Consistency and recovery

Distribution and consistency

Mastering, indexing, and delivery can complete at different times. A design must distinguish:

• the write or review request being accepted;
• the entity or record becoming searchable;
• the mastering task reaching a terminal state; and
• each downstream destination accepting the result.

Use tasks, index status, product events, and destination-side checks for their respective scopes. Never treat the first successful response as proof of the entire end-to-end outcome.

Recovery design

Before production use, establish procedures for incorrect merge, disconnect, ignore, deletion, ambiguous timeout, and failed delivery. Record the entity, candidate and record identifiers, time, actor, comment, task or event, and trace identifier without copying sensitive record contents into an ordinary support channel.

See Integration approaches and Recover tasks and deliveries.