Container Storage Interface (CSI)

[Jie Yu]

Hi,

Container Storage Interface (CSI)
is a joint work between
container orchestrators (Mesos, Kubernetes, Docker and Cloud Foundry)
trying to standardize the interface between container orchestrator and
storage platforms.

The goal is to let storage platforms just write one plugin that works for
all container orchestrators.

The initial discussion is captured in this doc:
https://docs.google.com/document/d/1JMNVNP-ZHz8cGlnqckOnpJmHF-DNY7IYP-
Di7iuVhQI/edit?usp=sharing

Feel free to comment on that doc if you have any thoughts! This is the same
doc shared in the k8s and Docker community.

The relevant Mesos work to integrate CSI is captured in the Resource
Provider  design doc:
https://docs.google.com/document/d/125YWqg_5BB5OY9a6M7LZcby5RSqBwo2PZzpVL
uxYXh4/edit?usp=sharing

- Jie

Re: Detect when mesos agent needs work directory cleanup

[Tomek Janiszewski]

1. Cleanup in required when agent configuration is not compatible with
previous version. This mean task runtime change. This occurs when:
resources, isolators, attributes or containerizers change.
2. IMO it's OK. When you reboot node all its tasks are gone so you won't
lose anything. Node will be registered as a new node.

śr., 22.03.2017, 15:55 użytkownik Michele Bertasi <
michele.bert...@brightcomputing.com> napisał:

> Hi everybody,
>
> I'm having troubles with the cleanup of the mesos agent work directory.
> When we change some configuration parameters to the agent (e.g. the
> --resource flag), the agent refuses to start and asks the removal of
> */var/lib/mesos-agent/meta/slaves/latest* in the logs.
>
> Since we are automating these changes, we would like to know when to
> perform this cleanup programmatically.
>
> Right now we are doing that *after every single change*, but that's not
> ideal, since every time mesos master assigns a new ID to the slave. It
> seems unnecessary to me.
>
> My questions are:
>
>- is there a way to know when this cleanup is needed (without looking
>at the logs)?
>- do we have problems if we clean up the work directory every time the
>node boots (so the configuration might or might not have been changed at
>all)?
>
>
> Thanks, kind regards,
>
> Mike
>