fix(kv-router): stamp request identity on decode-affinity scoring row

[nv-yna]

Summary

DefaultWorkerScorer::worker_logit has three tracing::debug! sites that emit one candidate scoring row each. Two of them carry request_id and worker_type so the candidates of a single routing decision can be grouped by id in one hop. The third — the decode-affinity branch — carries neither, and it returns before reaching the other two.

The result is that any decision taking the decode-affinity branch emits candidate rows without request identity, so grouping them back into a decision requires the line-adjacency heuristic that stamping was added to replace.

This is not occasional. The branch predicate reads the configured overlap_score_credit (default 1.0), not any measured overlap:

if self.worker_type == "decode"
    && !context.track_prefill_tokens
    && weights.overlap_score_credit > 0.0

so on an affected deployment every decision takes the branch and 100% of candidate rows lose their identity fields — even when the measured overlap is zero.

Affected configurations

Deployment	worker_type	track_prefill_tokens	overlap_score_credit	Affected
Aggregated + --no-router-track-prefill-tokens	decode	false (flag)	1.0 (default)	Yes — all decisions
Aggregated, default flags	decode	true	1.0	No
Disaggregated, decode pool	decode	false (forced)	0.0 (forced)	No
Disaggregated, prefill pool	prefill	true	1.0	No

Aggregated deployments route through worker_type = "decode" unconditionally, so the aggregated case only needs the user-facing --no-router-track-prefill-tokens flag to be affected. The disaggregated decode pool is spared because build_decode_router_override forces overlap_score_credit: Some(0.0), making the third condition false.

Origin

Two PRs landed the same day touching this function. #12370 (15:24 PT) added the identity fields to the two rows that existed then. #11720 (19:39 PT) added this third row from a branch that predated the stamping. git merge-base --is-ancestor confirms #12370 was already in #11720's ancestry, so this is a parallel-development merge miss rather than an intentional change — the two PRs touched disjoint lines, so git saw no conflict.

Fix

Add the two fields so all three candidate rows share one shape. context.request_id and self.worker_type are already in scope in this branch (it already reads self.worker_type and context.track_prefill_tokens in its own predicate), both are borrowed &str with no allocation, and both are evaluated inside the macro so they cost nothing when DEBUG is disabled.

Severity is observability only. Routing behaviour, metrics and API surfaces are unchanged, and these rows are DEBUG-level, so a default INFO deployment sees no change.

Follow-ups not in this PR

• No in-repo test asserts these fields on selector rows, which is why CI did not catch the original gap. tests/utils/router_logs.py already parses request_id out of router logs, so the assertion has a natural home.
• The three emit sites duplicate the identity fields by hand, which is what let them drift apart. Factoring the emit into one helper would make the next scoring branch inherit the fields automatically.

Validation

• cargo check -p dynamo-kv-router — clean
• cargo clippy -p dynamo-kv-router — no new warnings
• cargo fmt --all -- --check — clean

Behavioural verification was done on the 1.4.0 line, where the same defect is present, by running an aggregated deployment (dynamo.frontend --router-mode kv + dynamo.mocker --num-workers 3, Qwen3-0.6B) twice with only --no-router-track-prefill-tokens differing:

• control (flag off): 24 of 24 candidate rows carried both fields
• subject (flag on): 24 of 24 candidate rows carried neither, and overlap_credit_blocks: 0.000 confirmed the branch fired with no measured overlap

The control arm establishes that the log directives, deployment and field parsers all work, so the subject result is attributable to the branch and not to test setup.

An internal Linear ticket also tracks this (DYN-3804).

---

Summary by CodeRabbit

• Bug Fixes
  • Enhanced decode-worker scoring logs with request IDs and worker types for more complete diagnostic information.

[github-actions]

👋 Hi nv-yna! Thank you for contributing to ai-dynamo/dynamo.

Just a reminder: The NVIDIA Test Github Validation CI runs an essential subset of the testing framework to quickly catch errors.Your PR reviewers may elect to test the changes comprehensively before approving your changes.

🚀

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 695f7b7b-7d9d-4a6f-9aaf-1be7a9a9ed21

📥 Commits

Reviewing files that changed from the base of the PR and between c0b5e55 and e6c76ef.

📒 Files selected for processing (1)

• lib/kv-router/src/scheduling/selector/default.rs

---

Walkthrough

The decode-worker scoring debug log now includes request_id and worker_type fields.

Changes

Decode scoring logging

Layer / File(s)	Summary
Add decode scoring metadata lib/kv-router/src/scheduling/selector/default.rs	The decode scoring debug event records request_id and worker_type before the existing formula details.

Estimated code review effort: 1 (Trivial) | ~2 minutes

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The description gives detailed scope, impact, affected configurations, fix, and validation, but it omits the required Related Issues section and reviewer-start guidance.	Add the required Related Issues section and confirm either the linked issue number or that no related issue exists; also identify where reviewers should start.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the primary change: adding request identity to decode-affinity scoring logs.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

---

_{Comment @coderabbitai help to get the list of available commands.}

[datadog-official]

🎯 Code Coverage (details)
• Patch Coverage: 100.00%
• Overall Coverage: 41.92% (-0.39%)

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: e6c76ef | Docs | Datadog PR Page | Give us feedback!}