[
https://issues.apache.org/jira/browse/MESOS-4113?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15070045#comment-15070045
]
Anthony Scalisi commented on MESOS-4113:
----------------------------------------
Sorry for the delay here, on PST timezone.
[~jieyu] unfortunately, I'm not familiar with how Marathon does health checks
(I don't speak Java at all). I opened an issue on Github:
https://github.com/mesosphere/marathon/issues/2870#issuecomment-166848308
so hoping to get some leads over there.
The only thing for sure is that 0.26.0 changed what task IP is reporting and
Marathon 0.13.0 (or 0.14 RC for that matter) grabs the IP reported now
(defaulting to the internal Docker one) which make all the health checks
failing.
I saw mention of Mesos-DNS and I would like to add that even though Marathon
health checks stopped working, our other discovery mechanisms didn't:
in the case of Mesos-consul for example
(https://github.com/CiscoCloud/mesos-consul), you can see here
(https://github.com/CiscoCloud/mesos-consul/blob/aac6c2828a46c3a54efe1fbc41003dbcd69a6a40/main.go#L120-L123)
that it has a flag to specify how the task IP is registered in Consul. We had
it set to "host,mesos,docker,netinfo", so pretty much registering the task IP
as the Mesos slave IP so everything worked fine on that end on v0.26.0.
I'd like to mention that Mesos-DNS has the same kind of flag:
https://github.com/mesosphere/mesos-dns/blob/9a8aa106a05339c79fb189d435c68f64e876414c/records/config.go#L49
I also multiple issues relevant:
- https://github.com/mesosphere/mesos-dns/issues/332
- https://github.com/mesosphere/mesos-dns/issues/369
Unfortunately, I don't know enough of the internal of Mesos itself so just
coming from an Ops point of view (and a long night of nuked containers
unfortunately).
I'd like to mention also that I don't have enough bandwidth to be able to
support Mesos compiled from sources as I rely on stable and testing packages
from Mesosphere repositories.
If it is any help, I'm available on IRC in the #mesos and #marathon channels
and more than happy to help debug things.
> Docker Executor should not set container IP during bridged mode
> ---------------------------------------------------------------
>
> Key: MESOS-4113
> URL: https://issues.apache.org/jira/browse/MESOS-4113
> Project: Mesos
> Issue Type: Bug
> Components: docker
> Affects Versions: 0.25.0, 0.26.0
> Reporter: Sargun Dhillon
> Assignee: Artem Harutyunyan
> Labels: mesosphere
>
> The docker executor currently sets the IP address of the container into
> ContainerStatus.NetworkInfo.IPAddresses. This isn't a good thing, because
> during bridged mode execution, it makes it so that that IP address is
> useless, since it's behind the Docker NAT. I would like a flag that disables
> filling the IP address in, and allows it to fall back to the agent IP.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
