Publish to multiple output ports

A product may expose the same checked table through more than one target. The first output port is the primary release. Each later port uses an adapter with dr_write_target(). Sources and transformations run once, and every output receives the checked delivery.

contract <- dataraft.core::dr_contract("customers",
  columns = c(id = "integer"), key = "id")
product <- dataraft.core::dr_product("customers",
  data.frame(id = 1:3), contract = contract) |>
  dataraft.core::dr_add_output(dataraft.core::dr_output(
    "primary", dataraft.adapters::dr_target_rds("customers-primary"),
    sla = dataraft.core::dr_sla(
      available_by = "08:00", timezone = "Europe/Berlin"))) |>
  dataraft.core::dr_add_output(dataraft.core::dr_output(
    "secondary", dataraft.adapters::dr_target_rds("customers-secondary")))

result <- dataraft.core::dr_run(product,
  business_date = as.character(Sys.Date()), evidence = "run-evidence")
result$port_outputs
result$metadata$sla$primary
dataraft.core::dr_read_run("run-evidence", result$run_id)$port_outputs

The SLA check uses the successful publication time and the explicit business date. A late delivery remains published and records late in the result and durable evidence. dr_check_sla() also evaluates missing deliveries when no run takes place.

Recover from a partial publication

Outputs commit in order. They do not share a transaction across destinations. If a later writer fails, the run has status error, while result$port_outputs names the committed and failed ports. The primary release remains available. Inspect each destination and the saved run evidence before retrying, especially when a target appends data. Use stop_on_failure = FALSE to receive the result directly; the default error includes it as condition$result.

result <- dataraft.core::dr_run(product,
  business_date = as.character(Sys.Date()),
  stop_on_failure = FALSE, evidence = "run-evidence")
result$status
result$port_outputs
# After repairing the failed destination, resume only unpublished ports:
resumed <- dataraft.core::dr_retry_ports(product, result,
  evidence = "run-evidence")
resumed$port_outputs

Keep the original in-memory result: the saved JSON evidence contains no data rows. The retry retains the run ID and never rewrites a port marked published. If an unpublished port has an SLA, provide business_date again. If a writer fails during retry, its receipt stays failed and later ports remain pending.

Cache reuse is disabled for products with multiple output ports, so an unchanged primary release cannot silently skip a secondary publication. RDS, database and Parquet targets are available through dataraft.adapters; choose a target with publication behavior that fits each consumer.