Hide Forgot
Description of problem: Sometimes, ixgbe nics will take up to 40 seconds before they are all up and os-net-config will run before all nics are up and will fail noticeably when trying to map nicX to pXpX. Version-Release number of selected component (if applicable): How reproducible: Random Steps to Reproduce: 1. Deploy an overcloud using more than 2 ixgbe nics 2. 3. Actual results: Fails at network configuration Expected results: succeeds Additional info: Patch provided by the customer: --- 20-os-net-config.orig 2016-03-17 18:47:05.634188672 -0400 +++ 20-os-net-config 2016-03-17 18:47:19.111195755 -0400 @@ -51,6 +51,7 @@ NET_CONFIG=$(os-apply-config --key os_net_config --type raw --key-default '') if [ -n "$NET_CONFIG" ]; then + if [[ $(awk '{printf "%d",$1}' /proc/uptime) -lt 600 ]];then echo "nic delay fix, wait for 60 sec";sleep 60; fi os-net-config -c /etc/os-net-config/config.json -v --detailed-exit-codes RETVAL=$? if [[ $RETVAL == 2 ]]; then
This bug did not make the OSP 8.0 release. It is being deferred to OSP 10.
(In reply to David Hill from comment #0) > Description of problem: > Sometimes, ixgbe nics will take up to 40 seconds before they are all up and > os-net-config will run before all nics are up and will fail noticeably when > trying to map nicX to pXpX. I strongly suspect that this is due to switch configuration. If spanning-tree is enabled, and the host ports are not configured to enable fast forwarding, then it can take up to a minute before STP will start forwarding traffic to the port. The customer should have better results if they configure "portfast" or "edge port" or "host port" on the ports attached directly to servers (name will vary depending on switch vendor). I also don't believe that this bug would affect newer versions of OSP-Director. Considering that this is an old ticket, I'm going to close this ticket for now. Please feel free to reopen if the root cause analysis is incorrect.