Rules for Software Delivery ...
These are rules I would love to enforce for any 3rd party application vendor when you’re shipping stuff to me. I love the 12 Factor rules but they are lengthy. I get most of the value from these…
- Version all artifacts
- Use Semantic Versioning
- Artifacts are sacrosanct - once an artifact is produced at a specific version, it is essentially a shippable product. Don’t open it up to change something. Need to fix something? Revision and fix forward.
- Package using a non-distribution & non-platform specific package format first (e.g. .zip/.tar.gz) Then build distribution / platform specific packages.
- If you’re using a binary packaging model (e.g. rpm), make your source available for packaging.
- Include the version in the package name Don’t package up third party open source tools and deliver them on someone else’s behalf.
- List dependencies in the packages. If possible, don’t tie to specific minor versions of tools. i.e. depend on MySQL 5.1 not MySQL 5.1.73
- Include documentation in your package about installing & configuring your software. A README.txt or README.md is pretty normal.
- Please do not add do a bunch of pre or post install magic in your packaging upon installation. Copying files to directories, setting up permissions, relevant users etc. is all fine.
- Where using distribution specific packages, a simple yum install top-level-package should install your package and all of it’s dependencies.
- Please try and adhere to FHS and the distributions standards for installation on *nix operating systems.
- Please integrate in with the local system startup for the distribution you’re targeting when providing distribution specific packages.
- Install with sane defaults that preferably allow me just to start your application without configuring stuff, or with minimal (obvious) configuration. Helps me play with stuff in dev.
Support / Production
- LTS - Have a support strategy that lines in with common distribution LTS support strategies.
- Have a debug mode which enables more verbose logging and allows you to capture inputs & outputs from your various channels/services. Particularly in a distributed application.
- If configuration is complicated (i.e. you need to edit more than one file), provide a way to show configuration
With the amount of code, libraries and full blown systems now out there, delivery is becoming more important. Don’t try and re-invent the wheel. Plenty of other people have been doing this for some time.