Kubernetes vs osm2pgsql

osm2pgsql ships two or three releases a year and the last two carry substantial new machinery. 2.3.0 reworked tile expiry: polygons are now expired by shape rather than bounding box, and an opt-in diff expire mode computes the symmetric difference between an object's old and new geometry so a small edit to a large feature only invalidates the tiles it actually touched. It also added configurable limits on how many tiles one geometry or one run can expire. Before that, 2.2.0 introduced Locators for fast region lookup during import and process_deleted_* callbacks for handling removed objects in Lua.

The project is adding primitives rather than features, and saying so explicitly — 2.2.0 describes Locators and deleted callbacks as building blocks for things users have requested for years. The pattern is to move work that previously happened in the database after import into the import pipeline itself: region tagging via Locators instead of a post-import spatial join, deletion handling via callbacks instead of database triggers. The expiry work follows the same logic, pushing precision upstream so downstream re-rendering does less. The new expiry limits are a defensive addition, added specifically because vandalism or misconfiguration can otherwise generate billions of tiles and exhaust memory.

Expect diff expire to move from opt-in toward default once it has been exercised, following the pattern of the middle table format that became default in 1.11.0. The experimental osm2pgsql-expire command introduced in 2.2.0 is the other loose thread — it is explicitly marked as subject to change and has not stabilised.