Record a database's grid, so tile pruning cannot guess it wrong

cell_grid_ncol() resolves the grid width from a cell_grid table in the database and otherwise falls back to v8's 7200. No release ever wrote that table, so the fallback was always taken — correct for global05 by coincidence and wrong for every usa05 release. The consequence was silent and total: cell_model is Hive-partitioned by a tile id computed from the grid width, so a wrong width computes a different, valid tile id, WHERE tile IN (…) prunes away the only partition holding the cells, and the clicked-cell species list came back empty on v1-v7 rather than failing. Measured on the promoted release: a v7 cell with 477 models returned 0 species.

Usage

cell_grid_write(con, grid_id)

Arguments

con

a writable DBI connection to a serving database

grid_id

grid registry key, e.g. "usa05" or "global05"

Value

the written data frame, invisibly

Details

Call this when building a serving database, before cell_model_tile_check().