RabbitMQ-specific IPv6 sysctl flags were dropped when migrating to the kernel-baremetal-ansible service

[Damien Ciabrini]

When kernel-baremetal-ansible.yaml was introduced, various sysctl settings dealing with IPv6 were not ported from kernel-baremetal-puppet.yaml. This had at least one side effect of breaking the fix we did in lp#1780065 [1]:

              # always allow ipv6 on loopback for rabbitmq, see +bug/1780065
              net.ipv6.conf.lo.disable_ipv6:
                value: 0

Without that setting, the Erlang VM cannot start on a host where IPv6 is entirely disabled. This in turn prevents rabbitmq from starting and make the deployment fail.

[1] https://bugs.launchpad.net/tripleo/+bug/1780065

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.opendev.org/731241

[OpenStack Infra (hudson-openstack)]

Change abandoned by Damien Ciabrini (<email address hidden>) on branch: master
Review: https://review.opendev.org/731241
Reason: after discussion with ysandeep, that patch is not needed, the original issue was due to downstream nodepool customization

[Damien Ciabrini (dciabrin)]

After further investigation, it seems the error seen on that environment was not a due to a t-h-t regression but to a misconfiguration. Quoting ysandeep:

"
 I don't think we need that anymore.. In our case that was due to downstream nodepool customization. 99-tripleo.conf is correctly poppulated by tripleo ansible.
"

I'm closing the bug accordingly.

[Takashi Kajinami (kajinamit)]

I think this problem is still valid. I noticed that net.ipv6.conf.lo.disabled_ipv6=0 comes BEFORE net.ipv6.conf.all.disabled_ipv6=1, which ends up in 99-sysctl.conf when KernelDisableIPv6, which results in net.ipv6.conf.lo.disabled_ipv6=1 actually.

However seems like erlang vm can start even with this setup, since we allow net.ipv6.ip_nonlocal_bind by default recently.

I think we can reproduced the issue by setting both KernelDisableIPv6: 1 and KernelIpNonLocalBind: 0 in the template.

[Takashi Kajinami (kajinamit)]

correction

I noticed that net.ipv6.conf.lo.disabled_ipv6=0 comes BEFORE net.ipv6.conf.all.disabled_ipv6=1, which ends up in 99-sysctl.conf when KernelDisableIPv6

I meant to say;

I noticed that net.ipv6.conf.lo.disabled_ipv6=0 comes BEFORE net.ipv6.conf.all.disabled_ipv6=1 in 99-sysctl.conf when KernelDisableIPv6, which results in ...

[Damien Ciabrini (dciabrin)]

Thanks for the heads up Takashi, I wonder if we should reopen this bug and correct subject and description, or it we should track your issue in a dedicated lp. Even if the end-result is the same, the cause looks sufficiently different that I would prefer doing the latter.