cache, we had a 48 millisecond response. Closer analysis showed that since Varnish already handled the same use case as the Drupal page cache (caching per exact URL), the dynamic page cache was the most performant option.

We automated a nightly deployment and subsequent crawl of site pages in a “Continuous Delivery” environment. While this was originally intended as a check for fatal errors, it gave us a very consistent snapshot of the site’s performance, since we were hitting the same pages every night. This allowed us to predict the performance impact of upcoming changes, which is critical to catching performance-killers before they go to production.

As a result of all the work done over the previous 5 months, we were able improve our content freshness (cache lifetime) from 60 minutes to 30 minutes.

April, 2018

We enabled HTTP2, an addition to the HTTP protocol that allows you to stream multiple static assets over the same pipeline.

We discovered that the HTML response was coming across the wire with, in some cases, up to a megabyte of data. That entire chunk of data had to be downloaded first before the page could proceed onto the static assets. We traced this back to the embedded SVG icons. Any time an icon appeared, the XML was being embedded in the page. In some cases, we were ending up with the exact same XML SVG content embedded in the page over 100 times. Our solution for this was to replace the embedded icon with an SVG “use” statement pointing to the icon’s SVG element. Each “used” icon was embedded in the page once. This brought pages that were previously over a megabyte down to under 80 kilobytes, and cut page load time for the worst offenders from more than 30 seconds to less than three seconds.

We reformulated the URL of the emergency alerts we’d added previously to specify exactly the fields that we wanted to receive in that response, and we were able to cut it down from 781 kilobytes to 16 kilobytes for the exact same data, with no change for the end users.

We switched from WOFF web fonts to WOFF2 for any browsers that would support it.

We used preloading to make those fonts requested immediately after the HTML response was received, shortening the amount of time it took for the first render of the page pretty significantly.

We added the ImageMagick toolkit contrib module, and enabled the “Optimize Images” option. This reduced the weight of our content images, with some of the hero images being cut by over 100 kilobytes.

We analyzed the “Did you find what you were looking for?” form at the bottom of every page, and realized that the JavaScript embed method we were using was costing our users a lot in terms of bandwidth and load time. Switching this to a static HTML embed with a snippet of JavaScript to submit the form had a dramatic improvement in load time.

The Mass.gov logo was costing the site over 100 kilobytes, because it existed as one large image. We broke it up so that the image of the seal would be able to be reused between the header and the footer, and then utilized the site web font as live text to create the text to the right of the seal.

LCM worked with the Commonwealth to build an extract-transform-load tool (ETL) that runs whenever an updated data set is provided by GIS. The ETL takes millions of records from the FGDB dataset, normalizes them, and imports them into an AWS RDS database. The state had a proof-of-concept process for this that took hours to run. By leveraging AWS Fargate and the open-source GDAL library the required time was brought to under 10 minutes.

After the ETL, we built an API endpoint using Serverless.js (AWS) that takes in addresses in a typical mailing address format (which allows for many potential variations). We leverage libpostal to separate the address into distinct address components, and then perform a query against the RDS database to see if any matches are returned.

The serverless architecture of both the ETL and the API endpoints are highly scalable and secure, and the state is only charged while they are actually being used. This allowed us to create a flexible address matching system that took advantage of their internal dataset of unique addresses, and allowed user input in a wide variety of formats, for a very small cost.

The Commonwealth of Massachusetts was faced with a difficult to use web experience that stood in the way of a constituent base that wanted to interact with government services through the web. The goal was set to create new primary face of Massachusetts government that was constituent centric to enable fast, easy, and wicked awesome interactions with state services.

Prior to the migration to Drupal 8, the Commonwealth’s antiquated, proprietary content management system reflected its internal hierarchical structure instead of organizing content in a way that made sense to its visitors. That meant, for example, that a visitor looking to start a business in Massachusetts had to visit four or more separate department sites in order to get started. The Commonwealth sought to improve the user’s experience by focusing on guiding the visitor through the various services the Commonwealth provides over helping them navigate the complexities often found in a massive state organization. 

Massachusetts Digital Service (MassGovDigital) had two core philosophies that drove their decision making: focus on the constituent experience and use data to support all decisions. They envisioned a unified user experience, where tasks like preparing to renew a driver’s license matched the ease of use of shopping on a major retail website. The MassGovDigital team was challenged to apply private-sector data analytics savvy, like defining and tracking constituent “conversions.” Gaining access to analytics at their fingertips was key for site authors to make fully-informed decisions to optimize their content for readers. A high-performing site search would ensure content findability. 

Last Call Media and the MassGovDigital teams worked together to improve the Drupal authoring experience in order to deliver the next chapter of the state’s digital future. The end result represents a significant leap towards the goal of a new primary face of Massachusetts government that centers on constituents—enabling fast, easy, and wicked awesome interactions with state services.

Improving the Drupal authoring experience

If you can improve the authors’ ability to create and improve content, then the constituents will be better enabled for fast, easy, and wicked awesome interactions with state services. For Last Call Media, improving the authoring experience meant a better interface for communicating data about each piece of content. This data was a mix of content scoring and direct feedback from constituents. For example, a site visitor submits the feedback form (available on any page) and says whether they found the page to be helpful or not (“Did you find what you were looking for on this webpage?”). The raw data goes into the calculation of their “grades.” The raw feedback comments also become visible in a tab in the editor interface, where the editor can focus in on what is or isn’t working well about the page. Surfacing this data, along with making the interfaces more intuitive and the content easier to add and manage, enabled authors to more rapidly improve their content in a more targeted way. According to ForeSee results, the release of the reimagined authoring experience directly correlated with a never before seen increase in Customer Satisfaction.

Retheming Drupal’s default admin site was a critical opportunity to improve usability for authors. With a pattern library, Last Call Media applied these improvements at a large scale without the prohibitively high effort of creating designs for every single page. Paying extra attention to vertical spacing intervals Last Call Media created a clearly defined page hierarchy. At the same time, flipping the brand color scheme provided a user experience that’s consistent with the main site, yet unique.

By using a shared design vocabulary, these ideas could be applied site wide—even the pages that weren’t yet designed—to start development and apply the broad strokes to the theme, while the design team focused on some areas of the site that required more attention for usability. Faced with the challenge to deliver complex content with forms assembled from a number of differently-styled modules, form elements became a focus area. In addition to a unified styling system to improve usability for all form elements, horizontal lines now mark the end of one form element and the beginning of the next. Vertical lines do the same for nested re-orderable elements.

A clear path to publish content 

Last Call Media also created a new and improved version of the Admin Toolbar to streamline navigation for content editors and creators. In addition to the Admin Toolbar module, a separate custom toolbar module specific to the Admin Theme was created to handle customizations. Now, buttons provide easy access back to the login page, with a separate button to return to the main site. A custom tab directs content editors to their most important pages.

Before, there was no direct way for authors to quickly access their top two pages: the page to add content and the page to create new documents. To balance that functionality with a slim line top bar experience, buttons were added to those pages right to the tab menu. Last Call Media built a custom “Add Content” page with attentional experience in mind; Mass.gov, at the time of this project, had about 35 content types for an author to scroll through. 

A faster and easier authoring experience has allowed authors to keep improving and adding to their content. This keeps constituents in the know and continues to grow Mass.gov as a single source of all the information they need to interact with government services. Read more about how Last Call Media did this here.

In the discovery phase, we reviewed the marketplace to find the most appropriate CDN for the State’s use case, balancing security, performance, and cost considerations. Ultimately, Cloudflare was selected as the best fit because of its extensive firewall and DDOS protections, and granular cache control using “Cache Tags,” which have the potential to boost performance for the constituents and reduce the risk of site instability.  

The first, and perhaps most critical concern we addressed in the course of this project was that the CDN needed to be resilient, serving pages even if the site itself was not functioning properly. For example, the development team does code releases periodically that take the backend of the site completely offline, but constituents still need to be able to access content during this time. To meet this requirement, we adjusted the site’s caching headers to include directives to serve cached responses in the event of an error response received from the origin.  As a result, constituents are able to access the majority of Mass.gov, even if a catastrophic event takes the web servers completely offline.

As a government website, Mass.gov is always at risk of attack from malicious actors. To mitigate this risk, Last Call Media undertook extensive configuration and testing of Cloudflare’s various security features, including the Web Application Firewall (WAF), DDOS protections, and custom firewall rules. We had a few hiccups along the way with configuring the security features (at one point, content authors were receiving CAPTCHA verifications when submitting their changes), but were ultimately able to work through these issues to dial in the right balance of security and ease-of-use.

Next, we implemented Cloudflare’s brand new “Workers” feature, which gives granular control over CDN functionality using a javascript “service worker.” The Worker we wrote for this project handles more than 6 million requests a day, and gives the Commonwealth the ability to test and deploy CDN level changes to development, staging, and production environments independently, making it much faster and safer to verify and release changes. The worker implementation benefits the Commonwealth giving them flexibility for the future, while also reducing cost over the previous CDN.

These workers were also integral to the success of this migration beyond what we had initially imagined. During the testing and release phases of the project, they gave us a mechanism for fixing changes that was reviewable and testable. Having a well-defined review and deployment process improved the team’s visibility into what changes were being made, and let us avoid silly mistakes. Overall, we felt the development team’s velocity was greatly improved by using this workflow.

Discovery

The success of Mass.gov hinges on getting the right feedback from constituents to site authors. Our first step in overhauling the way Mass.gov collects feedback was to define what we needed to know about each page in order to improve it, so we could design the feedback component around that. It consisted of the following:

• Whether or not users found what they were looking for, and what that was.
• Contextualize the above by knowing how satisfied users are with the page, and what they came to the site to do.
• Very detailed feedback that could only be provided through their user panel, a list of nearly 500 constituents who have volunteered to test new features for the site.

With our broad goals defined, we wanted to make sure the feedback component was working on a more granular level as well. We conducted a series of interviews with site authors asking how to best reach their users, and gained some valuable insight. Here’s what they told us:

• Too much information in the feedback form would scare users away.
• Feedback was being submitted with the expectation of a response, and organizations wanted to be able to respond.
• But, not all organizations would be able to respond, so a variety of contact options needed to be available to them.

Strategy

We combined what we learned above with our best practices to make a set of requirements that we used to define a strategy. It was immediately obvious that this feedback component needed to do a lot! And like site authors told us, if we showed that to users all at once, we might scare them away.

So to maximize the amount of responses we’d get, we decided to lower the effort for submission by presenting these options one at a time, starting with the step that takes the least amount of commitment, and increasing with each step. So users can submit a little bit of feedback, and then opt into submitting a little more, and then keep going.

Designing the feedback form

With a clear strategy in place, we designed the following component.

On first load, the component is very simple — it’s only asking users if they found what they were looking for.

Once users have made a selection, the component expands with fields asking them what they were looking for.

Site authors have the option of including an alert here that tells users this form is not for urgent assistance, and directs them to a better place where they can do that.

In the above example, the organization who is responsible for this page is able to respond directly to feedback. So if users say they would like a response, a form opens up for them to enter their contact information. If the organization was not able to respond directly to feedback, a brief explanation of why would appear there instead.

 After submitting, users are thanked for their feedback. 

Seen above, organizations are given the option to link to their contact page. This is commonly used if the organization is unable to respond directly to feedback.

Users are then given the option to take a short survey, where they can provide more detailed feedback.

After submitting the survey, users are given the opportunity to join the Mass.gov user panel. This is the largest commitment available for providing feedback, so it’s at the very end!

So that’s how feedback is collected on the site. But what happens to it after that?

Displaying feedback to site authors

Feedback submitted through the site can be viewed per node, i.e. a site author can go to a specific page through the backend of the site and view all the feedback submitted for that page. But a lot of feedback can be submitted for a single page, and on top of that, site authors are often responsible for multiple or even many pages. Combing through all that feedback can be a prohibitively daunting task, or simply not possible.

To help with this, we designed the “Content that needs attention” panel for site authors.

The “Content that needs attention” panel appears on the welcome page on the backend of the site, making it one of the first things site authors see after logging in. It displays the page titles of their 10 pages with the lowest scores from users, sorted by page views. By showing site authors their content that’s seen by the most people first, we’re helping them prioritize what to work on next.

We’re giving site authors additional information about the content right in the component, helping them make decisions at a glance. In addition to the aforementioned page titles, scores, and page views, we’re showing them the content type (since some titles can be very similar on this site), the date they last revised it (in case that helps them know how badly this content needs attention), and something a little surprising… a “Snooze” button!

We put a snooze button in because once site authors make an improvement to content, it’s no longer helpful for them to see it here. So, the way it works is that they make an improvement to content, then hit “Snooze,” and it’ll disappear from this list for one month. At the end of that month, one of two things will have happened: 1) the content will have improved enough to no longer appear on this list, or 2) the content needs more improvement, and will appear back on this list.

Our first step in initiating the migration was to investigate what we were dealing with. At the time, mass.gov had around 600,000 revisions with a moderation state that needed to be migrated from Workbench Moderation to Content Moderation. We started off by digging into Content Moderation’s code to fully understand the complexities and layers of the switch. We found the systems to be very different and with no pre-designed migration path, we deduced that we needed to create a handmade migration path from scratch, outside of a standard Drupal 8 migration. Once the initial configuration of this path was set up, the process was just to keep building the migration through trial and error and figuring out the fastest plan of action.

We knew that switching from Workbench Moderation to the new core module, Content Moderation meant the mass.gov site and its authors would benefit. For a government site, security is always a concern and therefore a top priority. When working with a core module, it is actively supported for security and for any updates that core has as opposed to still running on a contributed module. 

After we felt sound on the coding portion of the switch, we wanted to make sure we were in alignment about expected workflows, transitions, non-transitions, and revision states. We started with around 20 transitions in Workbench Moderation that we were able to consolidate to 5 transitions in Content Moderation for a more optimized workflow. 

We also worked on rebuilding some views such as the “all content view” and mass.gov’s specific dashboard called, “MyContent Block”, which contains all the content the logged in author is watching. 

After a successful switch, mass.gov users are leveraging Content Moderation to moderate content. We are implementing a patch to Content Moderation for the view for “filter by moderation state”. This filter was missing some indices that would cause MySQL to incorrectly do a full table scan instead of an indexed scan of content that caused a lot of performance issues. We ended up writing a patch to include the missing indices that would bring down the query load time on the “MyContent Block” view from 15,000 milliseconds to 200 milliseconds. We plan for this patch to be made available for other sites experiencing the same issue with a lot of content like Mass.gov does.

The Commonwealth of Massachusetts maintains several APIs used by a variety of internal and external teams.

The APIs are built leveraging AWS API Gateway, which allows each granted user to be assigned an API key with various functionality such as rate limiting and access one or more of the state’s APIs. This system is great to work with, but requires a user who is managing the applications to be very familiar with AWS and API gateway in order to add or modify a user’s access to each API.

The application requires OAuth2 authentication through GitHub in order to access the tool, and has a ReactJS frontend that allows authenticated users to easily administer the keys.