I found a few other posts which looked to be asking similar questions, but none with satisfactory answers.
I noticed that after the update to Nougat my Nexus 5x has been disconnecting from wifi periodically (possibly it was doing so before, but I'm also seeing a distinct uptick in cellular data usage on days where I should be connected to my home wifi all day). I took a packet caputre and watched the logs on my EdgeRouter Lite during several of these disconnections and noted that it coincides with a restart of radvd. When this happens the router sends out a couple of ICMPv6 RA packets with a router lifetime of 0. Android seems to respond to this by completely disconnecting to the wifi for about 10 seconds and then reconnecting. I'm not sure whether or not Android is behaving correctly here, but I wanted to ask a couple of questions about the router behaviour as it strikes me as less than ideal.
A packet capture on the WAN interface shows that the radvd restarts are happening after the router renews its IPv6 and prefix. The DHCPv6 renew messages have T1=900, and the server responds with the same value, so this happens every 15 minutes, which seems kind of short. As far as I can tell from looking at some source for dhcp6c I found online, this is an arbitrary default:
duration = 1800; /* 30min. XXX: no rationale */
where T1 is set to 0.5 * duration. RFC 3315 states that the client can set T1 and T2 to 0 if it has no preference for the values, so I'm curious if there's a reason why dhcp6c doesn't do that instead of picking some arbitrary value.
After the renewal the IPv6 prefix hasn't changed. However it looks like both dhcpv6-pd-response.pl and dhcpv6-pd-client.pl are going to restart radvd even if the configuration hasn't changed at all, which seems unneccesary. Is there a reason for doing this? If not, it would be nice if the scripts were updated so that radvd won't be restarted if nothing has changed.