mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Avinash sridharan <avin...@mesosphere.io>
Subject Re: Review Request 51643: Modified network file setup in `network/cni` isolator.
Date Wed, 07 Sep 2016 19:05:23 GMT


> On Sept. 7, 2016, 1 a.m., Joseph Wu wrote:
> > src/slave/containerizer/mesos/isolators/network/cni/cni.cpp, lines 1540-1542
> > <https://reviews.apache.org/r/51643/diff/2/?file=1491981#file1491981line1540>
> >
> >     Given that this case is very specific for systems without a dependency on `/etc/hosts`
or `/etc/hostname`, would it make sense to create the files in the agent's work directory?
 i.e. to avoid polluting the host?
> 
> Avinash sridharan wrote:
>     We create these files in the agents work directory as well (a directory specific
to the `network/cni` isolator). The reason we create these files specifically in the host
fs is for the `executor` and containers that don't have their own images. Reason being that
both these entities see the host file system, albiet with a different mount name space. Also,
since the hostname and IP address is different from the hosts IP and hostname, we need to
give them a different view of the /etc/hosts and /etc/hostname files. We do this by bind mounting
the files from the containers work dir on to these target mount points on the host file system.

>     
>     We do the same operation below for the containers rootfs as well. However, in case
of the containers rootfs we are guaranteed that these mount points exist (if not we can create
them).

Modified the comments to clarify this point.


- Avinash


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/51643/#review147956
-----------------------------------------------------------


On Sept. 7, 2016, 7:02 p.m., Avinash sridharan wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/51643/
> -----------------------------------------------------------
> 
> (Updated Sept. 7, 2016, 7:02 p.m.)
> 
> 
> Review request for mesos, Gilbert Song, Jie Yu, Joseph Wu, and Qian Zhang.
> 
> 
> Bugs: MESOS-6052
>     https://issues.apache.org/jira/browse/MESOS-6052
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> In case /etc/hosts and /etc/hostname files are not present in the host
> filesystem, we were ignoring these files and assuming that they would
> not be required by the executor when it is launched in a new network
> namespace. This assumption is incorrect, since the executor needs
> /etc/hosts in the new network namespace to resolve its hostname.
> Hence, we are explicitly creating these files in the host file system
> in case they are not present, so that containers /etc/hosts and
> /etc/hostname can be mounted on these mount points. This solves the
> problem in distributions such as CoreOS that don't have /etc/hosts in
> their host filesystem.
> 
> 
> Diffs
> -----
> 
>   src/slave/containerizer/mesos/isolators/network/cni/cni.cpp d17a45fe17bb74cbf9ae421dc8a492e5dc5f1a00

> 
> Diff: https://reviews.apache.org/r/51643/diff/
> 
> 
> Testing
> -------
> 
> make check
> 
> Also, build an open DC/OS image with this patch to run on CoreOS and tried launch a unified
containerizer through Marathon using dcos-cli to test this patch on CoreOS.
> 
> 
> Thanks,
> 
> Avinash sridharan
> 
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message