s3migrate

Bulk delete/copy/move files or modify Hive/Drill/Athena partitions using pythonic pattern matching

Example

Imagine we have a dataset as follows:

s3://bucket/training_data/2019-01-01/part1.parquet 
s3://bucket/validation_data/2019-06-01/part13.parquet
... 

To make this dataset Hive-friendly, we want to includ explicit key-value pairs in the paths, e.g.:

s3://bucket/data/split=training/execution_date=2019-01-01/part1.parquet
s3://bucket/data/split=training/execution_date=2019-06-01/part13.parquet
...

This can be achieved using the s3migrate.mv (aka move) command with intutitive pattern matching:

old_path = "s3://bucket/{split}_data/{execution_date}/{filename}"
new_path = "s3://bucket/data/split={split}/execution_date={execution_date}/{filename}"
s3migrate.mv(
    from=old_path,
    to=new_path,
    dryrun=False
)

If instead we want to delete all files matching old_path pattern, we can use s3migrate.rm:

s3migrate.rm(
    from=old_path,
    dryrun=False
)

Supported commands

File-system-like operations

The module provides the following commands:

command	number of patterns	action
cp/copy	2	copy (duplicate) all matched files to new location
mv/move	2	move (rename) all matched files
rm/remove	1	remove all matched files

Eeach takes one or two patterns, as well as the dryrun argument.

NB when two patterns are provided, both must contain the same set of keys

General-purpose generators

command	usecase
iter	iterate over all matching filenames, e.g. to read each file
iterformats	iterate over all matched format dictionaries, e.g. to collect all Hive key values

s3migrate.iter(pattern) will yield file names filename matching pattern. This allows custom file processing logic downstream.

s3migrate.iterformats(pattern) will instead yield dictionaries fmt_dict such that pattarn.format(**fmt_dict) is equivalent to the matched filename.

Dry run mode

Dry run mode allows testing your patterns without performing any destructive operations.

With dryrun=True (default), information about operations to be performed is logged at INFO and DEBUG level - make sure to set your logging accordingly, e.g. inside a Jupyter Notebook:

import logging

logger = logging.getLogger()
logger.setLevel(logging.DEBUG)
logger.handlers = [logging.StreamHandler()]