Repository navigation
Replies: 1 comment
|
A data point, and a rule that worked for us (gbrain 0.60.565, managed Postgres, database-only source). Before: 10,923 versions holding 424 MB of text. One company hub page had 915 versions averaging 178 KB. Nearly every version came from a single remembered fact being added to the page's Rule: drop a version when a later version of the same page, or the live page, has the same title, frontmatter and timeline, and the same body outside the Result: 2,711 versions dropped across 20 pages, which freed 295 MB of the 424 MB. No claims were lost, and the hub page went from 915 versions to 47. The rule left alone the versions that held content found nowhere else (612 in our first count): in-between generated lists, and prose edited and then rewritten the same day. This is the third direction above ("skip a version for system-owned fence-only changes"), applied after the fact. Done at write time, these rows would never be created. A system-owned write that changes only a fence ( |
Uh oh!
There was an error while loading. Please reload this page.
Moved from #7566, opened by @mattverlaque on 2026-10-25. Feature requests now live in Discussions > Ideas.
What we see
On a managed Postgres brain (two weeks old),
page_versionsis 175 MB, 12.3k rows for 7.7k pages. Growth concentrates on hub entity pages:Most snapshots come from small additions: a fact filed into the page's Facts table, a timeline entry, a
remember. Each one archives the whole page. As the page grows, every new fact costs a larger snapshot, so storage grows roughly with the square of activity on busy entities. I couldn't find a retention or pruning path forpage_versions(noDELETE FROM page_versionsoutside page deletes).Possible directions
gbrain configkeys like the receipt retention.Happy to help with a PR once there's a preferred direction.
All reactions