[PR #722] [CLOSED] [Bugfix]Fix 'invalid argument' error when setting QP state to INIT vi… #1771

New issue

Closed

opened 2026-10-29 21:30:12 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-29 21:30:12 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/kvcache-ai/Mooncake/pull/722
Author: @hezhiqian01
Created: 8/6/2025
Status: ❌ Closed

Base: main ← Head: dev

📝 Commits (2)

2ebbce6 [Bugfix]Fix 'invalid argument' error when setting QP state to INIT via ibv_modify_qp on some rdma devices.
0098b54 Implement QP state query before resetting

📊 Changes

1 file changed (+30 additions, -10 deletions)

View changed files

📝 mooncake-transfer-engine/src/transport/rdma_transport/rdma_endpoint.cpp (+30 -10)

📄 Description

This PR fix the issue as below:
When using the Intel E810 RDMA NIC, calling the ibv_modify_qp function to change the QP state to INIT results in an 'Invalid argument (22)' error.

The error might be caused by the QP already being in the RESET state, and setting it to RESET again could lead to an internal state inconsistency.

The solution is to remove the step that sets the QP to RESET.

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/kvcache-ai/Mooncake/pull/722 **Author:** [@hezhiqian01](https://github.com/hezhiqian01) **Created:** 8/6/2025 **Status:** ❌ Closed **Base:** `main` ← **Head:** `dev` --- ### 📝 Commits (2) - [`2ebbce6`](https://github.com/kvcache-ai/Mooncake/commit/2ebbce6e70b0d0a08d8697d73cfd89b7df7aca28) [Bugfix]Fix 'invalid argument' error when setting QP state to INIT via ibv_modify_qp on some rdma devices. - [`0098b54`](https://github.com/kvcache-ai/Mooncake/commit/0098b54cbfdeb0d8ea2aa3c0aed77d281d2fe601) Implement QP state query before resetting ### 📊 Changes **1 file changed** (+30 additions, -10 deletions) <details> <summary>View changed files</summary> 📝 `mooncake-transfer-engine/src/transport/rdma_transport/rdma_endpoint.cpp` (+30 -10) </details> ### 📄 Description This PR fix the issue as below: When using the Intel E810 RDMA NIC, calling the **ibv_modify_qp** function to change the QP state to **INIT** results in an **'Invalid argument (22)'** error. The error might be caused by the QP already being in the **RESET** state, and setting it to **RESET** again could lead to an internal state inconsistency. The solution is to remove the step that sets the QP to **RESET**. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-29 21:30:12 +00:00