Bug #6380
closed
Monitor tries to bind on IPv6 address while not available yet after boot
Added by Wido den Hollander over 10 years ago.
Updated over 8 years ago.
Description
Not sure yet what the problem is, but creating a issue so it's documented.
When using IPv6 only and binding the monitor on a IPv6 address the monitor doesn't come online after boot.
This is what the logs show:
2013-09-23 11:46:08.037723 7f7110b13780 -1 accepter.accepter.bind unable to bind to [2a00:XXX:XXX:200::6789:2]:6789: Cannot assign requested address
I think this has something to do with DAD (Duplicate Address Detection) of IPv6 where it takes a couple of seconds for the address to actually be available on the nic.
I'll try to find the root cause.
We've seen similar issues on fedora with ipv4; there it was just a problem with the init/systemd ordering (networkmanager vs normal network service). This is ubuntu I assume? I would expect the ip to be fully configured by the time the network service finishes and/or we get to runlevel 2 (where ceph-all starts)...
Sage Weil wrote:
We've seen similar issues on fedora with ipv4; there it was just a problem with the init/systemd ordering (networkmanager vs normal network service). This is ubuntu I assume? I would expect the ip to be fully configured by the time the network service finishes and/or we get to runlevel 2 (where ceph-all starts)...
This is indeed Ubuntu, but with IPv6 there is something different.
When you add a IPv6 address to the NIC it will be attached, but it will take about 3 to 5 seconds before it to become functional (and bind-able) due do the Duplicate Address Detection the kernel performs.
So the network service exits and the booting continues and within that 3 to 5 seconds the monitor starts and tries to bind.
That is my educated guess for now.
So I did a small test:
[mon]
pre start command = "sleep 3"
That works, the mon now start nicely and binds to the IPv6 address.
So it seems that when the monitor tries to start the IPv6 address isn't available yet and this short sleep fixes that. 1 second might work, but I only tried with 3 seconds.
- Status changed from New to Resolved
Also available in: Atom
PDF