Stores¶

Stores are the core implementations of the AsyncKeyValue protocol. They provide the actual storage backend for your key-value data.

Store Categories¶

Stores are organized into three categories based on their storage location and use case:

Local Stores: In-memory or on-disk storage local to the application
Secret Stores: Secure storage for sensitive data in OS secret stores
Distributed Stores: Network-based storage for multi-node applications

Stability Levels¶

Each store has a stability rating that indicates the likelihood of backwards-incompatible changes to how data is stored:

Stable: No planned backwards-incompatible changes
Preview: Unlikely to change in backwards-incompatible ways
Unstable: May change in backwards-incompatible ways
N/A: Not applicable (e.g., in-memory stores)

If you're using py-key-value for caching, stability may not matter. For long-term storage, prefer stable stores.

Local Stores¶

Local stores are stored in memory or on disk, local to the application.

Store	Stability	Description
Memory	N/A	Fast in-memory storage for development and caching
Disk	Stable	Persistent file-based storage in a single file
Disk (Per-Collection)	Stable	Persistent storage with separate files per collection
DuckDB	Unstable	In-process SQL OLAP database with native JSON storage
chDB	Unstable	Embedded ClickHouse engine with SQL access to stored data
FileTree	Stable	Directory-based storage with JSON files for visual inspection
Null (test)	N/A	No-op store for testing without side effects
RocksDB	Unstable	High-performance embedded database
Simple (test)	N/A	Simple in-memory store for testing
Windows Registry	Unstable	Windows Registry-based storage

MemoryStore¶

Fast in-memory storage ideal for development, testing, and caching.

from key_value.aio.stores.memory import MemoryStore

store = MemoryStore()

Installation:

pip install py-key-value-aio[memory]

Use Cases:

Development and testing
Fast caching
Session storage
Temporary data

Characteristics:

No persistence (data lost on restart)
Extremely fast
No external dependencies
Thread-safe

DiskStore¶

Persistent file-based storage using a single JSON file.

from key_value.aio.stores.disk import DiskStore

store = DiskStore(directory="./cache")

Installation:

pip install py-key-value-aio[disk]

Use Cases:

Local caching with persistence
Development environments
Single-node applications
Small datasets

Characteristics:

Persists across restarts
Simple file format (JSON)
Suitable for small to medium datasets
All data in one file

MultiDiskStore¶

Persistent storage with separate files per collection.

from key_value.aio.stores.multi_disk import MultiDiskStore

store = MultiDiskStore(directory="./cache")

Installation:

pip install py-key-value-aio[disk]

Use Cases:

Organizing data by collection
Better performance with many collections
Easier to manage individual collections

Characteristics:

One file per collection
Better suited for many collections
Easier collection management
JSON-based storage

FileTreeStore¶

Directory-based storage for visual inspection and debugging.

from key_value.aio.stores.filetree import FileTreeStore

store = FileTreeStore(directory="./debug-store")

Installation:

pip install py-key-value-aio[filetree]

Use Cases:

Visual inspection of store contents
Debugging store behavior
Development and testing
Understanding data structure

Characteristics:

Collections as directories
Keys as JSON files ({key}.json)
Human-readable filesystem layout
Easy to inspect and modify
NOT for production use

Directory Structure:

{base_directory}/
  {collection_1}/
    {key_1}.json
    {key_2}.json
  {collection_2}/
    {key_3}.json

Important Limitations:

Poor performance with many keys
No atomic operations
No automatic cleanup of expired entries
Filesystem path length constraints
Subject to filesystem limitations

When to Use:

Use FileTreeStore when you need to:

Visually inspect what's being stored
Debug complex data structures
Understand how the store organizes data
Manually modify stored data for testing

When NOT to Use:

Production environments
High-performance requirements
Large datasets
Concurrent access scenarios

DuckDBStore¶

In-process SQL OLAP database with native JSON storage.

from key_value.aio.stores.duckdb import DuckDBStore

# In-memory database (default)
store = DuckDBStore()

# Persistent database
store = DuckDBStore(database_path="./my_store.duckdb")

# With existing connection
import duckdb
connection = duckdb.connect("./my_store.duckdb")
store = DuckDBStore(connection=connection)

Installation:

pip install py-key-value-aio[duckdb]

Use Cases:

Analytics on stored key-value data
Local development with SQL querying capabilities
Applications needing both key-value and SQL access
Data exploration and reporting

Characteristics:

Native JSON column support for queryable values
In-memory or persistent storage
Standard SQL access to stored data
Efficient timestamp handling for TTL queries
Stable storage format: Unstable

Direct SQL Access:

DuckDB allows you to query your key-value data directly with SQL:

# After storing some data
await store.put("user:1", {"name": "Alice", "age": 30}, collection="users")
await store.put("user:2", {"name": "Bob", "age": 25}, collection="users")

# Query directly via DuckDB
result = store._connection.execute("""
    SELECT key, value->>'name' as name
    FROM kv_entries
    WHERE collection = 'users'
""").fetchall()

ChDBStore¶

Embedded ClickHouse engine (chDB) with SQL access to your key-value data.

from key_value.aio.stores.chdb import ChDBStore

# In-memory database (default)
store = ChDBStore()

# Persistent database (chDB stores data in a directory)
store = ChDBStore(database_path="./my_store.chdb")

# With an existing chDB session
from chdb.session import Session
session = Session("./my_store.chdb")
store = ChDBStore(session=session)

Installation:

pip install py-key-value-aio[chdb]

Use Cases:

Analytics on stored key-value data with full ClickHouse SQL
Embedded OLAP workloads alongside cache-style usage
Applications needing both key-value access and ClickHouse functions

Characteristics:

Embedded ClickHouse — no external server required
In-memory or persistent storage (single directory)
Process-global singleton: only one database_path active per process at a time
Full ClickHouse SQL on stored data
ReplacingMergeTree-backed schema (latest write wins via FINAL)
Synchronous I/O (chDB has no async API; operations block the event loop)
Expired entries are not automatically culled — see class docstring for cleanup guidance
Stable storage format: Unstable

RocksDBStore¶

High-performance embedded database using RocksDB.

from key_value.aio.stores.rocksdb import RocksDBStore

store = RocksDBStore(path="./rocksdb")

Installation:

pip install py-key-value-aio[rocksdb]

Use Cases:

High-throughput applications
Large datasets
Performance-critical applications

Characteristics:

Very fast reads and writes
Efficient storage
Requires native dependencies
Stable storage format: Unstable

WindowsRegistryStore¶

Storage using the Windows Registry.

from key_value.aio.stores.registry import WindowsRegistryStore

store = WindowsRegistryStore(
    hive="HKEY_CURRENT_USER",
    registry_path="Software\\py-key-value"
)

Installation:

pip install py-key-value-aio[registry]

Use Cases:

Windows-specific applications
System configuration storage
Integration with Windows settings

Characteristics:

Windows-only
Persists in registry
Subject to registry limitations
Stable storage format: Unstable

NullStore¶

No-op store that discards all data. Useful for testing.

from key_value.aio.stores.null import NullStore

store = NullStore()

Use Cases:

Testing without side effects
Disabling storage temporarily
Performance baseline testing

SimpleStore¶

Simple in-memory store for testing.

from key_value.aio.stores.simple import SimpleStore

store = SimpleStore()

Use Cases:

Basic testing
Minimal implementation reference

Secret Stores¶

Secret stores provide secure storage for sensitive data, typically using operating system secret management facilities.

Store	Stability	Description
Keyring	Stable	OS-level secure storage (Keychain, Credential Manager, etc.)
Vault	Unstable	HashiCorp Vault integration for enterprise secrets

KeyringStore¶

Secure storage using the operating system's keyring (macOS Keychain, Windows Credential Manager, Linux Secret Service).

from key_value.aio.stores.keyring import KeyringStore

store = KeyringStore(service_name="py-key-value")

Installation:

pip install py-key-value-aio[keyring]

Use Cases:

Storing API keys and tokens
User credentials
Sensitive configuration
Encrypted local storage

Characteristics:

OS-level encryption
Secure by default
Cross-platform
Windows limitation: Strict value length limits

Important: Windows Keyring has strict limits on value length which may cause issues with large values.

VaultStore¶

Integration with HashiCorp Vault for enterprise secret management.

from key_value.aio.stores.vault import VaultStore

store = VaultStore(
    url="http://localhost:8200",
    token="your-token"
)

Installation:

pip install py-key-value-aio[vault]

Use Cases:

Enterprise secret management
Multi-environment deployments
Centralized secret rotation
Audit logging

Characteristics:

Enterprise-grade security
Centralized management
Audit logging
Stable storage format: Unstable

Distributed Stores¶

Distributed stores provide network-based storage for multi-node applications.

Store	Stability	Description
Aerospike	Unstable	High-performance distributed NoSQL database
Azure Tables	Unstable	Azure Table Storage key-value storage
DynamoDB	Unstable	AWS DynamoDB key-value storage
S3	Unstable	AWS S3 object storage
Elasticsearch	Unstable	Full-text search with key-value capabilities
Firestore	Unstable	Google Cloud Firestore key-value storage
Memcached	Unstable	High-performance distributed memory cache
MongoDB	Unstable	Document database used as key-value store
OpenSearch	Unstable	OpenSearch search engine with key-value capabilities
PostgreSQL	Unstable	PostgreSQL database with JSONB storage
Redis	Stable	Popular in-memory data structure store
Valkey	Stable	Open-source Redis fork

AerospikeStore¶

High-performance distributed NoSQL database optimized for speed and scale.

from key_value.aio.stores.aerospike import AerospikeStore

# Using host list
store = AerospikeStore(
    hosts=[("localhost", 3000)],
    namespace="test",
    set_name="kv-store"
)

# Or with an existing client
import aerospike
client = aerospike.client({"hosts": [("localhost", 3000)]})
client.connect()
store = AerospikeStore(client=client, namespace="test")

Installation:

pip install py-key-value-aio[aerospike]

Use Cases:

High-throughput applications
Real-time data processing
Large-scale distributed systems
Low-latency caching

Characteristics:

Sub-millisecond latency
Horizontal scaling
Native TTL support
Namespaces must be pre-configured on server
Sets created automatically
Stable storage format: Unstable

RedisStore¶

High-performance in-memory store using Redis.

from key_value.aio.stores.redis import RedisStore

store = RedisStore(url="redis://localhost:6379/0")

tls_store = RedisStore(
    url="rediss://redis.example.com:6380/0",
    ssl_ca_certs="/etc/ssl/certs/redis-ca.pem",
)

Redis TLS can be enabled with a rediss:// URL or with ssl=True when using host and port. Pass ssl_ca_certs to verify the server certificate, and ssl_certfile plus ssl_keyfile for mutual TLS deployments. For development or test environments with self-signed certificates, ssl_cert_reqs="none" and ssl_check_hostname=False are available, but should not be used for production.

Installation:

pip install py-key-value-aio[redis]

Use Cases:

Distributed caching
Session storage
Real-time applications
High-throughput systems

Characteristics:

Very fast (in-memory)
Production-ready
Rich feature set
Horizontal scaling support
SSL/TLS and mutual TLS connection options
Stable storage format

ValkeyStore¶

Open-source Redis fork with similar performance characteristics.

from key_value.aio.stores.valkey import ValkeyStore

store = ValkeyStore(host="localhost", port=6379)

For Valkey Cluster deployments, pass a GLIDE cluster configuration. The store will create the GlideClusterClient lazily on first use, so applications can configure the store before an async event loop is running:

from glide_shared.config import GlideClusterClientConfiguration, NodeAddress

from key_value.aio.stores.valkey import ValkeyStore

cluster_store = ValkeyStore(
    config=GlideClusterClientConfiguration(
        addresses=[NodeAddress("valkey-cluster.example.com", 6379)]
    )
)

Installation:

pip install py-key-value-aio[valkey]

Use Cases:

Same as Redis
Open-source preference
Redis API compatibility

Characteristics:

Redis-compatible
Open-source governance
Production-ready
Standalone and cluster-mode GLIDE clients
Stable storage format

FirestoreStore¶

Google Cloud Firestore used as a key-value store.

from key_value.aio.stores.firestore import (
    FirestoreStore,
    FirestoreV1CollectionSanitizationStrategy,
    FirestoreV1KeySanitizationStrategy,
)

store = FirestoreStore(
    credentials=google_credentials,
    database="firestore-db",
    key_sanitization_strategy=FirestoreV1KeySanitizationStrategy(),
    collection_sanitization_strategy=FirestoreV1CollectionSanitizationStrategy(),
)

Installation:

pip install py-key-value-aio[firestore]

Use Cases:

Google Cloud-native applications
Serverless / managed infrastructure
Existing Firestore deployments

Characteristics:

Managed cloud database
Document/collection model
Optional key and collection sanitization for Firestore ID constraints
Firestore sanitizers reserve H_ and S_ prefixes for keys and collections to prevent collisions
Stable storage format: Unstable

AzureTablesStore¶

Azure Table Storage integration for Azure-native applications.

from key_value.aio.stores.azure_tables import (
    AzureTablesSanitizationStrategy,
    AzureTablesStore,
)

store = AzureTablesStore(
    account_name="mystorageaccount",
    credential=credential,
    table_name="kvstore",
    collection_sanitization_strategy=AzureTablesSanitizationStrategy(),
    key_sanitization_strategy=AzureTablesSanitizationStrategy(),
)

Installation:

pip install py-key-value-aio[azure-tables]

Use Cases:

Azure-native applications
Serverless / managed infrastructure
Low-cost persistent state
Managed Identity authentication
OAuth/session state on Azure Container Apps

Characteristics:

Fully managed
Pay-per-use pricing
No native background TTL; entries are lazily expired on read
Optional PartitionKey/RowKey sanitization for Azure Tables limits
Stable storage format: Unstable

DynamoDBStore¶

AWS DynamoDB integration for serverless and cloud-native applications.

from key_value.aio.stores.dynamodb import DynamoDBStore

store = DynamoDBStore(
    table_name="kv-store",
    region_name="us-east-1"
)

Installation:

pip install py-key-value-aio[dynamodb]

Use Cases:

AWS-native applications
Serverless architectures
Global distribution
Managed infrastructure

Characteristics:

Fully managed
Auto-scaling
Global tables
Pay-per-use pricing
Stable storage format: Unstable

S3Store¶

AWS S3 object storage for durable, scalable key-value storage.

from key_value.aio.stores.s3 import S3Store

store = S3Store(
    bucket_name="my-kv-bucket",
    region_name="us-east-1"
)

Installation:

pip install py-key-value-aio[s3]

Use Cases:

Large value storage (up to 5TB per object)
Durable, long-term storage
Cost-effective archival
Multi-region replication

Characteristics:

99.999999999% durability
Automatic key sanitization for S3 path limits
Supports lifecycle policies
Pay-per-use pricing
Stable storage format: Unstable

ElasticsearchStore¶

Full-text search engine used as a key-value store.

from key_value.aio.stores.elasticsearch import ElasticsearchStore

store = ElasticsearchStore(
    url="https://localhost:9200",
    api_key="your-api-key",
    index="kv-store"
)

Installation:

pip install py-key-value-aio[elasticsearch]

Use Cases:

Applications already using Elasticsearch
Need for search capabilities
Analytics and logging

Characteristics:

Search capabilities
Distributed by default
Rich querying
Stable storage format: Unstable

OpenSearchStore¶

OpenSearch search engine used as a key-value store. Collections are stored in separate indices with values in flattened fields.

from key_value.aio.stores.opensearch import OpenSearchStore

# Using URL and API key
store = OpenSearchStore(
    url="https://localhost:9200",
    api_key="your-api-key",
    index_prefix="kv_store"
)

# With existing client
from opensearchpy import AsyncOpenSearch
client = AsyncOpenSearch(hosts=["https://localhost:9200"])
store = OpenSearchStore(
    opensearch_client=client,
    index_prefix="kv_store"
)

Installation:

pip install py-key-value-aio[opensearch]

Use Cases:

Applications already using OpenSearch
Need for search capabilities with AWS-managed service
Analytics and logging

Characteristics:

Search capabilities
Distributed by default
Collections stored in separate indices
Values stored in flat_object fields
Stable storage format: Unstable

Sanitization Strategies:

OpenSearch has specific restrictions on keys and index names. Use the built-in sanitization strategies to avoid issues:

from key_value.aio.stores.opensearch import (
    OpenSearchStore,
    OpenSearchV1KeySanitizationStrategy,
    OpenSearchV1CollectionSanitizationStrategy,
)

store = OpenSearchStore(
    url="https://localhost:9200",
    index_prefix="kv_store",
    key_sanitization_strategy=OpenSearchV1KeySanitizationStrategy(),
    collection_sanitization_strategy=OpenSearchV1CollectionSanitizationStrategy(),
)

MongoDBStore¶

Document database used as a key-value store.

from key_value.aio.stores.mongodb import MongoDBStore

store = MongoDBStore(url="mongodb://localhost:27017/test")

Installation:

pip install py-key-value-aio[mongodb]

Use Cases:

Applications already using MongoDB
Document-oriented data
Flexible schemas

Characteristics:

Document storage
Rich querying
Horizontal scaling
Stable storage format: Unstable

PostgreSQLStore¶

PostgreSQL database with JSONB storage for flexible key-value data.

Note: PostgreSQL is async-only. This store uses asyncpg which provides native async/await operations.

from key_value.aio.stores.postgresql import PostgreSQLStore

# Using connection URL
store = PostgreSQLStore(url="postgresql://localhost:5432/mydb")

# Using connection parameters
store = PostgreSQLStore(
    host="localhost",
    port=5432,
    database="mydb",
    user="myuser",
    password="mypass"
)

async with store:
    await store.put(key="user_1", value={"name": "Alice"}, collection="users")
    user = await store.get(key="user_1", collection="users")

Installation:

pip install py-key-value-aio[postgresql]

Use Cases:

Applications already using PostgreSQL
Need for SQL querying on stored data
ACID transaction requirements
Complex data relationships

Characteristics:

JSONB storage for efficient querying
TTL support via expiration timestamps
Single table design (collections as column values)
Async-only (uses asyncpg)
Stable storage format: Unstable

MemcachedStore¶

High-performance distributed memory caching system.

from key_value.aio.stores.memcached import MemcachedStore

store = MemcachedStore(host="127.0.0.1", port=11211)

Installation:

pip install py-key-value-aio[memcached]

Use Cases:

Distributed caching
Session storage
High-throughput caching

Characteristics:

Very fast
Simple protocol
Distributed by design
No persistence
Stable storage format: Unstable

Choosing a Store¶

Development¶

Recommended: MemoryStore or DiskStore

Fast iteration
No setup required
Easy debugging

# Development
from key_value.aio.stores.memory import MemoryStore
store = MemoryStore()

Production Caching¶

Recommended: RedisStore or ValkeyStore

High performance
Distributed
Production-ready
Stable storage format

# Production caching
from key_value.aio.stores.redis import RedisStore
store = RedisStore(url="redis://localhost:6379/0")

Long-Term Storage¶

Recommended: Stores with Stable stability rating

RedisStore
ValkeyStore
DiskStore
MultiDiskStore
KeyringStore

Avoid unstable stores for data you can't afford to lose or migrate.

Sensitive Data¶

Recommended: KeyringStore or VaultStore

OS-level encryption
Secure by default
Audit logging (Vault)

# Sensitive data
from key_value.aio.stores.keyring import KeyringStore
store = KeyringStore(service_name="my-app")

Serverless/Cloud¶

Recommended: DynamoDBStore (AWS)

Fully managed
Auto-scaling
No servers to maintain

# AWS Lambda
from key_value.aio.stores.dynamodb import DynamoDBStore
store = DynamoDBStore(table_name="kv-store", region_name="us-east-1")

Store Compatibility¶

All stores implement the same protocol, making it easy to switch:

# Development
store = MemoryStore()

# Production
store = RedisStore(url="redis://localhost:6379/0")

# Your code works with both!
await store.put(key="user:123", value={"name": "Alice"}, collection="users")

See the API Reference for complete store documentation.