azure-data-tables-py – skill DevOps & Hạ tầng cho agent AI

Azure Tables SDK for Python

NoSQL key-value store for structured data (Azure Storage Tables or Cosmos DB Table API).

Installation

pip install azure-data-tables azure-identity

Environment Variables

# Azure Storage Tables
AZURE_STORAGE_ACCOUNT_URL=https://<account>.table.core.windows.net

# Cosmos DB Table API
COSMOS_TABLE_ENDPOINT=https://<account>.table.cosmos.azure.com

Authentication

from azure.identity import DefaultAzureCredential
from azure.data.tables import TableServiceClient, TableClient

credential = DefaultAzureCredential()
endpoint = "https://<account>.table.core.windows.net"

# Service client (manage tables)
service_client = TableServiceClient(endpoint=endpoint, credential=credential)

# Table client (work with entities)
table_client = TableClient(endpoint=endpoint, table_name="mytable", credential=credential)

Client Types

| Client | Purpose |

|--------|---------|

| TableServiceClient | Create/delete tables, list tables |

| TableClient | Entity CRUD, queries |

These are independent fragments. Confirm account/table/partition/row and exact write/delete scope; protect existing entities, use ETags/match conditions for concurrent changes and reconcile timeouts before retries. Storage Table and Cosmos Table authentication/feature/transaction limits differ; verify the selected endpoint rather than assuming one identity configuration fits both. Bound queries/pages/bytes and do not log entity contents. Close async credentials as well as clients on failure. Currency double examples are illustrative; choose exact asset units for money.

Table Operations

# Create table
service_client.create_table("mytable")

# Create if not exists
service_client.create_table_if_not_exists("mytable")

# Delete table
service_client.delete_table("mytable")

# List tables
for table in service_client.list_tables():
    print(table.name)

# Get table client
table_client = service_client.get_table_client("mytable")

Entity Operations

Important: Every entity requires PartitionKey and RowKey (together form unique ID).

Create Entity

entity = {
    "PartitionKey": "sales",
    "RowKey": "order-001",
    "product": "Widget",
    "quantity": 5,
    "price": 9.99,
    "shipped": False
}

# Create (fails if exists)
table_client.create_entity(entity=entity)

# Upsert explicitly replaces; omitted mode defaults can merge instead.
from azure.data.tables import UpdateMode
table_client.upsert_entity(entity=entity, mode=UpdateMode.REPLACE)

Get Entity

# Get by key (fastest)
entity = table_client.get_entity(
    partition_key="sales",
    row_key="order-001"
)
print(f"Product: {entity['product']}")

Update Entity

# Replace entire entity
entity["quantity"] = 10
table_client.update_entity(entity=entity, mode="replace")

# Merge (update specific fields only)
update = {
    "PartitionKey": "sales",
    "RowKey": "order-001",
    "shipped": True
}
table_client.update_entity(entity=update, mode="merge")

Delete Entity

table_client.delete_entity(
    partition_key="sales",
    row_key="order-001"
)

Query Entities

Query Within Partition

# Query by partition (efficient)
entities = table_client.query_entities(
    query_filter="PartitionKey eq 'sales'"
)
for entity in entities:
    print(entity["RowKey"])  # Do not dump property payloads

Query with Filters

# Filter by properties
entities = table_client.query_entities(
    query_filter="PartitionKey eq 'sales' and quantity gt 3"
)

# With parameters (safer)
entities = table_client.query_entities(
    query_filter="PartitionKey eq @pk and price lt @max_price",
    parameters={"pk": "sales", "max_price": 50.0}
)

Select Specific Properties

entities = table_client.query_entities(
    query_filter="PartitionKey eq 'sales'",
    select=["RowKey", "product", "price"]
)

List All Entities

# List all (cross-partition - use sparingly)
for entity in table_client.list_entities():
    print(entity["RowKey"])  # Do not dump property payloads

Batch Operations

from azure.data.tables import TableTransactionError

# Batch operations (same partition only!)
operations = [
    ("create", {"PartitionKey": "batch", "RowKey": "1", "data": "first"}),
    ("create", {"PartitionKey": "batch", "RowKey": "2", "data": "second"}),
    ("upsert", {"PartitionKey": "batch", "RowKey": "3", "data": "third"}),
]

try:
    table_client.submit_transaction(operations)
except TableTransactionError as e:
    print(f"Transaction failed: {e}")

Async Client

from azure.data.tables.aio import TableServiceClient, TableClient
from azure.identity.aio import DefaultAzureCredential

async def table_operations():
    credential = DefaultAzureCredential()
    
    async with TableClient(
        endpoint="https://<account>.table.core.windows.net",
        table_name="mytable",
        credential=credential
    ) as client:
        # Create
        await client.create_entity(entity={
            "PartitionKey": "async",
            "RowKey": "1",
            "data": "test"
        })
        
        # Query
        async for entity in client.query_entities("PartitionKey eq 'async'"):
            print(entity["RowKey"])  # Do not dump property payloads

import asyncio
asyncio.run(table_operations())

Data Types

| Python Type | Table Storage Type |

|-------------|-------------------|

| str | String |

| int | Int64 |

| float | Double |

| bool | Boolean |

| datetime | DateTime |

| bytes | Binary |

| UUID | Guid |

Best Practices

1. Design partition keys for query patterns and even distribution

2. Query within partitions whenever possible (cross-partition is expensive)

3. Use batch operations for multiple entities in same partition

4. Choose upsert mode deliberately; idempotence also depends on business effects and concurrency

5. Use parameterized queries to prevent injection

6. Keep entities small — max 1MB per entity

7. Use async client for high-throughput scenarios

When to Use

This skill is applicable to execute the workflow or actions described in the overview.

Limitations

• Use this skill only when the task clearly matches the scope described above.
• Do not treat the output as a substitute for environment-specific validation, testing, or expert review.
• Stop and ask for clarification if required inputs, permissions, safety boundaries, or success criteria are missing.