Configuration Languages Are the Bane of Our Existence

Wednesday, 30 October, 2024 - 14:4515:05 GMT

Paul Komkoff

Abstract: 

It is probably a good idea to make it possible to change some constants in your program without recompiling it. So why it then gets incredibly hard to control these configurations? At which point configuration becomes a program with no tests, written in untyped language, which requires a lot of compute to evaluate and can't be checked in advance? Is it at all possible (and enough) to get rid of these languages and go back to ini files?

If you are like me, you want to know the answers to these questions, and this is what I'm going to talk about. Plus:

  • sendmail.cf was an early sign everybody ignored
  • if you use regular expressions for matching and selecting in your configuration, start writing a premortem
  • when your configuration is more complicated than your program, who is your program now?

Paul Komkoff[node:field-speakers-institution]

Out of 33 years of working with computers and networks, Paul spent 17 in SRE organization. He believes that complexity needs to be actively managed and to know the better ways to fix things we need to explore the depths of failure.

BibTeX
@conference {302251,
author = {Paul Komkoff},
title = {Configuration Languages Are the Bane of Our Existence},
year = {2024},
address = {Dublin},
publisher = {USENIX Association},
month = oct
}