What the platform should not do
Most writing about platform engineering is about what to build. Very little of it is about what to refuse, which is strange, because refusing is most of the job.
Blog¶
Moving the firewall off the application server
For a long time the firewall in front of one of our applications was CSF, running on the EC2 instance itself. It worked. Changing a rule meant connecting to the box and running a command, and that sentence is the entire reason we moved away from it.
What an artifact signature proves, and what it does not
I wrote earlier about verifying commit signatures in the pipeline, and mentioned in passing that the build signs its own output. That sentence deserves a post of its own, because signing an artifact is easy and the claims people make about it are usually larger than what it actually supports.
Why our Java builds run on Fargate and our Docker builds do not
Most discussions about self-hosted CI runners start with cost. Managed minutes are expensive, the argument goes, so run your own and save money. That was not why we did it, and cost is usually the weakest reason on the list.
Measuring developer experience without a survey
The usual first move when someone decides to take developer experience seriously is to send a survey. It is the obvious thing to do and it is rarely the most useful thing available, because the data you actually need is already sitting in systems you run.
Local development environments that match production closely enough
Two failure modes, and I have caused both. The first is a local environment so far from production that "it works on my machine" stops being a joke and becomes a daily fact. The second is chasing full parity until starting the thing locally takes fifteen minutes, eats most of a laptop, and developers quietly stop using it.
What migrating logging from Datadog to Dynatrace actually cost us
Every platform team eventually has to take something away. The technical part is usually straightforward. What goes wrong is everything between announcing it and switching it off, and the cost almost never lands where you expect.
Secrets management that developers do not route around
Every secrets policy I have seen written down says the same sensible things. Do not commit credentials. Do not paste them in chat. Do not email them. Every organisation I have worked in has done all three, regularly, including the people who wrote the policy.
The support channel is your best product research
Platform teams are usually bad at user research. We do not run interviews, we do not have a researcher, and the survey we send once a year gets answered by the people who already like us. Meanwhile the same team runs a support channel that receives dozens of real, unprompted, specific accounts of the platform failing its users, every single day.
Observability for developers, not just for on-call
Almost every dashboard I have built or inherited was designed for an operator. Cluster health, node pressure, error rates across the estate, a wall of graphs that makes sense if you are holding the pager and looking for the one thing that is on fire.