have a bond interface thats failing on cable pull test intermittently.

bond0
eth6
eth7

master slave mode. eth6 active by default.

testing ping from other server in same subnet

1. use ifenslave to detach eth6, eth7 becomes active, bond is ok. reverse detach, ifenslave change active eth, bond is fine.
2. use if/ifdown to bring cards up and down , altering active card. bond is fine.
3. use ifconfig to bring cards up and down, cards dont detach but are marked as "down". eth6 is active, ifconfig eth6 down, eth7 is then marked as active, Bond fails, ping hangs. Use "service network restart" to re-enable. ping starts again.
4. cable pull tests (used different cables, different switches). same behaviour as ifconfig tests, eth6 is active, pull eth6, ping hangs. "service network restart" to start it again.

I tried a tcpdump on eth6 when the card was down and the ping started again!! wtf? running tcpdump when pulling cable and ping stays running.

any ideas?

A totally wild guess on my part: hardware conflicts between your network cards, or between hardware on the motherboard itself. What does "lspci -vv" (without the quotes) say? Also, do you get any error messages when you first boot your system?

thanks for reply xode.

we had thought this, we're running RH oel 5.9, the bonds were working fine with 5.7, we've upgraded to 10gb cards as well. All cards working, its jsut the bond that has the problem. but have done some more config testing since which is pointing at a more probable root cause.

We've started using jumboframes so have mtu=9000, the problem ONLY occurs when mtu=9000, if i set the mtu back to 1500, the bond survives any cable pull. Some incompatibility between the cards, the OS, the switch when mtu=9000? theres a couple of references to this online if you search for "linux jumboframes bond not working.