Description of problem: I must do a sleep of roughly 10 seconds before letting the iscsi login to all portals, because only 4 out of 8 targets are logged into. The first 4 targets that are logged into are the ones setup for eth0, the next 4 get a connection timed out unless I sleep for 10 seconds or so. This became a problem after patching recently and getting the updated initscript package. Version-Release number of selected component (if applicable): How reproducible: EMC-CX320 Configure eth0 and eth1 for the 8 iscsi targets I also have a 4 port nic with 2 bonds setup, bond0 - balance-alb (public interface) and bond1 - 802.3ad (cluster interconnect) Steps to Reproduce: 1. Update your host with this set of packages Oct 16 11:28:19 Updated: krb5-libs - 1.6.1-25.el5_2.1.x86_64 Oct 16 11:28:20 Updated: initscripts - 8.45.19.1.EL-1.x86_64 Oct 16 11:28:21 Updated: nss - 3.12.1.1-1.el5.x86_64 Oct 16 11:28:21 Updated: cups-libs - 1:1.2.4-11.18.el5_2.2.x86_64 Oct 16 11:28:21 Updated: OpenIPMI-libs - 2.0.6-6.el5_2.2.x86_64 Oct 16 11:28:22 Updated: krb5-libs - 1.6.1-25.el5_2.1.i386 Oct 16 11:28:28 Updated: nss - 3.12.1.1-1.el5.i386 Oct 16 11:28:28 Updated: pam_krb5 - 2.2.14-1.el5_2.1.x86_64 Oct 16 11:28:28 Updated: nss-tools - 3.12.1.1-1.el5.x86_64 Oct 16 11:28:29 Updated: OpenIPMI - 2.0.6-6.el5_2.2.x86_64 Oct 16 11:28:30 Updated: cups - 1:1.2.4-11.18.el5_2.2.x86_64 Oct 16 11:28:30 Updated: dhcpv6-client - 1.0.10-4.el5_2.3.x86_64 Oct 16 11:28:30 Updated: krb5-workstation - 1.6.1-25.el5_2.1.x86_64 Oct 16 11:28:30 Updated: pam_krb5 - 2.2.14-1.el5_2.1.i386 Oct 16 11:28:31 Updated: kernel-headers - 2.6.18-92.1.13.el5.x86_64 Oct 16 11:28:32 Updated: tzdata - 2008f-3.el5.noarch Oct 16 11:28:49 Installed: kernel - 2.6.18-92.1.13.el5.x86_64 2. Reboot 3. Check connectivity status on EMC Navisphere, 4 of 8 will be logged in Actual results: 4 out of 8 initiators logged in Expected results: 8 out of 8 initiators logged in Additional info:
Are you using static IPs? What sort of switching infrastructure do you have (is STP enabled?)
Yes, using static IPs. I believe spanning tree is enabled even though the network admin says portfast should be enabled. Reason being, when I pull one of the interfaces out of the bond and reconnect it, I have to wait roughly 30 seconds before I can start talking on that interface. Arwin
Given that, there's very little you can do outside of setting NETWORKDELAY in /etc/sysconfig/network - there's no programmatic way for the init scripts to know that the switch is actually passing traffic.