Input: 2014q2 post-mortem
I'm going to start doing quarterly post-mortems for Input development. The goal is to be more communicative about what happened, why, what's in the works and what I need more help with.
NB: "Fjord" is the name of the codebase that runs Input.
Bug and git stats
Bugzilla ======== Bugs created: 63 Bugs fixed: 54 git === Total commits: 151 Will Kahn-Greene : 143 (+15123, -4602, files 446) ossreleasefeed : 3 (+197, -42, files 9) Joshua Smith : 2 (+65, -31, files 5) Anna Philips : 1 (+367, -3, files 12) Swarnava Sengupta : 1 (+2, -2, files 1) Ricky Rosario : 1 (+0, -0, files 0) Total lines added: 15754 Total lines deleted: 4680 Total files changed: 473
We added a lot of lines of code this quarter:
April 1st, 2014: 15195 total, 6953 Python
July 1st, 2014: 20456 total, 9247 Python
That's a pretty big jump in LOC. I think a bunch of that is the translation-related changes.
Contributor stats
5 non-core people contributed to Fjord development.
I spent some time over the weekend finishing up Vagrant provisioning script and rewriting the docs. I'm planning to spend some more time in 2014q3 reducing the complexity and barriers for setting up a Fjord development environment to the point where someone can contribute.
Additionally, I'm planning to create more bugs that are contributor-friendly. I started doing that in the last week. I think a good goal for Input is to have around 20 contributor-y bugs hanging around at any given time.
Accomplishments
Site health dashboard: I wrote a mediocre site health dashboard that's good enough to give me a feel for how the site is performing before and after a deployment. This still needs some work, but I'll schedule that for a rainy day.
Client side smoke tests: I wrote smoke tests for the client side. I based it on the defunct input-tests code that QA was maintaining up until we rewrote Input. There are still a bunch of tests that I want to write to have a better coverage of things, but having something is way better than nothing. I'm hoping the smoke tests will reduce the amount of manual testing I'm doing, too.
Vagrant: I took some inspiration from Erik Rose and DXR and wrote a Vagrant provisioning shell script. This includes a docs overhaul as well. This work is almost done, but needs some more testing and will probably land in the next week or two. This will make peoples' lives easier.
Automated translation system (human and machine): I wrote an automated translation system. It's generalized so that it isn't model/field specific. It's also generalized so that we can add plugins for other translation systems. It's currently got plugins for Dennis, Gengo machine translation and Gengo human translation. I turned the automated human translation on yesterday and it seems to be working well. That was a HUGE project. I'm glad it's done.
One thing it includes is a lot of auditing and metrics gathering. This will make it possible for me to go back in time and look at how the translation system worked on various Input feedback responses and hone the system going forward to reduce the number of human translations we're doing and also reduce the number of problems we have doing them.
Better query syntax: We were upgraded to Elasticsearch 0.90.10. I
switched the query syntax for the dashboard search field to use
Elasticsearch simple_query_string
. That allows users to express
search queries they weren't previously able to express.
utm_source and utm_campaign handling: I finished the support for
handling utm_source
and utm_campaign
querystring
parameters. This allows us to differentiate between organic feedback
and non-organic feedback.
More like this: I added a "more like this" section to the response view. This makes it possible for UA analyzers to look at a response and see other responses that are similar.
Dashboards for you, dashboards for everyone!
I'm putting this in its own section because it's intriguing. I'll write another blog post about it later in July as things gel.
On Thursday, a couple of days after d3 training that Matt organized, I threw together a better GET API for Input feedback responses. It's not documented, it probably has some bugs and it's probably going to change a bit, but the gist of it is that it lets you more easily build a dashboard that meets your needs against live Input data.
Here's a proof-of-concept:
http://bl.ocks.org/willkg/c4d5a272f86ae4510750
That's looking at live Input data using the new GET API. The code is in a GitHub gist. It auto-updates every 2 minutes.
The problem is that I've got a ton of Input work to do and I just can't write dashboard code on Input fast enough. Further, of the people I've talked to that use the front page dashboard, they all have really different questions they're asking of the data. I'm hoping this alleviates that bottleneck by letting you and everyone else write dashboards that meet your needs.
I encourage you to take my proof-of-concept, fork the gist, tweak it, use bl.ocks.org or something to "host" the gist. Build the dashboard that answers your questions. Share it with other people. Plus, let me know about it. If you have issues with the API, submit a bug and tell me.
If this scratches the itch I think needs scratching, it should result in a bunch of interesting dashboards. If that happens, I'll write some code in Input to create a curated list of them so people can find them more easily.
Summary
This was a really nuts quarter and parts of it really sucked, but we got a lot accomplished and we laid some groundwork for some really interesting things for 2014q3.
Update April 21st, 2015
LGuruprasad found a bug in the script that caused commits-by-author information to be wrong. Fixed the script and updated the stats!