Reconstructing a container deployment takes an identity for the image and a record of the surrounding choices. This TECH WEEKLY edition reviews established Docker behaviour and a proposed release-record check.
An image tag tells me what name was requested. It doesn't fully describe what was deployed.
Docker lets you pull an image by digest to identify fixed content. Tags can be updated, including tags that look like version numbers. If a deployment record stores only the tag, someone investigating a later failure may be looking at a different image behind the same name.
I'd retain the resolved digest with the deployment record. For a multi-platform image, I'd also keep the selected platform and the image reference resolved for it. An index identifies a collection of platform-specific images; two machines can legitimately select different members of that collection.
There is another part of the comparison. The image doesn't describe every runtime choice. A different command, mount or environment setting can change behaviour even when the image content matches. Record the relevant configuration version and secret references, while keeping secret values out of the report.
This gives a rollback discussion something concrete to work with. Which content ran? Which platform was selected? Which configuration accompanied it? Keep the observed result as well, because a matching identifier doesn't prove the service was healthy.
Pinning also creates an update responsibility. A fixed digest won't move itself to a patched image. Review the replacement, test it and record the new digest through the normal release process.
For this week's technical review, I'd check whether one deployment can be reconstructed from its record without guessing what the tag meant that day. This is a review of established Docker behaviour, not a claim about a new release.
