mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chun-Hung Hsiao <chhs...@apache.org>
Subject Re: Review Request 70213: Added the `VolumeManager` interface to manage CSI volumes.
Date Tue, 02 Apr 2019 05:29:09 GMT


> On March 19, 2019, 2:42 p.m., Benjamin Bannier wrote:
> > src/csi/volume_manager.hpp
> > Lines 74-75 (patched)
> > <https://reviews.apache.org/r/70213/diff/2/?file=2132465#file2132465line74>
> >
> >     Do we have a chance here to translate from CSI semantics to our own semantics
instead of leaking the CSI protocol? I think not using a magic value of `0` would be nice.
> >     
> >     Seems like we could either return a `Failure` if `GET_CAPACITY` is not supported
or some `Option<Bytes>::none()`. This would likely require capability knowledge.
> 
> Chun-Hung Hsiao wrote:
>     I'm actually debating if we should expose CSI capabilities. Given that the capabilities
might change, do you think it's a good idea to expose that?
> 
> Benjamin Bannier wrote:
>     Wouldn't they'd only be exposed in the documentation? We could replace that part
with _can permanently not be determined_ so we can use failed futures for retryable errors
and ready nones for the unsupported providers.
> 
> Chun-Hung Hsiao wrote:
>     The caller, i.e., SLRP, won't know why it fails. If the plugin doesn't support `GET_CAPACITY`,
SLRP should still work with handling pre-existing volumes. So we either needs to expose the
capability to SLRP, or don't fail here.

Dropping this issue. See later patches in this chain for uses.


> On March 19, 2019, 2:42 p.m., Benjamin Bannier wrote:
> > src/csi/volume_manager.hpp
> > Lines 115-117 (patched)
> > <https://reviews.apache.org/r/70213/diff/2/?file=2132465#file2132465line115>
> >
> >     Am I reading correctly that the return boolean is not dynamic and just maps
onto whether the controller has `CREATE_DELETE_VOLUME` or not?
> 
> Chun-Hung Hsiao wrote:
>     Yes you're right lol. This roughly matches the semantics defined here: https://github.com/apache/mesos/blob/master/include/mesos/mesos.proto#L2102
>     
>     Again, not sure if we should expose the CSI capabilities instead. WDYT?
> 
> Benjamin Bannier wrote:
>     It would be nice to see how this would be used. Do you want to distinguish transient
from permanent failures, or would a `Future<Nothing>` work just as well?

Dropping this issue. See later patches in this chain for uses.


- Chun-Hung


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/70213/#review213806
-----------------------------------------------------------


On March 27, 2019, 5:56 a.m., Chun-Hung Hsiao wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/70213/
> -----------------------------------------------------------
> 
> (Updated March 27, 2019, 5:56 a.m.)
> 
> 
> Review request for mesos, Benjamin Bannier, Jie Yu, and Jan Schlicht.
> 
> 
> Bugs: MESOS-9622
>     https://issues.apache.org/jira/browse/MESOS-9622
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> `VolumeManager` is a wrapper for SLRP to use v0 and v1 CSI plugins
> polymorphically. It will be managing volume lifecycles and checkpoints
> and making the actual CSI calls for SLRP and SERP in the future.
> 
> 
> Diffs
> -----
> 
>   src/CMakeLists.txt 3397c3b1d4e8a7900b2e5f870679cc7aa30b4be2 
>   src/Makefile.am bcafe48b2105575371464a29783bc6f3f1c2cf8d 
>   src/csi/volume_manager.hpp PRE-CREATION 
>   src/csi/volume_manager.cpp PRE-CREATION 
> 
> 
> Diff: https://reviews.apache.org/r/70213/diff/3/
> 
> 
> Testing
> -------
> 
> make check
> 
> 
> Thanks,
> 
> Chun-Hung Hsiao
> 
>


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