Hi folks, Need some help on using Liburing 0.6. I am trying io_uring in EC2 environment (kernel version 5.4.95-47.164.amzn2int.x86_64, github https://github.com/amazonlinux/linux/tree/kernel-5.4.95-47.164.amzn2int). The test code logic is simple. I use one uring and two threads. One thread is continuously submitting nonblocking recvmsg from a TCP socket struct msghdr msg; struct io_uring_sqe* sqe = nullptr; memset(&msg, 0, sizeof(msg)); msg.msg_iov = iov; msg.msg_iovlen = count; sqe = io_uring_get_sqe(m_ring); io_uring_prep_recvmsg(sqe, sockfd, &msg, 0); sqe->user_data = (uint64_t) context; int ret = io_uring_submit(m_ring); if (ret < 0) { GLOG(ERRO, "failed to submit io: %d", ret); } Another thread is continuously polling by calling io_uring_wait_cqes. struct io_uring_cqe *cqe = nullptr; while (!m_stopped) { int ret = ::io_uring_wait_cqes(m_ring, &cqe, 0, nullptr, nullptr); if (ret != 0) { if (ret != -EAGAIN) { GLOG(ERRO, "cqe wait failed: %d", ret); } continue; } ::io_uring_cqe_seen(m_ring, cqe); void *ctx = ::io_uring_cqe_get_data(cqe); if (ctx == nullptr || cqe->res == 0) { continue; } if (cqe->res != -EAGAIN) { // For some reason, randomly I will get EMSGSIZE or EINVAL. Retry on the same sockfd will succeed. GLOG(ERRO, "cqe: %p %d %llu", cqe, cqe->res, cqe->user_data); } int err = (cqe->res > 0) ? 0 : -cqe->res; int transferred = (err == 0) ? cqe->res : 0; } Would like to know what could be the reason for the random EMSGSIZE or EINVAL error (see comments above). Thanks a lot. Yihao
