A query may need data owned by another service, such as an author’s display name, without a join across the two stores. A read replica is a consumer-owned local copy of selected data from the source’s published contract. The source remains authoritative, and the copy is only as fresh as its latest replication update. section 11.6 section 11.7
Use a replica when several reads need a relatively broad set of published data, can tolerate lag, and can meet their response-time targets using the local representation. Map contract fields at the boundary into the consumer’s read model; do not mirror the source’s private database schema. section 11.6 section 11.7
Compare the alternatives for each query:
- Ask the source when the answer must be current. There is no local copy to reconcile, but the query inherits the source’s latency and availability. section 11.6
- Read a replica when a broader local representation serves the query within its response-time target and the consumer can operate the copy. section 11.7
- Consider a projected view when repeated filtering, joining, aggregation, or ordering makes request-time work too slow or complex. It stores a query-shaped result, but also requires updates and drift management. section 11.7
Table 11.3 Choosing a “read” strategy
|
Strategy
|
Model shape
|
Freshness
|
Coupling
|
Who owns
the copy
|
Consistency
|
|
Ask the
source context
|
The author
service’s model, returned through its API; WriteFlow keeps no copy
|
Current at
the moment of the call
|
Your query
inherits the source context’s availability and latency
|
The source
context holds the only copy
|
Source
context’s response at call time; no shared transaction or snapshot with
WriteFlow’s data.
|
|
Read a
replica
|
A broader
local copy of selected data from the author service’s published contract,
mapped into WriteFlow’s read model.
|
As fresh as
the last replication update
|
You run the
replication and depend on the source context for updates
|
The
consumer holds the copy; the source context holds the truth
|
Eventually
consistent; the copy can drift
|
|
Project a
view
|
A
WriteFlow-owned, query-shaped view with only the fields the query needs; it
may combine data from multiple source contexts if the query requires it
|
As fresh as
the last projection event
|
You own the
projection and depend on the source context’s events
|
The
consumer owns the view; each source context owns its data
|
Eventually
consistent; updates can arrive twice or out of order
|
Compare the read strategies by model shape, freshness, coupling, ownership, and consistency. section 11.7
A replica needs operational care. Updates can be duplicated, arrive out of order, or be missed; deletes and partial updates can also leave the copy inconsistent. The consumer must update it, check for drift, and be able to rebuild or reconcile it. section 11.6 section 11.9
An article query
An article stores an author identifier but needs the author’s display name. If the query tolerates lag, it can read a local replica of published author data. An output port lets the handler use consumer-owned types while an adapter hides whether the answer comes from the replica, the source API, or a projected view. This read choice does not replace author validation before approval: stale read-side data must not decide whether an action is allowed. section 11.6 section 11.7 section 11.9


