parallelism.md
In HeatStressR: Calculate Heat Stress Indices

Parallel calculations

wbgt_liljegren() defaults to one process. Set workers explicitly for a large call; temporary PSOCK workers compute ordered chunks and are stopped after success or failure. The requested count is capped at the input row count and must be within the detected logical CPU limit. R check core restrictions limit requests to two workers.

wbgt_liljegren(tas, dewp, wind_2m, radiation, time, lon, lat, workers = 2)
heat_indices(tas, hurs, dewp = dewp, wind_2m = wind_2m,
  radiation = radiation, time = time, lon = lon, lat = lat,
  indices = "wbgt_liljegren", liljegren_options = list(workers = 2))

Keep one parallel layer. When a workflow already parallelizes independent files or locations, leave workers = 1. Other indices execute in one process; parallelize independent partitions externally if needed. No automatic workload heuristic changes the requested worker count.

Numeric values, missing masks, component values and diagnostic ordering match single-process calculations. Diagnostics-off worker results contain compact failure summaries; full row-level solver records are transferred only when requested. Avoid requesting diagnostics for production runs that only need WBGT values.

The scalar reference engine remains an internal scientific comparison/fallback path and is accessible through the deprecated legacy API during migration.