Zulip

We've been using PostgreSQL since the very early days of Zulip, but we actually didn't use it from the beginning. Zulip started out as a MySQL project back in 2012, because we'd heard it was a good choice for a startup with a wide community. However, we found that even though we were using the Django ORM for most of our database access, we spent a lot of time fighting with MySQL. Issues ranged from bad collation defaults, to bad query plans which required a lot of manual query tweaks.

We ended up getting so frustrated that we tried out PostgresQL, and the results were fantastic. We didn't have to do any real customization (just some tuning settings for how big a server we had), and all of our most important queries were faster out of the box. As a result, we were able to delete a bunch of custom queries escaping the ORM that we'd written to make the MySQL query planner happy (because postgres just did the right thing automatically).

And then after that, we've just gotten a ton of value out of postgres. We use its excellent built-in full-text search, which has helped us avoid needing to bring in a tool like Elasticsearch, and we've really enjoyed features like its partial indexes, which saved us a lot of work adding unnecessary extra tables to get good performance for things like our "unread messages" and "starred messages" indexes.

I can't recommend it highly enough.

I have mixed feelings on GitHub as a product and our use of it for the Zulip open source project. On the one hand, I do feel that being on GitHub helps people discover Zulip, because we have enough stars (etc.) that we rank highly among projects on the platform. and there is a definite benefit for lowering barriers to contribution (which is important to us) that GitHub has such a dominant position in terms of what everyone has accounts with.

But even ignoring how one might feel about their new corporate owner (MicroSoft), in a lot of ways GitHub is a bad product for open source projects. Years after the "Dear GitHub" letter, there are still basic gaps in its issue tracker:

• You can't give someone permission to label/categorize issues without full write access to a project (including ability to merge things to master, post releases, etc.).
• You can't let anyone with a GitHub account self-assign issues to themselves.
• Many more similar issues.

It's embarrassing, because I've talked to GitHub product managers at various open source events about these things for 3 years, and they always agree the thing is important, but then nothing ever improves in the Issues product. Maybe the new management at MicroSoft will fix their product management situation, but if not, I imagine we'll eventually do the migration to GitLab.

We have a custom bot project, http://github.com/zulip/zulipbot, to deal with some of these issues where possible, and every other large project we talk to does the same thing, more or less.

We actually started out on Travis CI, but we've migrated our main builds to CircleCI, and it's been a huge improvement.

The reason it's been a huge improvement is that Travis CI has a fundamentally bad design for their images, where they start with a standard base Linux image containing tons of packages (several versions of postgres, every programming language environment, etc). This is potentially nice for the "get builds for a small project running quickly" use case, but it's a total disaster for a larger project that needs a decent number of dependencies and cares about the performance and reliability of their build.

This issue is exacerbated by their networking infrastructure being unreliable; we usually saw over 1% of builds failing due to transient networking errors in Travis CI, even after we added retries to the most frequently failing operations like apt update or pip install. And they never install Ubuntu's point release updates to their images. So doing an apt update, apt install, or especially apt upgrade would take forever. We ended up writing code to actually uninstall many of their base packages and pin the versions of hundreds of others to get a semi-fast, semi-reliable build. It was infuriating.

The CircleCI v2.0 system has the right design for a CI system: we can customize the base image to start with any expensive-to-install packages we need for our build, and we can update that image if and when we want to. The end result is that when migrating, we were able to delete all the hacky optimizations mentioned above, while still ending up with a 50% faster build latency. And we've also had 5-10x fewer issues with networking-related flakes, which means one doesn't have to constantly check whether a build failure is actually due to an issue with the code under test or "just another networking flake".

We've been incredibly happy with the choice of Markdown as the composition language for Zulip:

• Everyone can write it, since it's just like email and many other products use it.
• It has a WYSIWYG type experience, so we don't feel the need to integrate a fancy editor like Quill.
• There are mature markdown processors for every programming language. Having compatible, high-quality implementations for Python and JavaScript is important for doing local echo properly (see our linked Markdown documentation for details of why is necessary).
• Rendering to HTML is fast, easy to do automated tests for, and easy to debug.

We have had to customize our Markdown implementation somewhat for the chat context, because we don't have the guarantee that content is not split across multiple messages (a good example is that you don't want the "automatically change numbered lists to start at 1" feature in default markdown; you want to restrict that to numbered lists where all the numbers are 1, for example) .