abnosql 0.0.24

NoSQL Abstraction Library

NoSQL Abstraction Library

Basic CRUD and query support for NoSQL databases, allowing for portable cloud native applications

• AWS DynamoDB
• Azure Cosmos NoSQL
• Google Firestore

This library is not intended to create databases/tables, use Terraform/ARM/CloudFormation etc for that

Why not just use the name 'nosql' or 'pynosql'? because they already exist on pypi :-)

• NoSQL Abstraction Library
  • Installation
• Usage
  • API Docs
  • Querying
  • Indexes
  • Updates
  • Existence Checking
  • Schema Validation
  • Partition Keys
  • Pagination
  • Audit
  • Change Feed / Stream Support
  • Client Side Encryption
• Configuration
  • AWS DynamoDB
  • Azure Cosmos NoSQL
  • Google Firestore
• Plugins and Hooks
• Testing
  • AWS DynamoDB
  • Azure Cosmos NoSQL
  • Google Firestore
• CLI
• Future Enhancements / Ideas

Installation

pip install 'abnosql[dynamodb]'
pip install 'abnosql[cosmos]'
pip install 'abnosql[firestore]'

For optional client side field level envelope encryption

pip install 'abnosql[aws-kms]'
pip install 'abnosql[azure-kms]'
pip install 'abnosql[gcp-kms]'

By default, abnosql does not include database dependencies. This is to facilitate packaging abnosql into AWS Lambda or Azure Functions (for example), without over-bloating the packages

Usage

from abnosql import table
import os

os.environ['ABNOSQL_DB'] = 'dynamodb'
os.environ['ABNOSQL_KEY_ATTRS'] = 'hk,rk'

item = {
    'hk': '1',
    'rk': 'a',
    'num': 5,
    'obj': {
        'foo': 'bar',
        'num': 5,
        'list': [1, 2, 3],
    },
    'list': [1, 2, 3],
    'str': 'str'
}

tb = table('mytable')

# create/replace
tb.put_item(item)

# update - using ABNOSQL_KEY_ATTRS
updated_item = tb.put_item(
    {'hk': '1', 'rk': 'a', 'str': 'STR'},
    update=True
)
assert updated_item['str'] == 'STR'

# bulk
tb.put_items([item])

# note partition/hash key should be first kwarg
assert tb.get_item(hk='1', rk='a') == item

assert tb.query({'hk': '1'})['items'] == [item]

# scan
assert tb.query()['items'] == [item]

# be careful not to use cloud specific statements!
assert tb.query_sql(
    'SELECT * FROM mytable WHERE mytable.hk = @hk AND mytable.num > @num',
    {'@hk': '1', '@num': 4}
)['items'] == [item]

tb.delete_item({'hk': '1', 'rk': 'a'})

API Docs

See API Docs

Querying

query() performs DynamoDB Query using KeyConditionExpression (if key supplied) and exact match on FilterExpression if filters are supplied. For Cosmos, SQL is generated. This is the safest/most cloud agnostic way to query and probably OK for most use cases.

query_sql() performs Dynamodb ExecuteStatement passing in the supplied PartiQL statement. Cosmos uses the NoSQL SELECT syntax.

During mocked tests, SQLGlot is used to execute the statement, so results may differ...

Care should be taken with query_sql() to not to use SQL features that are specific to any specific provider (breaking the abstraction capability of using abnosql in the first place)

The Firestore plugin uses sqlglot to parse simple SQL statements (eg AND only supported)

Indexes

Beyond partition and range keys defined on the table, indexes currently have limited support within abnosql

• The DynamoDB implemention of query() allows a secondary index to be specified via optional index kwarg
• Cosmos has Range, Spatial and Composite indexes, however the abnosql library does not do anything yet with index kwarg in query() implementation.

Updates

put_item() and put_items() support update boolean attribute, which if supplied will do an update_item() on DynamoDB, and a patch_item() on Cosmos. For this to work however, you must specify the key attribute names, either via ABNOSQL_KEY_ATTRS env var as a comma separated list (eg perhaps multiple tables all share common partition/range key scheme), or as the key_attrs config item when instantiating the table, eg:

tb = table('mytable', {'key_attrs': ['hk', 'rk']})

If you don't need to do any updates and only need to do create/replace, then these key attribute names do not need to be supplied

All items being updated must actually exist first, or else exception raised

Firestore does not return updated item, so if this is required use put_get = True config variable

Existence Checking

If check_exists config attribute is True, then CRUD operations will raise exceptions as follows:

• get_item() raises NotFoundException if item doesnt exist
• put_item() raises ExistsException if item already exists
• put_item(update=True) raises NotFoundException if item doesnt exist to update
• delete_item() raises NotFoundException if item doesnt exist

This adds some delay overhead as abnosql must check if item exists

This can also be enabled by setting environment variable ABNOSQL_CHECK_EXISTS=TRUE

If for some reason you need to override this behaviour once enabled for put_item() create operation, you can pass abnosql_check_exists=False into the item (this gets popped out so not persisten), which will allow create operation to overwrite the existing item without throwing ExistsException

Schema Validation

config can define jsonschema to validate upon create or update operations (via put_item())

Combination of the following config attributes supported

• schema : jsonschema dict or yaml string, applied to both create and update
• create_schema : jsonschema dict/yaml only on create
• update_schema : jsonschema dict/yaml only on update
• schema_errmsg : override default error message on both create and update
• create_schema_errmsg : override default error message on create
• update_schema_errmsg : override default error message on update

You can get details of validation errors through e.to_problem() or e.detail

NOTE: key_attrs required when updating (see Updates)

Partition Keys

A few methods such as get_item(), delete_item() and query() need to know partition/hash keys as defined on the table. To avoid having to configure this or lookup from the provider, the convention used is that the first kwarg or dictionary item is the partition key, and if supplied the 2nd is the range/sort key.

Pagination

query and query_sql accept limit and next optional kwargs and return next in response. Use these to paginate.

This works for AWS DyanmoDB & Firestore, however Azure Cosmos has a limitation with continuation token for cross partitions queries (see Python SDK documentation). For Cosmos, abnosql appends OFFSET and LIMIT in the SQL statement if not already present, and returns next. limit is defaulted to 100. See the tests for examples

Audit

Table config attribute audit_user will add the following to the item being written to database:

• createdBy - value of audit_user, added if does not exist in item supplied to put_item()
• createdDate - UTC ISO timestamp string, added if does not exist
• modifiedBy - value of audit_user always added
• modifiedDate - UTC ISO timestamp string, always added

If snake_case over CamelCase is preferred, set env var ABNOSQL_CAMELCASE = FALSE

NOTE: created* will only be added if update is not True in a put_item() operation

Table config attribute audit_callback with value as a function callback can be used to hook into additional audit stores.

Callback function must accept the following positional args:

• table_name - table name
• dt_iso - ISO date timestamp
• operation - create, update, get or delete
• key - key of item serialised in =; format
• audit_user - user performing the operation

Change Feed / Stream Support

AWS DynamoDB Streams allow Lambda functions to be triggered upon create, update and delete table operations. The event sent to the lambda (see aws docs) contains eventName and eventSourceARN, where:

• eventName - name of event, eg INSERT, MODIFY or REMOVE (see here)
• eventSourceARN - ARN of the table name

This allows a single stream processor lambda to process events from multiple tables (eg for writing into ElasticSearch)

Like DynamoDB, Azure CosmosDB supports change feeds, however the event sent to the function (currently) omits the event source (table name) and only delete event names are available if a preview change feed mode is enabled, which needs explicit enablement for.

Because both the eventName and eventSource are ideally needed (irrespective of preview mode or not), abnosql library automatically adds the changeMetadata to an item during create, update and delete, eg:

item = {
    "hk": "1",
    "rk": "a",
    "changeMetadata": {
        "eventName": "INSERT",
        "eventSource": "sometable"
    }
}

Because no REMOVE event is sent at all without preview change feed mode above - abnosql must first update the item, and then delete it. This is also needed for the eventSource / table name to be captured in the event, so unfortunately until Cosmos supports both attributes, update is needed before a delete. 5 second synchronous sleep is added by default between update and delete to allow CosmosDB to send the update event (0 seconds results in no update event). This can be controlled with ABNOSQL_COSMOS_CHANGE_META_SLEEPSECS env var (defaults to 5 seconds), and disabled by setting to 0

This behaviour is enabled by default, however can be disabled by setting ABNOSQL_COSMOS_CHANGE_META env var to FALSE or cosmos_change_meta=False in table config. ABNOSQL_CAMELCASE = FALSE env var can also be used to change attribute names used to snake_case if needed

To write an Azure Function / AWS Lambda that is able to process both DynamoDB and Cosmos events, look for changeMetadata first and if present use that otherwise look for eventName and eventSourceARN in the event payload assuming its DynamoDB

Google Firestore should support triggering functions similar to DynamoDB Streams, so changeMetadata is not required

Client Side Encryption

If configured in table config with kms attribute, abnosql will perform client side encryption using AWS KMS, Azure KeyVault or Google KMS

Each attribute value defined in the config is encrypted with a 256-bit AES-GCM data key generated for each attribute value:

• aws uses AWS Encryption SDK for Python
• azure uses python cryptography to generate AES-GCM data key, encrypt the attribute value and then uses an RSA CMK in Azure Keyvault to wrap/unwrap (envelope encryption) the AES-GCM data key. The plugin uses the azure-keyvault-keys python SDK for wrap/unrap functionality of the generated data key (Azure doesnt support generate data key as AWS does - see also tink issue)
• gcp uses Google Tink

All providers use a 256-bit AES-GCM generated data key with AAD/encryption context (Azure provider uses a 96-nonce). AES-GCM is an Authenticated symmetric encryption scheme used by AWS, Azure & Google (and Hashicorp Vault)

See also AWS Encryption Best Practices

Example config:

{
    'kms': {
        # Azure example
        'key_ids': ['https://foo.vault.azure.net/keys/bar/45e36a1024a04062bd489db0d9004d09'],

        # AWS example
        # 'key_ids': ['arn:aws:kms:us-west-2:111122223333:key/1234abcd-12ab-34cd-56ef-1234567890ab'],

        # Google Example
        # 'key_ids': ['gcp-kms://projects/p1/locations/global/keyRings/kr1/cryptoKeys/ck1'],

        'key_attrs': ['hk', 'rk'],
        'attrs': ['obj', 'str']
    }
}

Where:

• key_ids: list of AWS KMS Key ARNs, Azure KeyVault identifier (URL to RSA CMK) or Google KMS URI. This is picked up via ABNOSQL_KMS_KEYS env var as a comma separated list (NOTE: env var recommended to avoid provider specific code)
• key_attrs: list of key attributes in the item from which the AAD/encryption context is set. Taken from ABNOSQL_KEY_ATTRS env var or table key_attrs if defined there
• attrs: list of attributes keys to encrypt
• key_bytes: optional for azure, use your own AESGCM key if specified, otherwise generate one

If kms config attribute is present, abnosql will look for the ABNOSQL_KMS provider to load the appropriate provider KMS module (eg "aws" or "azure"), and if not present use default depending on the database (eg cosmos will use azure, dynamodb will use aws)

In example above, the key_attrs ['hk', 'rk'] are used to define the encryption context / AAD used, and attrs ['obj', 'str'] what attributes to encrypt/decrypt

With an item:

{
    'hk': '1',
    'rk': 'b',
    'obj': {'foo':'bar'},
    'str': 'foobar'
}

The encryption context / AAD is set to hk=1 and rk=b and obj and str values are encrypted

If you don't want to use any of these providers, then you can use put_item_pre and get_item_post hooks to perform your own client side encryption

See also AWS Multi-region encryption keys and set ABNOSQL_KMS_KEYS env var as comma list of ARNs

Configuration

It is recommended to use environment variables where possible to avoid provider specific application code

if ABNOSQL_DB env var is not set, abnosql will attempt to apply defaults based on available environment variables:

• AWS_DEFAULT_REGION - sets database to dynamodb (see aws docs)
• FUNCTIONS_WORKER_RUNTIME - sets database to cosmos (see azure docs)
• K_SERVICE - sets database to firestore (though this could also get confused if running on knative)

AWS DynamoDB

Set the following environment variable and use the usual AWS environment variables that boto3 uses

• ABNOSQL_DB = "dynamodb"

Or set the boto3 session in the config

from abnosql import table
import boto3

tb = table(
    'mytable',
    config={'session': boto3.Session()},
    database='dynamodb'
)

Azure Cosmos NoSQL

Set the following environment variables:

• ABNOSQL_DB = "cosmos"
• ABNOSQL_COSMOS_ACCOUNT = your database account
• ABNOSQL_COSMOS_ENDPOINT = drived from ABNOSQL_COSMOS_ACCOUNT if not set
• ABNOSQL_COSMOS_CREDENTIAL = your cosmos credential, use Azure Key Vault References if using Azure Functions. Don't set to use DefaultAzureCredential / managed identity.
• ABNOSQL_COSMOS_DATABASE = cosmos database

OR - use the connection string format:

• ABNOSQL_DB = "cosmos://account@credential:database" or "cosmos://account@:database" to use managed identity (credential could also be "DefaultAzureCredential")

Alternatively, define in config (though ideally you want to use env vars to avoid application / environment specific code).

from abnosql import table

tb = table(
    'mytable',
    config={'account': 'foo', 'database': 'bar'},
    database='cosmos'
)

Google Firestore

Set the following environment variables:

• ABNOSQL_DB = "firestore"
• ABNOSQL_FIRESTORE_PROJECT or GOOGLE_CLOUD_PROJECT = google cloud project
• ABNOSQL_FIRESTORE_DATABASE = Firestore database
• ABNOSQL_FIRESTORE_CREDENTIALS = oauth, optional - if using google CLI, its also picked up from ~/.config/gcloud/application_default_credentials.json if found

OR - use the connection string format:

• ABNOSQL_DB = "firestore://project@credential:database"

Alternatively, define in config (though ideally you want to use env vars to avoid application / environment specific code).

from abnosql import table

tb = table(
    'mytable',
    config={'project': 'foo', 'database': 'bar'},
    database='firestore'
)

See also https://cloud.google.com/firestore/docs/authentication

Plugins and Hooks

abnosql uses pluggy and registers in the abnosql.table namespace

The following hooks are available

• set_config - set config
• get_item_pre
• get_item_post - called after get_item(), can return modified data
• put_item_pre
• put_item_post
• put_items_post
• delete_item_post

See the TableSpecs and example test_hooks()

Testing

AWS DynamoDB

Use moto package and abnosql.mocks.mock_dynamodbx

mock_dynamodbx is used for query_sql and only needed if/until moto provides full partiql support

Example:

from abnosql.mocks import mock_dynamodbx 
from moto import mock_dynamodb

@mock_dynamodb
@mock_dynamodbx  # needed for query_sql only
def test_something():
    ...

More examples in tests/test_dynamodb.py

Azure Cosmos NoSQL

Use requests package and abnosql.mocks.mock_cosmos

Example:

from abnosql.mocks import mock_cosmos
import requests

@mock_cosmos
@responses.activate
def test_something():
    ...

More examples in tests/test_cosmos.py

Google Firestore

Use python-mock-firestore and pass MockFirestore() to table config as client attribute, or patch get_client()

Example:

from unittest.mock import patch
from mockfirestore import MockFirestore
from abnosql.plugins.table.firestore import Table as FirestoreTable


@patch.object(FirestoreTable, 'get_client', MockFirestore)
def test_something():
    tb = table('mytable', {})
    item = tb.get_item(foo='bar')

More examples in tests/test_firestore.py

CLI

Small abnosql CLI installed with few of the commands above

Usage: abnosql [OPTIONS] COMMAND [ARGS]...

Options:
  --help  Show this message and exit.

Commands:
  delete-item
  get-item
  put-item
  put-items
  query
  query-sql

To install dependencies

pip install 'abnosql[cli]'

Example querying table in Azure Cosmos, with cosmos.json config file containing endpoint, credential and database

$ abnosql query-sql mytable 'SELECT * FROM mytable' -d cosmos -c cosmos.json
partkey      id      num  obj                                          list       str
-----------  ----  -----  -------------------------------------------  ---------  -----
p1           p1.1      5  {'foo': 'bar', 'num': 5, 'list': [1, 2, 3]}  [1, 2, 3]  str
p2           p2.1      5  {'foo': 'bar', 'num': 5, 'list': [1, 2, 3]}  [1, 2, 3]  str
p2           p2.2      5  {'foo': 'bar', 'num': 5, 'list': [1, 2, 3]}  [1, 2, 3]  str

Future Enhancements / Ideas

• client side encryption
• test pagination & exception handling
• Google Firestore support, ideally in the core library (though could be added outside via use of the plugin system). Would need something like FireSQL implemented for python, maybe via sqlglot
• Google Vault KMS support
• Hashicorp Vault KMS support
• Simple caching (maybe) using globals (used for AWS Lambda / Azure Functions)
• PostgresSQL support using JSONB column (see here for example). Would be nice to avoid an ORM and having to define a model for each table...
• blob storage backend? could use something similar to NoDB but maybe combined with smart_open and DuckDB's Hive Partitioning
• Redis..
• Hook implementations to write to ElasticSearch / OpenSearch for better searching. Useful when not able to use AWS Stream Processors Azure Change Feed, or Elasticstore. Why? because not all databases support stream processing, and if they do you don't want the hastle of using CDC