perfSONAR User Q&A and Other Discussion

[Andrew Lake <>]

Hi Trond,

I think there is indeed an issue with cassandra’s default config and IPv6-only hosts. Can you try the following:

1. Open  /etc/cassandra/conf/cassandra-env.sh

2. Find and comment-out the line JVM_OPTS=“$JVM_OPTS -Djava.net.preferIPv4Stack=true”

3. Save the changes.

4. Restart cassandra with systemctl restart cassandra

Might want to double check there are no other references to -Djava.net.preferIPv4Stack=true in that file, mine just had the one. Also since you have been changing /etc/hosts, make sure localhost still resolves to ::1. Its possible we may need to play with the Java opts some more, but hopefully that is all you need. Seems to be a known issue with cassandra. We aren’t the authors of the cassandra RPM, but if that indeed fixes your issue we can look at having one of our packages correct that issue on install (it won’t be the first time we have had to some hand-holding for cassandra’s RPM).

Also, as an aside, there should be no requirement to disable your firewall. I think you said that didn’t work anyways but all communication with cassandra is done on localhost and should not be affected by the host firewall.

Thanks,

Andy

On October 6, 2017 at 7:37:39 AM, Trond Endrestøl () wrote:

On Fri, 6 Oct 2017 13:22+0200, Trond Endrestøl wrote:

> On Fri, 6 Oct 2017 13:18+0200, Trond Endrestøl wrote:
>
> > On Fri, 6 Oct 2017 12:59+0200, Trond Endrestøl wrote:
> >
> > > On Fri, 6 Oct 2017 09:48-0000, Kashif Mohammad wrote:
> > >
> > > > I was also having the same problem with Cassandra not talking to
> > > > other components. It looks like that it is a problem with firewall.
> > > > I disabled firewall and its working now.
> > >
> > > Well, yes and no. Running these commands made it eventually work on an
> > > isolated system with IPv4 and IPv6 addresses assigned to the NIC:
> > >
> > > systemctl stop firewalld
> > > systemctl disable firewalld
> > > shutdown -r now Preventive reboot after disabling firewalld.;sleep 1;logout
> > >
> > > Systems running with IPv6 only, not so much.
> > >
> > > Further testing shows that the web browser determines success or
> > > failure.
> > >
> > > Access the web UI by specifying the system's IPv4 address, numerically
> > > or as a domain name, and everything works; access the web UI by
> > > specifying the system's IPv6 address, numerically or as a domain name,
> > > and it all breaks. And happens in 2017?
> > >
> > > I was hoping to avoid assigning IPv4 addresses to my perfSONAR nodes ...
> >
> > A crude workaround is to:
> >
> > 1. Assign a private IPv4 address to the NIC, say 192.0.2.3/24 (TEST-NET-1).
> > 2. Don't bother adding a default gateway for IPv4.
> > 3. Edit /etc/hosts, specifying the IPv6 address prior to the IPv4 address:
> >
> > 2001:db8:dead:beef::3 yournode.fqdn yournode
> > 192.0.2.3 yournode.fqdn yournode
> >
> > Each node's internals are now able to translate yournode.fqdn to an
> > IPv4 address and resume internal comms. Dear God, make it stop!
>
> It's even better adding "yournode.fqdn yournode" to the loopback
> addresses in /etc/hosts.

Sorry guys, the latter is just bad advice. My bad. I stand by my first
suggestion of adding a dummy IPv4 address and making the hostname
resolvable via /etc/hosts.

Users running IPv4 only or dual-stack will only know the difference
once they turn off IPv4.

--
Trond.