Prismarine is a Pythonic ORM for DynamoDB, designed to simplify interactions with DynamoDB by providing a structured and Python-friendly interface. It leverages Python's type hinting and decorators to define models, which are then used to generate client code for database operations.
Key features include:
- Model Definition: Models are defined using Python's
TypedDict(default) or, optionally,pydantic.BaseModelclasses and are decorated with theCluster.modeldecorator to specify primary and sort keys. - Automatic Client Generation: The
prismarine_client.pyfile is auto-generated, containing classes and methods for interacting with DynamoDB tables based on the defined models. - Easy Integration: The generated client code integrates seamlessly with existing Python applications, providing methods for common database operations.
Prismarine aims to streamline the development process by reducing boilerplate code and ensuring that database interactions are type-safe and maintainable.
Prismarine works best with EasySAM.
pip install prismarine<base-path>/
<package-name>/
- models.py
- db.py
- prismarine_client.py // Auto-generated
Models are defined in the models.py file. Each model is a TypedDict, decorated with the Cluster.model decorator. You can also opt into Pydantic models (see Using Pydantic Models).
The Cluster class is used to group extension models together. It also sets a prefix for the table names.
from typing import TypedDict, NotRequired
from prismarine import Cluster
c = Cluster('TapgameExample')
@c.model(PK='Foo', SK='Bar')
class Team(TypedDict):
Foo: str
Bar: str
Baz: NotRequired[str]If we place this code in <base-path>/<package-name>/models.py and the following command is run, it will generate a prismarine_client.py file in the same directory:
prismarine generate-client --base <base-path> <package-name>The prismarine_client.py file will contain the following code:
class TeamModel(Model):
table_name = 'TapgameExampleTeam'
PK = 'Foo'
SK = 'Bar'
class UpdateDTO(TypedDict, total=False):
Foo: str
Bar: str
Baz: NotRequired[str]
@staticmethod
def list(*, foo: str) -> List[Team]:
...
@staticmethod
def get(*, bar: str, foo: str, default: Team | EllipsisType = ...) -> Team:
...
@staticmethod
def put(team: Team) -> Team:
...
@staticmethod
def update(
team: UpdateDTO, *, foo: str, bar: str, default: Team | EllipsisType = ...
) -> Team:
...
@staticmethod
def save(updated: Team, *, original: Team | None = None) -> Team:
...
@staticmethod
def delete(*, bar: str, foo: str):
...
@staticmethod
def scan() -> List[Team]:
...As you can see, the TeamModel class has static methods for all the CRUD operations. The UpdateDTO class is similar to the Team class, but all fields are optional.
Now, let's create a db.py file in the same directory:
import example.prismarine_client as pc
class TeamModel(pc.TeamModel):
passAlthough you can import and use prismarine_client.py directly, it is recommended to create a db.py file that imports the generated client and extends it with your own methods.
You can now use the TeamModel class in your code:
from sam.common.example.db import TeamModel
from sam.common.prismarine import DbNotFound
# Create a new team
new_team = TeamModel.put({'Foo': 'foo', 'Bar': 'bar', 'Baz': 'baz'})
# List teams by a primary key
teams_by_foo = TeamModel.list(foo='foo')
# Get a team
try:
team = TeamModel.get(foo='foo', bar='bar')
except DbNotFound:
print('Team not found')
# Update a team
updated_team = TeamModel.update(
{'Baz': 'new_baz'},
foo='foo',
bar='bar'
)
# List all teams
all_teams = TeamModel.scan()
# Delete a team
TeamModel.delete(foo='foo', bar='bar')You may notice that Prismarine mostly requires named arguments. This ensures that changes to field names do not cause silent code failures. For example, if the Sort Key name is changed, all usages of get and update methods will break and be highlighted by the IDE and linter. This approach also makes the code more readable.
Prismarine can optionally generate clients that work with pydantic.BaseModel schemas rather than TypedDict.
- Install the optional dependency:
pip install "prismarine[pydantic]"- Define your models as
BaseModelsubclasses inmodels.py. - Run the generator with the Pydantic model library enabled:
prismarine generate-client --model-library pydantic --base <base-path> <package-name>With this flag disabled (the default typed-dict mode), Prismarine behaves exactly as before. The Pydantic mode keeps the same API surface but returns/accepts BaseModel instances and automatically converts data during CRUD operations.
The Cluster.model decorator accepts several arguments to customize the model:
PK(required): The name of the partition key attributeSK(optional): The name of the sort key attributetable(optional): Sets a full custom table name (without prefix)name(optional): Sets a custom model name (used with prefix)trigger(optional): Configures a DynamoDB stream trigger for the table (when using with EasySAM)ttl(optional): Configures a DynamoDB Time To Live (TTL) attribute for the table (when using with EasySAM)
For example, if the Cluster has a prefix TapgameExample, by default the Team model will have the table name TapgameExampleTeam. If we set name='Custom', the table name will be TapgameExampleCustom. And if we set table='CustomTable', the table name will simply be CustomTable, without the prefix.
When using Prismarine with EasySAM, you can configure DynamoDB stream triggers directly on your models using the trigger parameter. This allows a Lambda function to be automatically invoked whenever items in the table are inserted, modified, or removed.
Simple trigger (string format):
@c.model(PK='Foo', SK='Bar', trigger='itemlogger')
class Item(TypedDict):
Foo: str
Bar: strAdvanced form (with options):
@c.model(
PK='Foo',
SK='Bar',
trigger={
'function': 'my-lambda',
'viewtype': 'new-and-old', # Optional: keys-only, new, old, new-and-old (default: new-and-old)
'batchsize': 10, # Optional: number of records per batch
'batchwindow': 5, # Optional: seconds to wait for batch
'startingposition': 'latest' # Optional: trim-horizon, latest (default: latest)
}
)
class Item(TypedDict):
Foo: str
Bar: strThe trigger configuration options:
- function: The name of the Lambda function to trigger
- viewtype: What data to include in the stream record (default:
new-and-old)keys-only: Only the keys of the modified itemnew: Only the new item imageold: Only the old item imagenew-and-old: Both old and new item images
- batchsize: Number of records to process per batch (improves throughput)
- batchwindow: Maximum number of seconds to wait for a batch (reduces latency)
- startingposition: Where to start reading the stream (default:
latest)trim-horizon: Start from the oldest record availablelatest: Start from the most recent record
When EasySAM generates the CloudFormation template, it will automatically:
- Enable DynamoDB Streams on the table
- Create an EventSourceMapping to connect the stream to your Lambda function
- Configure the appropriate IAM permissions for stream access
The trigger Lambda function will receive DynamoDB stream events with information about inserted, modified, or removed items.
When using Prismarine with EasySAM, you can configure DynamoDB Time To Live (TTL) directly on your models using the ttl parameter. This allows DynamoDB to automatically delete items after a specified expiration time.
Example:
from typing import TypedDict, NotRequired
from prismarine.runtime import Cluster
c = Cluster('PrismaTTL')
@c.model(PK='Foo', SK='Bar', ttl='ExpireAt')
class Item(TypedDict):
Foo: str
Bar: str
Baz: NotRequired[str]
ExpireAt: int # Unix timestamp (seconds since epoch)The ttl parameter specifies the attribute name that will store the expiration timestamp. When you create or update an item, set this attribute to a Unix timestamp (number of seconds since epoch). DynamoDB will automatically delete items within 48 hours after the TTL timestamp has passed.
Benefits:
- Automatic Cleanup: Items are automatically deleted without additional code
- Cost Effective: TTL deletion is free and doesn't consume write capacity units
- Declarative: Define TTL directly in your model configuration
When EasySAM generates the CloudFormation template, it will automatically:
- Enable TTL on the DynamoDB table
- Configure the
TimeToLiveSpecificationwith the specified attribute name
index decorators must be used before the model decorator.
The Cluster.index decorator is used to define a secondary index. It accepts PK, SK, and index arguments.
@c.index(index='by-bar', PK='Bar', SK='Foo')
@c.model(PK='Foo', SK='Bar')
class Team(TypedDict):
Foo: str
Bar: str
Baz: NotRequired[str]This will add a subclass ByBar to the TeamModel class:
class TeamModel(Model):
...
class ByBar:
PK = 'Bar'
SK = 'Foo'
@staticmethod
def list(
*,
bar: str,
limit: int | None = None,
direction: Literal['ASC', 'DESC'] = 'ASC'
) -> List[Team]:
...
@staticmethod
def get(*, bar: str, foo: str) -> Team:
...The Cluster.export decorator is used to define a class that is not a model, but is exported from the cluster. It accepts a class as an argument. It is required to used on all classes that serve as types for model elements.
@c.export
class Team(TypedDict):
Foo: str
Bar: strPrints the version of Prismarine.
prismarine version