I'm not sure what it says about Seattle that one of our biggest yearly events is a May Day protest that wrecks havoc across big chunks of downtown. What even competes? The Blue Angels shut down traffic on the bridges once a summer, and there's the Sea Fair downtown, but reception to those is always pretty muted in my experience. International Workers Day is the big show.
The May Day map I put together to track our reporters has quickly become one of my favorite projects for the Seattle Times. It was real-time, it posed interesting data challenges, and it really exploited our <leaflet-map> element more than anything else we've done so far. While I also wrote a post on it for our dev blog at work, I wanted to call out a couple of other interesting points here.
The most interesting technical detail here is the use of the Twitter streaming API, which delivers nearly instant updates for a search query (either on users, geolocation, or keyword). Node is a great fit for this, with the twitter module offering a readable stream that fires events as new items come in. Our scaffolding, on the other hand, is not intended to be run as a long-standing process, and I didn't really want to retrofit Grunt into a general-purpose application framework. I ended up writing the Twitter part of the app as a completely separate, continuous Node process, which then dumped out its data as a JSON file and started a standard build/deploy in a child process whenever new data arrived.
To store the tweets from the stream, the application uses a SQLite3 database, since that's the easiest way to query and update data. A static data store like this is not something that we've used on projects before, and I don't know if I'd re-use it again. Using SQLite itself is always a pleasure, but reliance on a local database means that I couldn't just clone the project from home and update it when I wanted to change the coloring on Saturday morning. Using cloud storage, like Google Sheets, has a lot of advantages for distributed and remote development.
Working with Twitter itself is an interesting problem, because it's clear that the company has no real coherent plan for outside developers. Over the last few years, the API for user access has been increasingly limited and broken as Twitter tried to drive third-party clients (which don't show ads and don't make money) out of existence. On the other hand, if you are building a Twitter bot, which our map effectively is, it remains a pretty useful and effective service for pub/sub communication. I'm not sure it says very much about Twitter's strategy that they'll let bots run wild while ordinary people are locked into a client monoculture, but that's honestly the least of my frustrations with them at this point.
All that said, I would personally use with this stack again in a heartbeat. Twitter is not the highest social traffic source for the Seattle Times, but almost all of our reporters use it anyway, and it's much nicer to program against compared to Facebook. The impending dilemma is if (or when) Twitter will decide to switch to a "curated" (read: algorithmically-tampered) stream a la Facebook's timeline. When that happens, its value to me as a news developer drops basically to nothing, because I won't be able to guarantee message delivery any more.
Which brings me to the most boring but probably most profound lesson of this project: we need a better build server. The May Day map ran on a box in the office we've affectionately dubbed "Cronda," which also currently tests our traffic alert application and previously powered the Seahawks fan map. In each of those cases, we've jury-rigged together a solution for pulling the latest source code and running builds at regular intervals (the cron Grunt task), but it's not optimal. We can't check on those builds remotely, or restart them if something goes wrong.
At some point, we'll probably move our builds from Cronda to an EC2 box that we can access remotely, but doing so doesn't honestly solve the problem — it just makes it less fragile. Eventually, I think we'll need to look into a real build monitor like Jenkins, which can automate deployments, track error logs, and respond to queries in our teach chat. I'm not entirely looking forward to that, since it feels like a very heavyweight solution, but the more complex our applications get, the more a little up-front rigor will pay off.