On Fri, 17 Dec 2004, Michael Parker stated: > I don't think it's so much a dynamic-IP thing as a NAT thing. Without > lots of activity (we can go a day or so at a time without a commit) > the slaves just stop talking to the master and with the connection > gone behind the NAT the master can't find them.
With most NAT systems worth their salt, this timeout is configurable. For example, on Linux systems using netfilter (2.4 and 2.6 kernels), writing a time in seconds to the file /proc/sys/net/ipv4/netfilter/ip_conntrack_tcp_timeout_established will fix it for IPv4 TCP streams: the default value is 432000 (five days). If you knock it up *too* far, you might see scalability problems, so don't multiply it by more than, say, ten. If SA goes fifty days without a commit something is very wrong. :) -- `The sword we forged has turned upon us Only now, at the end of all things do we see The lamp-bearer dies; only the lamp burns on.'
