Skip to contents

Write one replicate's ForCS log_Summary to a partitioned parquet

write_forcs_log_summary_parquet.Rd

Reads one replicate's log_Summary.csv via read_forcs_log_summary(), masks to core cells, and writes it to <scenario_dir>/<subdir>/replicate=<rep>/part-0.parquet, where scenario_dir is inferred as dirname(dirname(src_path)). This is the writer counterpart to open_forcs_log_summary_dataset(); scenario is embedded as a data column so several per-scenario roots can be unioned and filtered without touching the directory layout.

Usage

write_forcs_log_summary_parquet(
  src_path,
  cell_mask = NULL,
  subdir = "_aggregates/forcs_log_summary",
  staging_dir = NULL
)

Arguments

src_path

Path to one replicate's log_Summary.csv.

cell_mask

Optional data.frame with row/column columns identifying the core-area cells to retain (see read_forcs_log_summary()).

subdir

Path within the scenario directory for the dataset root (default "_aggregates/forcs_log_summary", matching open_forcs_log_summary_dataset()).

staging_dir

Optional directory for the temporary parquet before it is moved into place; NULL (default) stages in the destination directory (same-filesystem atomic rename).

Value

The written parquet path.

Details

The publish is atomic: the parquet is written to a temporary file and then fs::file_move()d into place, so a concurrent reader or a retried write never observes a partial file – safe for many replicate writers running at once against an NFS output directory. When staging_dir is supplied the temporary is written there (e.g. per-host scratch, keeping the interim bytes off NFS) and moved cross-filesystem; the default stages in the destination directory so the move is a same-filesystem atomic rename.

See also

Other ForCS output helpers: open_forcs_log_summary_dataset(), read_forcs_log_summary()