Copy Into : Pattern for sub-folders

[Anup]

While trying to ingest data from the S3 bucket, we are running into a situation where the data in s3 buckets is in sub-folders of multiple depths.

Is there a good way of specifying patterns for the above case?

We tried using the following for a depth of 4, and it works.

%sql

COPY INTO table_name

FROM 's3://bucket_name'

FILEFORMAT = BINARYFILE

PATTERN = '/*/*/*/*'

FORMAT_OPTIONS ('mergeSchema' = 'true',

'header' = 'true')

COPY_OPTIONS ('mergeSchema' = 'true');

However, it is not always possible to know the exact depth for a large amount of random data.

Has anyone run into this problem and was able to solve it?

[Kaniz_Fatma]

Hi @Anup, When dealing with data in S3 buckets that are organized into sub-folders of varying depths, specifying patterns can be challenging.

However, there are some approaches you can consider:

Wildcard Patterns:

• You’ve already used a wildcard pattern in your example: '/*/*/*/*'. This works well when you know the exact depth, but as you mentioned, it’s not always feasible.
• If you have a rough idea of the maximum depth, you can use a wildcard pattern like '/*/*/*/*/*' to cover a broader range of depths.

Recursive Wildcard Patterns:

• Unfortunately, S3 doesn’t support recursive wildcard patterns directly. However, you can use tools like AWS Glue or AWS Lambda to recursively traverse the sub-folders and discover files.
• For example, in PySpark, you can read all Parquet files below a target folder using a pattern like t....

Programmatic Approaches:

• Write custom scripts or use AWS SDKs (such as Boto3 for Python) to programmatically explore the bucket structure and dynamically determine the depth.
• You can recursively list objects in a bucket and process them accordingly.

Flat File Structure:

• Remember that S3 itself has a flat structure with no inherent hierarchy like a typical file system. It uses key name prefixes for objects.
• While the Amazon S3 console supports the concept of folders for organizational purposes, these are not actual folders but rather key prefixes.
• Consider organizing your data in a way that minimizes the need for deep nesting.

Bucket Design:

• Decide whether to use multiple buckets or a single bucket with sub-folders.
• There’s no performance difference between using a single bucket or multiple buckets, so both options....
• If you choose sub-folders, ensure consistent naming conventions to make pattern matching easier.

Service Limit Increase:

• By default, you can create up to 100 buckets per AWS account. If you need more, you can request a service limit increase.
• However, there’s a hard upper limit of 1000 buckets per account that cannot be exceeded.

Remember that flexibility and scalability are essential when dealing with large amounts of data. Experiment with different approaches to find the one that best fits your specific use case. 🚀

[Anup · Accepted solution]

Thanks @Kaniz_Fatma ,

We ended up implementing the programmatic approach to calculate the depth (using boto3).