pydantic-glue 0.7.0

Convert pydantic model to aws glue schema for terraform

JSON Schema to AWS Glue schema converter

• JSON Schema to AWS Glue schema converter
  • Installation
  • What?
  • Why?
  • Example
  • Override the type for the AWS Glue Schema
  • How it works?
  • Future work

Installation

pip install pydantic-glue

What?

Converts pydantic schemas to json schema and then to AWS glue schema, so in theory anything that can be converted to JSON Schema could also work.

Why?

When using AWS Kinesis Firehose in a configuration that receives JSONs and writes parquet files on S3, one needs to define a AWS Glue table so Firehose knows what schema to use when creating the parquet files.

AWS Glue lets you define a schema using Avro or JSON Schema and then to create a table from that schema, but as of May 2022 there are limitations on AWS that tables that are created that way can't be used with Kinesis Firehose.

https://stackoverflow.com/questions/68125501/invalid-schema-error-in-aws-glue-created-via-terraform

This is also confirmed by AWS support.

What one could do is create a table set the columns manually, but this means you now have two sources of truth to maintain.

This tool allows you to define a table in pydantic and generate a JSON with column types that can be used with terraform to create a Glue table.

Example

Take the following pydantic class

from pydantic import BaseModel
from typing import List


class Bar(BaseModel):
    name: str
    age: int


class Foo(BaseModel):
    nums: List[int]
    bars: List[Bar]
    other: str

Running pydantic-glue

pydantic-glue -f example.py -c Foo

you get this JSON in the terminal:

{
  "//": "Generated by pydantic-glue at 2022-05-25 12:35:55.333570. DO NOT EDIT",
  "columns": {
    "nums": "array<int>",
    "bars": "array<struct<name:string,age:int>>",
    "other": "string"
  }
}

and can be used in terraform like that

locals {
  columns = jsondecode(file("${path.module}/glue_schema.json")).columns
}

resource "aws_glue_catalog_table" "table" {
  name          = "table_name"
  database_name = "db_name"

  storage_descriptor {
    dynamic "columns" {
      for_each = local.columns

      content {
        name = columns.key
        type = columns.value
      }
    }
  }
}

Alternatively you can run CLI with -o flag to set output file location:

pydantic-glue -f example.py -c Foo -o example.json -l

If your Pydantic models use field aliases, but you prefer to display the field names in the JSON schema, you can enable this behavior by using the --schema-by-name flag.

Here you can find the details regarding pydantic aliases.

The following model will be converted differently with --schema-by-name argument.

from pydantic import BaseModel, Field

class A(BaseModel):
    hey: str = Field(alias="h")
    ho: str

pydantic-glue -f tests/data/input.py -c A

2025-02-01 00:08:45,046 - INFO - Generated file content:
{
  "//": "Generated by pydantic-glue at 2025-01-31 23:08:45.046012+00:00. DO NOT EDIT",
  "columns": {
    "h": "string",
    "ho": "string"
  }
}

 pydantic-glue -f tests/data/input.py -c A --schema-by-name
2025-02-01 00:09:18,381 - INFO - Generated file content:
{
  "//": "Generated by pydantic-glue at 2025-01-31 23:09:18.380586+00:00. DO NOT EDIT",
  "columns": {
    "hey": "string",
    "ho": "string"
  }
}

Override the type for the AWS Glue Schema

Wherever there is a type key in the input JSON Schema, an additional key glue_type may be defined to override the type that is used in the AWS Glue Schema. This is, for example, useful for a pydantic model that has a field of type int that is unix epoch time, while the column type you would like in Glue is of type timestamp.

Additional JSON Schema keys to a pydantic model can be added by using the Field function with the argument json_schema_extra like so:

from pydantic import BaseModel, Field

class A(BaseModel):
    epoch_time: int = Field(
        ...,
        json_schema_extra={
            "glue_type": "timestamp",
        },
    )

The resulting JSON Schema will be:

{
    "properties": {
        "epoch_time": {
            "glue_type": "timestamp",
            "title": "Epoch Time",
            "type": "integer"
        }
    },
    "required": [
        "epoch_time"
    ],
    "title": "A",
    "type": "object"
}

And the result after processing with pydantic-glue:

{
  "//": "Generated by pydantic-glue at 2022-05-25 12:35:55.333570. DO NOT EDIT",
  "columns": {
    "epoch_time": "timestamp",
  }
}

Recursing through object properties terminates when you supply a glue_type to use. If the type is complex, you must supply the full complex type yourself.

How it works?

• pydantic gets converted to JSON Schema
• the JSON Schema types get mapped to Glue types recursively

Future work

• Not all types are supported, I just add types as I need them, but adding types is very easy, feel free to open issues or send a PR if you stumbled upon a non-supported use case
• the tool could be easily extended to working with JSON Schema directly
• thus, anything that can be converted to a JSON Schema should also work.