event-store-design

Event Store Design

Comprehensive guide to designing event stores for event-sourced applications.

Do not use this skill when

• The task is unrelated to event store design
• You need a different domain or tool outside this scope

Instructions

• Clarify goals, constraints, and required inputs.
• Apply relevant best practices and validate outcomes.
• Provide actionable steps and verification.
• If detailed examples are required, open resources/implementation-playbook.md.

Use this skill when

• Designing event sourcing infrastructure
• Choosing between event store technologies
• Implementing custom event stores
• Optimizing event storage and retrieval
• Setting up event store schemas
• Planning for event store scaling

Core Concepts

1. Event Store Architecture

┌─────────────────────────────────────────────────────┐
│                    Event Store                       │
├─────────────────────────────────────────────────────┤
│  ┌─────────────┐  ┌─────────────┐  ┌─────────────┐ │
│  │   Stream 1   │  │   Stream 2   │  │   Stream 3   │ │
│  │ (Aggregate)  │  │ (Aggregate)  │  │ (Aggregate)  │ │
│  ├─────────────┤  ├─────────────┤  ├─────────────┤ │
│  │ Event 1     │  │ Event 1     │  │ Event 1     │ │
│  │ Event 2     │  │ Event 2     │  │ Event 2     │ │
│  │ Event 3     │  │ ...         │  │ Event 3     │ │
│  │ ...         │  │             │  │ Event 4     │ │
│  └─────────────┘  └─────────────┘  └─────────────┘ │
├─────────────────────────────────────────────────────┤
│  Global Position: 1 → 2 → 3 → 4 → 5 → 6 → ...     │
└─────────────────────────────────────────────────────┘

2. Event Store Requirements

Requirement	Description
Append-only	Events are immutable, only appends
Ordered	Per-stream and global ordering
Versioned	Optimistic concurrency control
Subscriptions	Real-time event notifications
Idempotent	Handle duplicate writes safely

Technology Comparison

Technology	Best For	Limitations
EventStoreDB	Pure event sourcing	Single-purpose
PostgreSQL	Existing Postgres stack	Manual implementation
Kafka	High-throughput streaming	Not ideal for per-stream queries
DynamoDB	Serverless, AWS-native	Query limitations
Marten	.NET ecosystems	.NET specific

Templates

Template 1: PostgreSQL Event Store Schema

-- Events table
CREATE TABLE events (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    stream_id VARCHAR(255) NOT NULL,
    stream_type VARCHAR(255) NOT NULL,
    event_type VARCHAR(255) NOT NULL,
    event_data JSONB NOT NULL,
    metadata JSONB DEFAULT '{}',
    version BIGINT NOT NULL,
    global_position BIGSERIAL,
    created_at TIMESTAMPTZ DEFAULT NOW(),

    CONSTRAINT unique_stream_version UNIQUE (stream_id, version)
);

-- Index for stream queries
CREATE INDEX idx_events_stream_id ON events(stream_id, version);

-- Index for global subscription
CREATE INDEX idx_events_global_position ON events(global_position);

-- Index for event type queries
CREATE INDEX idx_events_event_type ON events(event_type);

-- Index for time-based queries
CREATE INDEX idx_events_created_at ON events(created_at);

-- Snapshots table
CREATE TABLE snapshots (
    stream_id VARCHAR(255) PRIMARY KEY,
    stream_type VARCHAR(255) NOT NULL,
    snapshot_data JSONB NOT NULL,
    version BIGINT NOT NULL,
    created_at TIMESTAMPTZ DEFAULT NOW()
);

-- Subscriptions checkpoint table
CREATE TABLE subscription_checkpoints (
    subscription_id VARCHAR(255) PRIMARY KEY,
    last_position BIGINT NOT NULL DEFAULT 0,
    updated_at TIMESTAMPTZ DEFAULT NOW()
);

Template 2: Python Event Store Implementation

from dataclasses import dataclass, field
from datetime import datetime, timezone
from typing import Any, Optional, List
from uuid import UUID, uuid4
import json
import asyncio
import asyncpg

@dataclass
class Event:
    stream_id: str
    event_type: str
    data: dict
    metadata: dict = field(default_factory=dict)
    event_id: UUID = field(default_factory=uuid4)
    version: Optional[int] = None
    global_position: Optional[int] = None
    created_at: datetime = field(default_factory=lambda: datetime.now(timezone.utc))


class EventStore:
    def __init__(self, pool: asyncpg.Pool):
        self.pool = pool

    async def append_events(
        self,
        stream_id: str,
        stream_type: str,
        events: List[Event],
        expected_version: Optional[int] = None
    ) -> List[Event]:
        """Append events to a stream with optimistic concurrency."""
        async with self.pool.acquire() as conn:
            async with conn.transaction():
                # Serialize ALL appenders before allocating positions. This simple global
                # lock trades throughput for commit-consistent polling; every writer must use it.
                await conn.execute("SELECT pg_advisory_xact_lock(784291)")
                # Check expected version
                if expected_version is not None:
                    current = await conn.fetchval(
                        "SELECT MAX(version) FROM events WHERE stream_id = $1",
                        stream_id
                    )
                    current = current or 0
                    if current != expected_version:
                        raise ConcurrencyError(
                            f"Expected version {expected_version}, got {current}"
                        )

                # Get starting version
                start_version = await conn.fetchval(
                    "SELECT COALESCE(MAX(version), 0) + 1 FROM events WHERE stream_id = $1",
                    stream_id
                )

                # Insert events
                saved_events = []
                for i, event in enumerate(events):
                    event.version = start_version + i
                    row = await conn.fetchrow(
                        """
                        INSERT INTO events (id, stream_id, stream_type, event_type,
                                          event_data, metadata, version, created_at)
                        VALUES ($1, $2, $3, $4, $5, $6, $7, $8)
                        RETURNING global_position
                        """,
                        event.event_id,
                        stream_id,
                        stream_type,
                        event.event_type,
                        json.dumps(event.data),
                        json.dumps(event.metadata),
                        event.version,
                        event.created_at
                    )
                    event.global_position = row['global_position']
                    saved_events.append(event)

                return saved_events

    async def read_stream(
        self,
        stream_id: str,
        from_version: int = 0,
        limit: int = 1000
    ) -> List[Event]:
        """Read events from a stream."""
        async with self.pool.acquire() as conn:
            rows = await conn.fetch(
                """
                SELECT id, stream_id, event_type, event_data, metadata,
                       version, global_position, created_at
                FROM events
                WHERE stream_id = $1 AND version >= $2
                ORDER BY version
                LIMIT $3
                """,
                stream_id, from_version, limit
            )
            return [self._row_to_event(row) for row in rows]

    async def read_all(
        self,
        from_position: int = 0,
        limit: int = 1000
    ) -> List[Event]:
        """Read all events globally."""
        async with self.pool.acquire() as conn:
            rows = await conn.fetch(
                """
                SELECT id, stream_id, event_type, event_data, metadata,
                       version, global_position, created_at
                FROM events
                WHERE global_position > $1
                ORDER BY global_position
                LIMIT $2
                """,
                from_position, limit
            )
            return [self._row_to_event(row) for row in rows]

    async def subscribe(
        self,
        subscription_id: str,
        handler,
        from_position: int = 0,
        batch_size: int = 100
    ):
        """Subscribe to all events from a position."""
        # Get checkpoint
        async with self.pool.acquire() as conn:
            checkpoint = await conn.fetchval(
                """
                SELECT last_position FROM subscription_checkpoints
                WHERE subscription_id = $1
                """,
                subscription_id
            )
            position = from_position if checkpoint is None else checkpoint

        while True:
            events = await self.read_all(position, batch_size)
            if not events:
                await asyncio.sleep(1)  # Poll interval
                continue

            for event in events:
                await handler(event)
                position = event.global_position

            # Save checkpoint
            async with self.pool.acquire() as conn:
                await conn.execute(
                    """
                    INSERT INTO subscription_checkpoints (subscription_id, last_position)
                    VALUES ($1, $2)
                    ON CONFLICT (subscription_id)
                    DO UPDATE SET last_position = $2, updated_at = NOW()
                    """,
                    subscription_id, position
                )

    def _row_to_event(self, row) -> Event:
        return Event(
            event_id=row['id'],
            stream_id=row['stream_id'],
            event_type=row['event_type'],
            data=json.loads(row['event_data']),
            metadata=json.loads(row['metadata']),
            version=row['version'],
            global_position=row['global_position'],
            created_at=row['created_at']
        )


class ConcurrencyError(Exception):
    """Raised when optimistic concurrency check fails."""
    pass

Template 3: EventStoreDB Usage

from esdbclient import EventStoreDBClient, NewEvent, StreamState
import json

# Connect
client = EventStoreDBClient(uri="esdb://localhost:2113?tls=false")

# Append events
def append_events(stream_name: str, events: list, expected_revision=None):
    new_events = [
        NewEvent(
            type=event['type'],
            data=json.dumps(event['data']).encode(),
            metadata=json.dumps(event.get('metadata', {})).encode()
        )
        for event in events
    ]

    if expected_revision is None:
        state = StreamState.ANY
    elif expected_revision == -1:
        state = StreamState.NO_STREAM
    else:
        state = expected_revision

    return client.append_to_stream(
        stream_name=stream_name,
        events=new_events,
        current_version=state
    )

# Read stream
def read_stream(stream_name: str, from_revision: int = 0):
    events = client.get_stream(
        stream_name=stream_name,
        stream_position=from_revision
    )
    return [
        {
            'type': event.type,
            'data': json.loads(event.data),
            'metadata': json.loads(event.metadata) if event.metadata else {},
            'stream_position': event.stream_position,
            'commit_position': event.commit_position
        }
        for event in events
    ]

# Subscribe to all
def subscribe_to_all(handler, from_position: int = 0):
    subscription = client.subscribe_to_all(commit_position=from_position)
    for event in subscription:
        handler({
            'type': event.type,
            'data': json.loads(event.data),
            'stream_id': event.stream_name,
            'position': event.commit_position
        })

# Category projection ($ce-Category)
def read_category(category: str):
    """Read all events for a category using system projection."""
    return read_stream(f"$ce-{category}")

Template 4: DynamoDB Event Store

import boto3
from boto3.dynamodb.conditions import Key
from datetime import datetime, timezone
import json
import uuid

class DynamoEventStore:
    def __init__(self, table_name: str):
        self.dynamodb = boto3.resource('dynamodb')
        self.table = self.dynamodb.Table(table_name)

    def append_events(self, stream_id: str, events: list, expected_version: int = None):
        """Append events with conditional write for concurrency."""
        # Single-event teaching example; no batch atomicity or automatic retry.
        if len(events) != 1 or expected_version is None or expected_version < 0:
            raise ValueError("Supply exactly one event and a nonnegative expected version")
        event = events[0]
        version = expected_version + 1
        item = {
            'PK': f"STREAM#{stream_id}",
            'SK': f"VERSION#{version:020d}",
            'event_id': str(uuid.uuid4()),
            'event_type': event['type'],
            'event_data': json.dumps(event['data']),
            'version': version,
        }
        self.table.put_item(Item=item, ConditionExpression='attribute_not_exists(PK)')
        return events

    def read_stream(self, stream_id: str, from_version: int = 0):
        """Read events from a stream."""
        response = self.table.query(
            ConsistentRead=True,
            KeyConditionExpression=Key('PK').eq(f"STREAM#{stream_id}") &
                                  Key('SK').gte(f"VERSION#{from_version:020d}")
        )
        return [
            {
                'event_type': item['event_type'],
                'data': json.loads(item['event_data']),
                'version': item['version']
            }
            for item in response['Items']
        ]

# Table definition (CloudFormation/Terraform)
"""
DynamoDB Table:
  - PK (Partition Key): String
  - SK (Sort Key): String
  - No global subscription ordering implemented here

Capacity: On-demand or provisioned based on throughput needs
"""

Best Practices

Do's

• Use stream IDs that include aggregate type - Order-{uuid}
• Include correlation/causation IDs - For tracing
• Version events from day one - Plan for schema evolution
• Implement idempotency - Use event IDs for deduplication
• Index appropriately - For your query patterns

Don'ts

• Don't update or delete events - They're immutable facts
• Don't store large payloads - Keep events small
• Don't skip optimistic concurrency - Prevents data corruption
• Don't ignore backpressure - Handle slow consumers

Resources

• EventStoreDB
• Marten Events
• Event Sourcing Pattern

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.

Implementation limits and verification

The PostgreSQL sequence alone is not commit order: concurrent transactions can allocate positions then commit out of order, causing a polling checkpoint to skip late commits. The teaching implementation serializes all appenders with one transaction advisory lock; bypassing it invalidates that guarantee and its throughput must be measured. The unique stream/version constraint already supplies the stream index. Append-only permissions, stable request/event IDs and duplicate request reconciliation still need implementation: a UUID primary key prevents a duplicate insert but does not turn retries into successful idempotent responses. Never reuse mutated in-memory event versions as proof a transaction committed.

The DynamoDB example is one conditional event write, not a complete event store: it prevents overwrite at that key but does not atomically check a stream-head version, fill gaps, deduplicate requests, implement global ordering or multi-event batches. Use a transaction with a conditional head record for those guarantees; read_stream returns only one query page and must follow LastEvaluatedKey to finish. Do not checkpoint an eventually consistent timestamp GSI as a total order.

The esdbclient example uses the synchronous client/iterator with a synchronous handler; use its version-matched async client for an async service. Its plaintext localhost URI is for an isolated development server only; use authenticated TLS for remote connections. Category streams need enabled/configured system projections. Bound reads, implement shutdown/backpressure and one checkpoint owner per subscription. Handlers can repeat after failure: make their external effects idempotent or transactional. Snapshots are an optimization, not the source of truth. Rehearse concurrent appends, interrupted commits, duplicate IDs, projection replay and restore before deployment. Avoid raw personal data in immutable events; define retention and lawful redaction/crypto-erasure policies rather than absolute claims that historical data can never be removed.