How to calculate a Directory size in ADLS using PySpark?

Question

I want to calculate a directory(e.g- XYZ) size which contains sub folders and sub files.
I want total size of all the files and everything inside XYZ.

I could find out all the folders inside a particular path. But I want size of all together.
Also I see

display(dbutils.fs.ls("/mnt/datalake/.../XYZ/.../abc.parquet"))

gives me data size of abc file.
But I want complete size of XYZ.

Asked By: Crime_Master_GoGo

||

Source

Answer 1

The dbutils.fs.ls doesn’t have a recurse functionality like cp, mv or rm. Thus, you need to iterate yourself. Here is a snippet that will do the task for you. Run the code from a Databricks Notebook.

from dbutils import FileInfo
from typing import List

root_path = "/mnt/datalake/.../XYZ"

def discover_size(path: str, verbose: bool = True):
  def loop_path(paths: List[FileInfo], accum_size: float):
    if not paths:
      return accum_size
    else:
      head, tail = paths[0], paths[1:]
      if head.size > 0:
        if verbose:
          print(f"{head.path}: {head.size / 1e6} MB")
        accum_size += head.size / 1e6
        return loop_path(tail, accum_size)
      else:
        extended_tail = dbutils.fs.ls(head.path) + tail
        return loop_path(extended_tail, accum_size)

  return loop_path(dbutils.fs.ls(path), 0.0)

discover_size(root_path, verbose=True)  # Total size in megabytes at the end

If the location is mounted in the dbfs. Then you could use the du -h approach (have not test it). If you are in the Notebook, create a new cell with:

%sh
du -h /mnt/datalake/.../XYZ

Answered By: Emer

Answer 2

Try using the dbutils ls command, get the list of files in a dataframe and query by using aggregate function SUM() on size column:

val fsds = dbutils.fs.ls("/mnt/datalake/.../XYZ/.../abc.parquet").toDF

fsds.createOrReplaceTempView("filesList")

display(spark.sql("select COUNT(name) as NoOfRows, SUM(size) as sizeInBytes from fileListPROD"))

Answered By: Manoj Pandey

Answer 3

The @Emer answer is good, but can hit a RecursionError: maximum recursion depth exceeded really quickly, because it does a recursion for each files (if you have X files you will have X imbricated recursions).

Here is the same thing with recursion only for folders:

%python
from dbutils import FileInfo
from typing import List

def discover_size2(path: str, verbose: bool = True):
  def loop_path(path: str):
    accum_size = 0.0
    path_list = dbutils.fs.ls(path)
    if path_list:
      for path_object in path_list:
        if path_object.size > 0:
          if verbose:
            print(f"{path_object.path}: {path_object.size / 1e6} MB")
          accum_size += path_object.size / 1e6
        else:
          # Folder: recursive discovery
          accum_size += loop_path(path_object.path)
    return accum_size

  return loop_path(path)

Answered By: robin loche

Answer 4

Love the answer by Emer!

Small addition:

If you met

"ModuleNotFoundError: No module named ‘dbutils’"

try this from dbruntime.dbutils instead of from dbutils. It works for me!

— Shizheng

Answered By: shizzhan

Answer 5

For anyone still hitting recursion limits with @robin loche’s approach, here is a purely iterative answer:

# from dbutils import FileInfo # Not required in databricks
# from dbruntime.dbutils import FileInfo # may work for some people

def get_size_of_path(path):
    return sum([file.size for file in get_all_files_in_path(path)])

def get_all_files_in_path(path, verbose=False):
    nodes_new = []

    nodes_new = dbutils.fs.ls(path)
    files = []

    while len(nodes_new) > 0:
        current_nodes = nodes_new
        nodes_new = []
        for node in current_nodes:
            if verbose:
                print(f"Processing {node.path}")
            children = dbutils.fs.ls(node.path)
            for child in children:
                if child.size == 0 and child.path != node.path:
                    nodes_new.append(child)
                elif child.path != node.path:
                    files.append(child)
    return files

path = "s3://some/path/"

print(f"Size of {path} in gb: {get_size_of_path(path) / 1024 / 1024 / 1024}")

Answered By: Jeremy

How to calculate a Directory size in ADLS using PySpark?

Question:

Answers: