MPI多节点执行:HYDU

tech2026-04-06  1

问题:

在多节点环境上运行MPI程序,出现错误:

HYDU_sock_connect (utils/sock/sock.c:145): unable to connect from "2" to "1" (No route to host) main (pm/pmiserv/pmip.c:183): unable to connect to server digital4.ocean at port 42355 (check for firewalls!) HYDU_sock_connect (utils/sock/sock.c:145): unable to connect from "3" to "1" (No route to host) main (pm/pmiserv/pmip.c:183): unable to connect to server digital4.ocean at port 42355 (check for firewalls!)

解决:

1、关闭防火墙

关闭防火墙,即可运行成功

systemctl stop firewalld

2、修改hosts文件

可能是客户端连接不到服务器引起的,在客户端的host文件中添加相关的ip和主机名,如node2中:

10.10.1.12 node1 10.10.1.56 node3

3、防火墙添加富规则

在node1(服务端)的防火墙中添加富规则:

firewall-cmd --add-rich-rule='rule family="ipv4" source address="(node2的ip)" accept' firewall-cmd --add-rich-rule='rule family="ipv4" source address="(node3的ip)" accept'

查看添加的富规则:

firewall-cmd --list-rich-rules

移除添加的富规则:

firewall-cmd --remove-rich-rule='rule family="ipv4" source address="(node2的ip)" accept' firewall-cmd --remove-rich-rule='rule family="ipv4" source address="(node3的ip)" accept'

 

最新回复(0)