scaleway/pigeon-cluster v2.0.0
Follows compute-instance v6.0.0, which renamed instance_config.enable_heic_transcoding to instance_config.enable_transcoding. This renames the matching per-job passthrough and re-pins the module.
jobs[*].enable_heic_transcodingis renamed tojobs[*].enable_transcoding. Its meaning is unchanged:false(the default) installs no ffmpeg,trueinstalls a general-purpose ffmpeg build on that job's instance. The old name described only the HEIC case, but the flag is exactly what anypigeon-cli transformjob needs regardless of input file type.- Both
compute-instancesource pins —module "job"andmodule "bastion"— move from v5.7.0 to v6.0.0.module "bastion"passes noinstance_config, so it takes the re-pin without a field change.
This is a breaking change that fails silently if you miss it. Terraform treats an unknown attribute on an object-typed variable as a warning, not an error, so a jobs entry still setting enable_heic_transcoding after upgrading to v2.0.0 will plan and apply cleanly — the key is simply dropped. The job's instance then comes up with no ffmpeg installed, and the failure appears later as the transform dying on its first file with ffmpeg's own "decoder not found" error.
When bumping to v2.0.0, grep your jobs list for enable_heic_transcoding and rename every occurrence. A clean plan does not mean you caught them all.
Recorded in ADR-0151.