Schema Path Pattern¶

Specifying a schema path pattern for your schema extraction job helps optimize the extraction. In large-scale analytics systems, or data lakes, data representing a single logical schema is stored as a multi-file resource set. A resource set is a logical group of multiple files which use exactly the same logical schema. Alation uses the schema path pattern to identify resource sets in your storage system containers or buckets.

For example, let’s assume we are storing data as Parquet files. The physical storage of these files looks like this:

"/path/to/sample.parquet/part-1.snappy.parquet",
"/path/to/sample.parquet/part-2.snappy.parquet",
<...>
"/path/to/sample.parquet/part-n.snappy.parquet"

As mentioned above, a resource set would have exactly the same type of files, so if we extract column metadata from each file, it will be the same for all these files. If there are hundreds or thousands of files under one directory, all with the same schema, it would not be efficient to perform column extraction for each file as this is a resource-intensive operation.

Using a schema path pattern in such cases helps discover a schema for the whole resource set, optimizing the amount of data read and streamed by the connector. Use scenarios below to write schema path patterns for your OCF file system sources that support schema extraction.

Scenario 1¶

Single logical schema with multi-file dataset:

container/warehouse/production-database/sales_data/sales_data_1.parquet
container/warehouse/production-database/sales_data/sales_data_2.parquet
<...>
container/warehouse/production-database/sales_data/sales_data_n.parquet

The text string that needs to be provided as input in the Schema Path Pattern field can be:

sales_data

production-database/sales_data

Detected Logical Schema¶

container/warehouse/production-database/sales_data—Columns for this logical schema will be cataloged on the catalog page for container/warehouse/production-database/sales_data folder by reading the columns for file sales_data1.parquet.

Scenario 2¶

Multiple logical schemas with corresponding multi-file datasets. All logical schemas at the same directory depth:

container/warehouse/production-database/sales_data/sales_data_1.parquet
container/warehouse/production-database/sales_data/sales_data_2.parquet
<...>
container/warehouse/production-database/sales_data/sales_data_n.parquet

container/warehouse/production-database/product_data/product_data_1.parquet
container/warehouse/production-database/product_data/product_data_2.parquet
<...>
container/warehouse/production-database/product_data/product_data_n.parquet

Expected behavior for this kind of scenario is to discover sales_data and product_data as two logical schemas.

The text string that needs to be provided as input in the Schema Path Pattern field can be:

production-database/.*

production-database/(.*)_data

production-database/(sales|product)_data

Detected Logical Schemas¶

container/warehouse/production-database/sales_data—Columns for this logical schema will be cataloged on the catalog page for container/warehouse/production-database/sales_data folder by reading the columns for file sales_data_1.parquet.

container/warehouse/production-database/product_data—Columns for this logical schema will be cataloged on the catalog page for container/warehouse/production-database/product_data folder by reading the columns for file product_data_1.parquet.

Scenario 3¶

Multiple logical schemas with corresponding multi-file datasets. Each unique directory at the deepest level for each multi-file dataset to be discovered as a logical schema.

container/warehouse/production-database/sales_data/sales_data_1.parquet
container/warehouse/production-database/sales_data/sales_data_2.parquet
<...>
container/warehouse/production-database/sales_data/sales_data_n.parquet

container/warehouse/production-database/product_data/product_data_1.parquet
container/warehouse/production-database/product_data/product_data_2.parquet
<...>
container/warehouse/production-database/product_data/product_data_n.parquet

container/warehouse/production-database/customer/customer_data/customer_data_1.parquet
container/warehouse/production-database/customer/customer_data/customer_data_1.parquet
<...>
container/warehouse/production-database/customer/customer_data/customer_data_n.parquet

Expected behavior for this kind of scenario is to discover sales_data, product_data and customer_data as logical schemas.

The text string that needs to be provided as input in the Schema Path Pattern field can be:

production-database/.*

Detected Logical Schemas¶

container/warehouse/production-database/sales_data—Columns for this logical schema will be cataloged on the catalog page for container/warehouse/production-database/sales_data folder by reading the columns for file sales_data_1.parquet.

container/warehouse/production-database/product_data—Columns for this logical schema will be cataloged on the catalog page for container/warehouse/production-database/product_data folder by reading the columns for file product_data_1.parquet.

container/warehouse/production-database/customer/customer_data—Columns for this logical schema will be cataloged on the catalog page for container/warehouse/production-database/customer/customer_data folder by reading the columns for file customer_data_1.parquet.

Scenario 4¶

Single logical schema with multi-file dataset, data divided by year/month/day.

container/warehouse/production-database/inventory/2021/01/01/data.parquet
container/warehouse/production-database/inventory/2021/01/02/data.parquet
<...>
container/warehouse/production-database/inventory/2021/01/31/data.parquet
container/warehouse/production-database/inventory/2021/02/01/data.parquet
container/warehouse/production-database/inventory/2021/02/02/data.parquet
<...>
container/warehouse/production-database/inventory/2021/02/28/data.parquet
<...>
container/warehouse/production-database/inventory/2021/12/31/data.parquet

container/warehouse/production-database/inventory/2022/01/01/data.parquet
container/warehouse/production-database/inventory/2022/01/02/data.parquet
<...>
container/warehouse/production-database/inventory/2022/01/31/data.parquet
container/warehouse/production-database/inventory/2022/02/01/data.parquet
container/warehouse/production-database/inventory/2022/02/02/data.parquet
<...>
container/warehouse/production-database/inventory/2022/02/28/data.parquet
<...>
container/warehouse/production-database/inventory/2022/12/31/data.parquet

The text string that needs to be provided as input in the Schema Path Pattern field can be:

production-database/inventory

production-database/[a-z0-9A-Z]*

Detected Logical Schemas¶

container/warehouse/production-database/inventory—Columns for this logical schema will be cataloged on the catalog page for container/warehouse/production-database/inventory folder by reading the columns for file container/warehouse/production-database/inventory/2021/01/01/data.parquet.

Scenario 5¶

Multiple logical schemas with multi-file dataset, data divided by year/month/day.

container/warehouse/production-database/inventory/2021/01/01/data.parquet
container/warehouse/production-database/inventory/2021/01/02/data.parquet
<...>
container/warehouse/production-database/inventory/2021/01/31/data.parquet
container/warehouse/production-database/inventory/2021/02/01/data.parquet
container/warehouse/production-database/inventory/2021/02/02/data.parquet
<...>
container/warehouse/production-database/inventory/2021/02/28/data.parquet
<...>
container/warehouse/production-database/inventory/2021/12/31/data.parquet

container/warehouse/production-database/inventory/2022/01/01/data.parquet
container/warehouse/production-database/inventory/2022/01/02/data.parquet
<...>
container/warehouse/production-database/inventory/2022/01/31/data.parquet
container/warehouse/production-database/inventory/2022/02/01/data.parquet
container/warehouse/production-database/inventory/2022/02/02/data.parquet
<...>
container/warehouse/production-database/order/2022/02/28/data.parquet
<...>
container/warehouse/production-database/order/2022/12/31/data.parquet

container/warehouse/production-database/order/2021/01/01/data.parquet
container/warehouse/production-database/order/2021/01/02/data.parquet
<...>
container/warehouse/production-database/order/2021/01/31/data.parquet
container/warehouse/production-database/order/2021/02/01/data.parquet
container/warehouse/production-database/order/2021/02/02/data.parquet
<...>
container/warehouse/production-database/order/2021/02/28/data.parquet
<...>
container/warehouse/production-database/order/2021/12/31/data.parquet

container/warehouse/production-database/order/2022/01/01/data.parquet
container/warehouse/production-database/order/2022/01/02/data.parquet
<...>
container/warehouse/production-database/order/2022/01/31/data.parquet
container/warehouse/production-database/order/2022/02/01/data.parquet
container/warehouse/production-database/order/2022/02/02/data.parquet
<...>
container/warehouse/production-database/order/2022/02/28/data.parquet
<...>
container/warehouse/production-database/order/2022/12/31/data.parquet

The text string that needs to be provided as input in the Schema Path Pattern field can be:

production-database/([a-z0-9A-Z]*)

Detected Logical Schemas¶

-container/warehouse/production-database/inventory—Columns for this logical schema will be cataloged on the catalog page for container/warehouse/production-database/inventory folder by reading the columns for file container/warehouse/production-database/inventory/2021/01/01/data.parquet.

-container/warehouse/production-database/order—Columns for this logical schema will be cataloged on the catalog page for container/warehouse/production-database/order folder by reading the columns for file container/warehouse/production-database/order/2022/02/28/data.parquet.

Scenario 6¶

Multiple logical schemas with multi-file dataset, each unique directory under a specific directory to be identified as logical schema.

container/folder/engineering/data/emp1_data/2022/01/1.parquet
container/folder/engineering/data/emp1_data/2022/02/1.parquet
container/folder/engineering/data/emp1_data/2022/03/1.parquet
container/folder/engineering/data/emp1_data/2022/04/1.parquet

container/folder/engineering/data/emp2_data/2022/01/1.parquet
container/folder/engineering/data/emp2_data/2022/02/1.parquet
container/folder/engineering/data/emp2_data/2022/03/1.parquet
container/folder/engineering/data/emp2_data/2022/04/1.parquet

The text string that needs to be provided as input in the Schema Path Pattern field can be:

engineering/data/([a-z_0-9]*)

Detected Logical Schemas¶

-container/folder/engineering/data/emp1_data—Columns for this logical schema will be cataloged on the catalog page for container/folder/engineering/data/emp1_data folder by reading the columns for file container/folder/engineering/data/emp1_data/2022/01/1.parquet.

-container/folder/engineering/data/emp2_data—Columns for this logical schema will be cataloged on the catalog page for container/folder/engineering/data/emp2_data folder by reading the columns for file container/folder/engineering/data/emp2_data/2022/01/1.parquet.

Scenario 7¶

Multiple logical schemas with multi-file dataset, complex path pattern.

container/warehouse/production-database/sales_data/sales_data_1.parquet
container/warehouse/production-database/sales_data/sales_data_2.parquet
<...>
container/warehouse/production-database/sales_data/sales_data_n.parquet

container/warehouse/production-database/product_data/product_data_1.parquet
container/warehouse/production-database/product_data/product_data_2.parquet
<...>
container/warehouse/production-database/product_data/product_data_n.parquet

container/warehouse/dev-database/sales_data/sales_data_1.parquet
container/warehouse/dev-database/sales_data/sales_data_2.parquet
<...>
container/warehouse/dev-database/sales_data/sales_data_n.parquet

container/warehouse/dev-database/product_data/product_data_1.parquet
container/warehouse/dev-database/product_data/product_data_2.parquet
<...>
container/warehouse/dev-database/product_data/product_data_n.parquet

The text string that needs to be provided as input in the Schema Path Pattern field can be:

(production|dev)-database/.*

production-database/(sales_data|product_data)|(dev-database/(sales|product)_data)

Detected Logical Schemas¶

-container/warehouse/production-database/sales_data -container/warehouse/production-database/product_data -container/warehouse/dev-database/sales_data -container/warehouse/dev-database/product_data

Scenario 8¶

Multiple logical schemas with multi-file dataset with hive style column partitioning.

container/warehouse/production-database/sales_data/city=IN/state=GJ/sales_data_1.parquet
container/warehouse/production-database/sales_data/city=IN/state=GJ/sales_data_2.parquet
container/warehouse/production-database/sales_data/city=IN/state=MH/sales_data_1.parquet
container/warehouse/production-database/sales_data/city=IN/state=MH/sales_data_2.parquet
<...>
container/warehouse/production-database/sales_data/city=XX/state=XX/sales_data_n.parquet

container/warehouse/production-database/product_data/year=2022/month=01/product_data_1.parquet
container/warehouse/production-database/product_data/year=2022/month=01/product_data_2.parquet
container/warehouse/production-database/product_data/year=2022/month=02/product_data_1.parquet
container/warehouse/production-database/product_data/year=2022/month=02/product_data_2.parquet
<...>
container/warehouse/production-database/product_data/year=XXX/month=XXX/product_data_n.parquet

The text string that needs to be provided as input in the Schema Path Pattern field can be:

production-database/(sales|product)_data

production-database/(.*)_data

production-database/[^=]*

Detected Logical Schemas¶

-container/warehouse/production-database/sales_data—Columns for this logical schema will be cataloged on the catalog page for container/warehouse/production-database/sales_data folder by reading the columns for file container/warehouse/production-database/sales_data/city=IN/state=GJ/sales_data1.parquet. Additionally, column partitions city and state will also be cataloged as columns.

-container/warehouse/production-database/product_data—Columns for this logical schema will be cataloged on the catalog page for container/warehouse/production-database/product_data folder by reading the columns for file container/warehouse/production-database/product_data/year=2022/month=01/product_data_1.parquet. Additionally, column partitions year and month will also be cataloged as columns.

Scenario 9¶

container/warehouse/production-database/inventory/spend/xyz/2021/01/01/data.parquet

container/warehouse/production-database/inventory/def/2021/01/01/data.parquet

container/warehouse/production-database/inventory1/2021/01/01/data.parquet

Expectation here is to discover spend, def and inventory1 as logical schemas; however, that is currently not supported. Within the discovered resource set, the first file (as per the order in the inventory report) with a supported file format is read for columns.

Example 1:

container/warehouse/production-database/sales_data/part1.parquet
container/warehouse/production-database/sales_data/part2.parquet
container/warehouse/production-database/sales_data/part3.parquet

part1.parquet will be read for columns.

Example 2:

container/warehouse/production-database/sales_data/part1.csv
container/warehouse/production-database/sales_data/part1.parquet
container/warehouse/production-database/sales_data/part2.csv
container/warehouse/production-database/sales_data/part1.parquet
container/warehouse/production-database/sales_data/part3.csv
container/warehouse/production-database/sales_data/part1.parquet

part1.csv will be read for columns.

Example 3:

container/warehouse/production-database/sales_data/part1.csv
container/warehouse/production-database/sales_data/part1.psv
container/warehouse/production-database/sales_data/part1.tsv
container/warehouse/production-database/sales_data/part3.parquet

part1.csv will be read for columns.

To do an offline validation of the pattern:

Prepare a set of sample file paths representative of real data:

container/folder/engineering/data/emp1_data/2022/01/1.parquet
container/folder/engineering/data/emp1_data/2022/02/1.parquet
container/folder/engineering/data/emp1_data/2022/03/1.parquet
container/folder/engineering/data/emp1_data/2022/04/1.parquet

container/folder/engineering/data/emp2_data/2022/01/1.parquet
container/folder/engineering/data/emp2_data/2022/02/1.parquet
container/folder/engineering/data/emp2_data/2022/03/1.parquet
container/folder/engineering/data/emp2_data/2022/04/1.parquet

Replace the placeholder Schema Path Pattern string in this regular expression:

(.(<PATTERN>))/((.).((?i)\bcsv\b|(?i)\btsv\b|(?i)\bpsv\b|(?i)\bparquet\b))

Example: (.(engineering/data/([a-z_0-9]*)))/((.).((?i)\bcsv\b|(?i)\btsv\b|(?i)\bpsv\b|(?i)\bparquet\b))

Use the inputs from point 1 and 2 on a website like regex101, selecting Java 8 as the FLAVOR. Group 1s will be discovered as logical schemas.

Schema Path Pattern¶

Scenario 1¶

Detected Logical Schema¶

Scenario 2¶

Detected Logical Schemas¶

Scenario 3¶

Detected Logical Schemas¶

Scenario 4¶

Detected Logical Schemas¶

Scenario 5¶

Detected Logical Schemas¶

Scenario 6¶

Detected Logical Schemas¶

Scenario 7¶

Detected Logical Schemas¶

Scenario 8¶

Detected Logical Schemas¶

Scenario 9¶

Alation Documentation PDF