November 2022 Retrospective
Three releases for Mudmap! 🎉
- Switched data model to sqlc
- Added multi-stage installer
- pfSense logs
Switching to sqlc
This isn’t something I needed to do so why do it? Two reasons; type safety for the data model and reduce friction when adding or updating models.
I experimented with gRPC for another project a couple of months ago and was awestruck by how much the gRPC compiler does for you. It autogenerates thousands of lines of type safe code which you can then call without needing to write anything yourself. As an average at best developer I loved this. Updating the yaml definition rebuilds the interface with a single command and all the callers get updated. Of course, you might have to change your business logic depending on what you’ve updated, but you don’t need to fiddle with any of the service layer. It’s a huge time saver.
sqlc gives you this same experience but with sql. If tomorrow my
Device model needs to include a
Serial field I would have to update every sql statement, all the
calls and then worry about the implementation within handler and service layers. Instead, by
sqlc I just update my sql queries and model definitions and the entire data model will be
auto generated for me with those changes. Now all I need to do is focus on business logic
changes (which the go type system will catch).
It was a bit of learning curve at first especially with
uuid fields and
null’s but that
speaks more to my flawed database schema design choices. Rant alert: for any product/project
I produce in the future I will never use another
uuid. A Stripe-like id (e.g.
epic_sawfish_123) would be more than sufficient.
Perhaps my biggest cross the bear in Mudmap is the flaky installer. It really gets me down, and I’ve still yet to find a good way to fix this issue - its churned one of my biggest customers too.
This feature does not fix the problem, instead it’s part of my pathway to making the installation process more pleasant.
How it works (starts at Password Confirmation):
A crude drawing
This change introduced a multi-stage modal instead of a single step. Using a multi-stage allowed for better error handling and branching when prompting users for input.
Additionally, the Device’s root password is now being stored in an in-memory database. I’ve made an ideological stand against storing this password in Mudmap’s actual database. I feel that storing the password in a memory datastore is a good practice and does not break the promise of not storing it in the database. go-memdb was an easy choice for this. The detractor is that in the future I may need to switch to an external key value store such as Redis if I spin up multiple services.
By storing a Device’s root password in memory it can now be retrieved during the installation process across multiple stages. Without doing this, each stage would require the user to re-enter the password. This is not only inconvenient but also introduces needless complexities like what happens if they re-enter it wrong - it is another error handler and conditional branch that needs to be dealt with.
Adding this also opens up future enhancements such as a proper Deletion event. When a Device is deleted or the installation fails, Mudmap must be removed from the firewall. This requires the removal of the Mudmap service account before removing the API. Once the Mudmap service account is removed every action taken must be done as the root user. This requires root access which with this new addition does not require user intervention.
A bad practice which I am trying to fix (and this fixes most of it) is not deleting Mudmap artifacts when an installation fails. It is possible that these artifacts on subsequent re-installation attempts may even cause issues. But, the biggest problem is Mudmap isn’t being a good citizen and reverting a user’s device back to the state it was when they first tried the platform out.
I do not expect this will fix failed installations. What I might try next is to let users install it manually. This would require a few steps and may be too much friction, but it gives the user power to do something when an issue is encountered. Another thing I am contemplating is storing some metadata about each device in an attempt to draw some correlations between successful and unsuccessful installation attempts. I am unsure about this though.
This month I also added a few more endpoints and user-facing pages for viewing the following logs:
A small-ish update but added functionality all the same. Users can now view the logging of the above from within Mudmap. The pfSenseAPI audit logger is particularly handy. Eventually I will need to augment this with the user controlling the API, not just the API itself.
Writing this feature required a series of regex parsers to be created. The response from the API returns a single string per event. By creating custom parsers it was possible to create useful struct fields which could then be returned as json objects. A methodology which will likely serve other endpoints as they are integrated. Not captured in this work (scope creep) but planned is reading, updating and deleting system packages. Installing packages will be its own ticket but the methodology to do that is already there.
These small features when stacked up actually make Mudmap much more useful. Having to SSH into every device just to do a system and third party package audit is painful, soon Mudmap will remove that burden.
I use a custom implementation of a zettelkasten which I use to store snippets of information. The storage backend is GitHub and each Zet is just a single markdown file stored in a timestamped directory within my zet repo. Sometimes I want to review something I’ve written, but I’m not at my computer and searching for it on GitHub is inefficient. So I wrote a Go tool which would embed it into this site.
At first, I wrote it to just link directly to the Zet’s markdown file on GitHub. I found this wasn’t a great experience as it would take a few seconds to transition and the layout shift was jarring. I did learn how to render a json file using Hugo’s custom shortcodes though.
Instead of this I decided host the markdown files directly on this site. I enjoyed
writing this because it leverages one of Go’s truly great strengths;
If you’ve come from python you’ve probably heard of or used Jinja2
as the templating engine. It’s a sweet module but its another dependency and having
an inbuilt templating engine within Go is one of its underappreciated features.
How I leverage Go’s
text/template to auto generate Hugo’s frontmatter can be found here.
The code is pragmatic and works but don’t look to it as an exemplar of good but instead
practical. file1, file2 and GitHub actions.
Shock horror FIFA is corrupt as hell. A well crafted overview of how FIFA quickly became a money under the table organisation and how it continues to do so today. When people can wield absolute power (without any oversight) they will abuse it for their own gain.
I’m testing out a new work/fun balance; 3-4 days on mudmap 3-4 days on whatever makes me happy per week. Going too hard too soon on Mudmap has been the thing that burns me out. Also, once I release a feature or update I take a couple of days off to build whatever as a reward. Doing this has actually lead to days where I’ve just jumped on Excalidraw and explored a Mudmap feature to build in the future. I’ll play with this a see where it takes me.
I want to release another Mudmap update before Christmas.
Keep up to date with my stuff
Subscribe to get new posts and retrospectives