You can't backdate an S3 object

I was testing a job that treats S3 objects as stale once they’re older than two weeks. I needed an object that looked two weeks old, so I copied one onto itself with a backdated metadata value:

aws s3 cp s3://my-bucket/path/obj s3://my-bucket/path/obj \
  --metadata "touched=$(date -u -v-14d +%Y-%m-%dT%H:%M:%SZ)" \
  --metadata-directive REPLACE

The command worked, and Last-Modified on the object became the current time. A copy in place is a write, and S3 sets Last-Modified on every write. It’s system metadata, so you can’t set it yourself, and aws s3 sync doesn’t carry it over either.

You can keep your own date in user metadata, as above, and read it back with head-object:

aws s3api head-object --bucket my-bucket --key path/obj

That only helps if the code under test reads that metadata instead of LastModified. For a test, it’s easier to make the age threshold or the clock injectable, or to run against a local S3 such as moto or LocalStack.

Also watch for a side effect of the copy. REPLACE throws away any metadata you don’t pass again, including Content-Type.