Events table exists

UUID primary key with gen_random_uuid()

stream_type column

event_data as JSONB

metadata as JSONB with default

global_position as BIGSERIAL

Unique stream version constraint

Stream query index

Global position index

Snapshots table

Subscription checkpoints table

Stream ID naming guidance

Uses asyncpg

Transaction wrapping

Optimistic concurrency check

Event ID for idempotency

Correlation/causation IDs in metadata

Stream ID with aggregate type

Subscription checkpoint persistence

Checkpoint upsert pattern

Backpressure / poll interval

read_stream ordered by version

read_all ordered by global_position

No update/delete of events

Uses esdbclient package

StreamState.ANY for unconditional append

StreamState.NO_STREAM for new streams

Category projection stream naming

current_version parameter used

NewEvent for constructing events

Data encoded as bytes

Technology selection rationale

Subscription or read implementation

No update or delete of events

Stream ID includes aggregate type

PK uses STREAM# prefix

SK uses VERSION# prefix with zero-padding

GSI for global ordering

GSI1PK set to constant 'EVENTS'

Uses batch_writer for appending

Optimistic concurrency via conditional write

read_stream uses KeyConditionExpression

Stream ID naming with aggregate type

Table definition provided

Uses boto3 DynamoDB resource

Stream ID includes aggregate type

Correlation ID in metadata

Causation ID in metadata

Schema/event version field

Event ID for idempotency

Payload excludes large/derived data

No UPDATE/DELETE guidance

Concrete event examples in JSON

Metadata separate from payload

Event type naming

Timestamp field

DynamoDB for serverless team

PostgreSQL for existing stack team

Kafka for high-throughput team

Kafka per-stream query limitation

Comparison table present

EventStoreDB described as pure event sourcing

PostgreSQL manual implementation limitation

DynamoDB query limitations noted

Justification per recommendation

Marten mentioned

Technology output file