0 = epoll_wait(_, _, _, 0) loop takes 100% cpu about ndhc

[niklata]

Thanks for the report! Something unexpected is happening with the ARP or DHCP timeouts. Please apply the following debug patch and provide a sample of some of the prints:

http://dpaste.com/3S3JX3K

[niklata]

I've just pushed a commit that adds a failsafe to prevent the rapid-busy spin (it will force ndhc to wake up every 10s instead).

I'd still be very interested in the output from the debugging patch I posted in the previous comment, as I'd like to correct any issue that may have caused the bad behavior you've encountered.

[thypon]

Thanks!

http://dpaste.com/362SXWS

[niklata]

I think these problems are being caused by the long lease time on your network (which is perfectly valid -- it's an issue in ndhc).

I've pushed some commits that should clean up some possible truncation/overflow issues related to time values being represented as long long or unsigned long long being truncated down to epoll_wait's int-sized timeout.

Please let me know if the current ndhc master still has the problem. Thank you for the strace and debug print trace; they've been very helpful!

[thypon]

It's working with the latest version. Thank you.