Convos is a chat application that runs in your browser. It allows you to be persistently connected to IRC servers, where you can leave the office and up the conversation on the bus without being disconnected.
Convos was born in June 2012. It has been developed to scratch an itch that both myself and Marcus Ramberg had: We wanted to be connected to IRC, but felt that IRSSI running in screen was quite cumbersome, especially seen from a mobile perspective: We wanted an web interface that was responsive and therefore fit on any medium we would carry around.
We have both been using Convos for a long time, but the project itself lost traction after we found some non-fixable issues in the backend that are seen when the load on the server goes up. The issue comes from the lack of error handling to the database, which I will come back to later
Note that even with this issue, we still use Convos on a daily basis. Both for our personal conversations and at work.
This post is about giving an insight into what went wrong, and what the next version of Convos will do to fix it.
Convos does its best to use code from CPAN and factor out re-usable code back to CPAN. This chapter will give an insight into the core techonologies in Convos:
Convos supports IRC. IRC is a very old, stable but also flexible protocol. The main reason why we support IRC is that we wanted to talk with people on irc.perl.org and freenode, but the protocol and server side is flexible enough to be extended if you want more functionality on an in-house IRC server.
Mojo::IRC started off as a part of Convos, but we released it to CPAN as soon as we thought it was stable enough. Since then we have had a numerous of contributors and the module as become quite mature. The latest addition to the distribution is Mojo::IRC::UA, which tries to emulate state and order on top of the “order less” IRC protocol.
Mojo-Redis and Mojo-Redis2
The current version of Convos use Mojo::Redis. We selected Redis as our backend, since we wanted the pub/sub feature. Pub/sub is a mehanism where one client can publish a message and all the listeners receives the message, without the need for polling.
Mojo::Redis has since then been deprecated, since it did not have any error handling! I think looking back, we thought that the only error we would see (because of the nature of Redis) would be disconnects, but once you get high load on a system (client and/or server), anything can happen – and it did.
We hardly ever notice when running with a few users and low load, so Convos is still usable, but it was still not the kind of product we wanted to ship.
So I made Mojo::Redis2. This is an improved version which will eventually replace the original namespace of Mojo::Redis. It has proper error handling and also support for blocking requests.
The work to replace Mojo::Redis started in December last year, but I got sidetracked since there was too much to deal with. Also, after we received feedback from users that wanted to try out Convos without a database backend, we decided that we wanted to make a version of Convos that could work with just memory or file system as “backend”.
The current in-progress branch of Convos does not require Mojo::Redis, nor Mojo::Redis2, just a file system or (a lot of) free memory.
We wanted more than just a plain text interface. After all, we are working with web technologies, so any link posted should result at least a link or preferrably rich content.
We decided to create Mojolicious::Plugin::LinkEmbedder. This module can take an URL, and extend it into into metadata. In some cases it’s enough to just look at the URL, but in other cases the URL need to be followed and the content must be parsed. Either way, if the LinkEmbedder has a module for parsing the URL, it will result in a richer experience.
The current version of LinkEmbedder has support for Imgur, XKCD, Spotify, Travis, Twitter, Appear In, a bunch of pastebins and many of the most used video services. There are open issues for support for more services, but the sky is really the limit for what to support.
The next version of Convos is focused around the API. Swagger2 is “The Worldâs Most Popular Framework for APIs”. We use this framework in Convos to document and define the API resources.
This should make it very easy for anyone else to make competing frontends to the Riotjs based frontend shipped with Convos. This could span from an iOS app to integration with IRSSI.
The next version of Convos will use the design principles from Google’s Material design, implemented in the Materialize project. We decided that instead of rolling our own CSS, we should try to stand on someone else’s shoulders to avoid wasting too much time on design decisions. And besides: Material design looks very good!
Internals in “Convos revamped”
The next version of Convos is completely redesigned. It’s supposed to be easier to follow the logic, the objects should have a clean separation between “backend storage” and Perl space, and the API should support hooks provided by third party plugins.
The working draft can be seen in the batcode branch.
The current version has a single process running the backend, while the frontend can be started as a daemon or under hypnotoad
The next version will ultimately be able to be started with a command like this:
$ curl https://convos.by/code-1.0 | perl - daemon
Even if Convos doesn’t reach that point, it will run as a single process – both the backend and frontend.
The application has embedded the code from Mojolicious::Plugin::RequestBase, which makes it easy to run behind a reverse proxy.
The controller classes holds just enough logic to pass data from the Swagger based REST API, to “Convos::Core”.
There are currently three controllers defined: “User” handles login, register profile data and deletion of an account. “Connection” handles add/remove connections to chat servers and joining/parting conversations. “Chat” handles actions related to a single conversation and should be able to stream messages in the future.
This is the heart of Convos. The core takes care of connections, conversations, can persist to a backend and provide hooks for plugins.
The core structure is as following:
.---------. ____| Backend | .------._/ '---------' | Core | '------' .------. .-------------. .---------------. \______| User |___| Connections |___| Conversations | '------' '-------------' '---------------'
This means that Convos is a multi user application, that can persist to any backend (memory, file storage, redis, …) and connect to any chat server, as well as have keeping any number of conversations active.
The way the “Backend” is hooked into the rest of the object graph is by events. Any user, connection or conversation can emit new events that the Backend can choose to persist to storage. The default backend is a file-based backend, which enables Convos to be started without any additional dependencies, except Perl and a couple of modules from CPAN.
The important thing when designing these events is to not think about what kind of “Backend” you have.
Status of the revamped version
The current batcode branch has a lot of the REST resources defined, implemented and tested, but there is still missing features such as no way to log out and no way to send/stream messages. The bottleneck now is however the frontend. We currently have no tests for the Riotjs code and the user interface is in best case infantile.
I think the next version of Convos has a structure that is quite simple to follow. The idea behind using events, instead of a third party database makes the separation of concerns clearer.
Got questions, feedback, want to contribute or just see what’s going on with the project? Follow Convos on Twitter or join the IRC channel #convos on irc.freenode.net.
Demo of the current version.