So are the two nics ultimately intended to be doing some sort of bonding? For reference, I was thinking of seeing about making *either* mac address work and get the same address, under the presumption that firmware boot activity would give way to the appropriate bonding in the OS.
One interesting thing that acts like you describe: # nodediscover list -t pxe-client -o node -f node,mac,switch,port | head Node| Mac| Switch| Port -----|------------------|------------|---------------- n888| 08:94:ef:60:44:1e| 172.16.2.76| Ethernet1/6 n886| 08:94:ef:60:44:28| 172.16.2.76| Ethernet1/4 n973| 3c:fd:fe:4e:48:2c| 172.16.2.65| Ethernet31 Confluent will detect mac addresses according to what actually attempts PXE boot, without getting into linux. This will cause it to ignore the non-pxe mac. Once gathered, using a command to generate a macs.csv for xCAT: $ confluent2xcat d3-d5 -m macs.csv $ cat macs.csv #node,mac d3,08:94:ef:3f:e1:32 d4,08:94:ef:41:01:f0 d5,08:94:ef:50:be:c2 -----Original Message----- From: Thomas HUMMEL <thomas.hum...@pasteur.fr> Sent: Thursday, June 6, 2019 2:33 PM To: xcat-user@lists.sourceforge.net Subject: [External] [xcat-user] Multi nic in same subnet Hello, Some time ago both in this list and on gitter I discussed in details about how to deal with stateless discovery and provisionning nodes which have 2 nics in the same subnet. No clear conclusion had been found at this time and my last discussion on gitter had been interrupted by someone on you team gone in vacation ;-) So, to come up again with the subjet, hopefully sorting it out once for all : Basically, where nodes - have 2 nics eth0/1G and eth2/10G connected on the same subnet/vlan - xCAT MN has only one nic eth0 on this subnet - only switched based discovery regarding eth2 has been configured - only one dns name and one ip has been configured for the node how can I make sure that, once discovered the node will be provisionned/booted on eth2 ? The reasoning behind this could be : - only eth2 could be on the subnet but a switch misconfiguration could lead to eth0 being also on the same subnet (so we want to make sure ultimately eth2 only will be configured no matter what nic was used for discovery) - or we could want to discover / provision via eth0 but make sure once booted, the faster eth2 would be used In what I tested at the time, the following points surfaced : - in some (all ?) scenarii, provisionning followed the bios order, ending up in the node getting the correct (eth2's) ip but on eth0 - there was some behavior related to if a <node>-eth2 dns entry would be present or not which I didn't fully understand - there could be some kind of fallback to the MTMS discovery method (if node still in discoverydata) - primarynic was deprecated - installnic attribute purpose or use was not clear - nicips seemed to serve another purpose than mine (not sure about that) I forgot the details hence the way I'm asking the question now : how, if possible, can I achieve this (basically beeing sure only eth2 will get the node's ip, no matter which nic was used for discovery), using the latest xCAT version (on x86_64 arch) ? Thanks -- Thomas HUMMEL _______________________________________________ xCAT-user mailing list xCAT-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xcat-user _______________________________________________ xCAT-user mailing list xCAT-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xcat-user
