diff --git a/python/ray/data/context.py b/python/ray/data/context.py index 076ea77813c3..2fed7c5b6101 100644 --- a/python/ray/data/context.py +++ b/python/ray/data/context.py @@ -36,7 +36,7 @@ # We will attempt to slice blocks whose size exceeds this factor * # target_max_block_size. We will warn the user if slicing fails and we produce # blocks larger than this threshold. -MAX_SAFE_BLOCK_SIZE_FACTOR = 1.5 +MAX_SAFE_BLOCK_SIZE_FACTOR = float(os.environ.get("RAY_DATA_MAX_SAFE_BLOCK_SIZE_FACTOR", "1.5")) # Dataset will avoid creating blocks smaller than this size in bytes on read. # This takes precedence over DEFAULT_MIN_PARALLELISM.