[sawsa307]

Overview

Remove the producer-side branch of NixlPushConnectorWorker._get_new_notifs, which no push-mode code path can reach. Split out of #61488 / #61580 as an independent simplification, as suggested in #61580 (comment).

Claims

• No behavior change for P or D in push mode: the removed branch only ran on a plain "<req_id>:<n>" notif arriving at P, which push mode never sends.
• _get_new_notifs now only counts consumer-side completion notifs and always returns an empty set; P keeps reporting finished sends from its WRITE handles in get_transfer_results.
• 26 insertions, 59 deletions.

Validation

• pytest tests/v1/kv_connector/unit/test_nixl_push_connector.py tests/v1/kv_connector/unit/test_nixl_heartbeat.py: 103 passed.
• pytest tests/v1/kv_connector/unit/test_nixl_connector.py: 182 passed. The two test_abort_timeout_on_prefiller cases fail locally on main as well.
• With a tripwire in the removed branch, the push, heartbeat and NIXL connector unit tests pass without reaching it.
• On GB200, the tripwire was not reached in end-to-end push runs with TP1 → TP1, producer TP2 → consumer TP1 and producer TP1 → consumer TP2, and those runs met their expected outputs. These runs used the same push_worker.py change on the [KV Transfer] Connector-agnostic KV checksums for P/D disaggregation #61488 branch.

Details

In push mode, D sends P only two kinds of notifs:

• PUSH_REG, consumed by the writer thread before notifs are forwarded to _get_new_notifs;
• HB:, handled at the top of _get_new_notifs before the counting.

Plain "<req_id>:<n>" notifs to P come from _pending_recv_notifs, which only pull_worker.py fills. P → D completion notifs are attached to the push WRITEs and are counted on D, which is the branch this PR keeps.

The removed code is the producer-side counting copied from pull mode, the transfer_topo assert it needed, and the counter cleanup for finished pushes in get_transfer_results. One edge case changes: on a kv_both instance where one request is both loading and saving, a completion notif for it is now counted as a load instead of a send. Push mode never sends such a notif to the producer side.

Not a duplicate: #55470 also edits _get_new_notifs (it adds a failed-WRITE notif path) but keeps the producer-side branch. Whichever merges second needs a small rebase.

---

Pull Request Checklist

• I used vLLM's /pr-checklist skill. (Mandatory for agents, optional for humans).

• AI assistance was used during the creation of this PR.

• Design Fit: Minimizes impact on core components, reuses existing functionality, and justifies added complexity.

• Testing and Validation: Validates the change and ensures any added tests are meaningful and reliable, with CI coverage or documented CI resource constraints and validation performed outside CI.

• Code Quality and Style: Keeps code and comments clear and concise, and updates relevant documentation and examples.

• Pull Request Contents: Includes a brief summary and relevant links, supports claims with evidence, explains root causes and implementation trade-offs, and follows the contributing guide.

🤖 Generated with Claude Code

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[sawsa307]

cc @njhill

[njhill]

Thanks @sawsa307!

[njhill]

/ci run

[github-actions]

✅ @sawsa307, CI is now available for this PR.

• /ci run starts upstream CI; /amd-ci run starts AMD CI only.
• Run commands test your branch with every commit currently on its upstream target branch. If it is behind, the bot merges the target branch into it first (pull before pushing again). Append --allow-stale to a run command to test an outdated branch as is, at your own risk.
• /ci retry retries failed jobs in the CI build for the current PR head. If the current head has no CI build, it starts a new CI build for the current head containing only jobs that failed in the latest earlier CI build for this PR.
• /amd-ci retry retries failed jobs in AMD CI for the current PR head. Use /amd-ci run when the current head has no AMD CI build.
• /ci cancel cancels scheduled or running CI builds for this PR branch; /amd-ci cancel does the same for AMD CI only.

[github-actions]

✅ Triggered Buildkite CI #95617 for commit 6736f5362cb1.

ℹ️ This PR is 4 commits behind upstream main. Merged upstream main into this branch as 6736f5362cb1 before starting CI. Pull the branch before pushing more changes.