Incremental instance sync

Status: backlog. Requested by the maintainer on 2026-11-06 during the instance image import cycle. Not planned or approved.

Request

"In the future, we will need a way to get optimized sync. We can't be downloading the whole archive every time."

Current behavior

• Every ./v3 import-instance run downloads the source's full repository archive (/downloads/repository.tar.gz), even when the destination already has nearly all of it.

• After the instance image import cycle, a run also downloads the full image ZIP (/downloads/images.zip) whenever any referenced image is missing. A run with no missing images skips that download.

• The source's export cache avoids rebuilding the archive per request, but the bytes still cross the network in full.

Goal

Repeated imports from the same source transfer only what changed since the last import: new or changed text records and the specific missing images.

Questions for a future Step 1

• How the destination tells the source what it already has: the last imported revision, a list of hashes, or a date.

• Whether images can be fetched one at a time by hash instead of as one ZIP, and how removed-post images stay excluded.

• How a changed or rewritten source history is detected and falls back to a full download.

• Which access rules apply, since the full archive is public only when backups are turned on.