fix(routing): dedup decision engine connectors by merchant_connector_id

[prajjwalkumar17]

Type of Change

• Bugfix

Description

transform_de_output_for_router builds the router-facing connector list from a decision engine /routing/evaluate response: evaluated_output first, then the remaining connectors from output as fallbacks, deduplicated.

The dedup set was keyed on the connector name alone:

if seen.insert(eval_conn.connector) { ... }   // merchant_connector_id not part of the key

RoutableConnectorChoice carries both connector and merchant_connector_id, but only connector went into the set. So for a merchant holding several MCAs of the same connector, every MCA after the first is silently dropped.

This shows up on any rule that spans two MCAs of one connector — e.g. a volume split across two paypal MCAs. DE returns the sampled winner in evaluated_output and both arms in output; the second paypal MCA is then discarded, so the router loses its fallback and the list length disagrees with the legacy euclid result.

Found while investigating a persistent shadow-diff (is_equal_length=false) on a sandbox profile with exactly that shape.

Change

Key the set on the full (connector, merchant_connector_id) pair.

The second loop now builds the RoutableConnectorChoice before keying off it, which also removes a duplicated RoutableConnectors::from_str — the TryFrom<ConnectorInfo> impl already performs that parse with equivalent error logging.

Note: entries with merchant_connector_id: None no longer collapse into a same-named entry with Some(id). Both lists are serialized from the same ConnectorInfo objects in the same rule, so this is not expected to trigger in practice.

Motivation and Context

Static routing via the decision engine must preserve MCA-level granularity; collapsing by connector name loses both the routing target and its fallbacks for multi-MCA merchants.

How did you test it?

Unit tests added in crates/router/src/core/payments/routing/utils.rs (the file had no test module before):

running 3 tests
test ...::still_dedups_exact_duplicates ... ok
test ...::preserves_evaluated_first_ordering_for_distinct_connectors ... ok
test ...::keeps_both_mcas_of_the_same_connector ... ok

test result: ok. 3 passed; 0 failed

Verified the coverage is meaningful by reverting the fix and re-running with the tests in place — the regression test fails exactly as the bug predicts, while the two guard tests still pass:

test ...::keeps_both_mcas_of_the_same_connector ... FAILED

assertion `left == right` failed: second paypal MCA was dropped
  left: 1
 right: 2

test result: FAILED. 2 passed; 1 failed

left: 1 / right: 2 is the same shape as the sandbox shadow diff that led here (DE one connector, HS two).

Also run: cargo check -p router, cargo clippy -p router, cargo +nightly fmt -p router --check — all clean.

Checklist

• I formatted the code cargo +nightly fmt --all
• I addressed lints thrown by cargo clippy
• I added unit tests

🤖 Generated with Claude Code

Closes #13759

[semanticdiff-com]

Review changes with  

Changed Files

File	Status
crates/router/src/core/payments/routing/utils.rs	5% smaller