> On March 26, 2016, 12:37 a.m., Jie Yu wrote:
> > src/slave/slave.cpp, line 2375
> > <https://reviews.apache.org/r/45121/diff/1/?file=1308847#file1308847line2375>
> >
> >     I would suggest we don't use CHECK here. We can just LOG(ERROR) if the 
> > deletion fails. Given the TODO above, we could leak disk space anyway.
> 
> Neil Conway wrote:
>     I used `CHECK_SOME` because we use the same error-handling strategy when 
> creating the persistent volume fails, or if the slave fails to checkpoint 
> resources to disk successfully. Are you sure we want to handle `rmdir` errors 
> differently?
> 
> Jie Yu wrote:
>     We cannot proceed when creating persistent volume fails because the 
> subsequent 'mount' will fail.
>     
>     For the deletion case, if it fails, we just leak some data in the volume. 
> The subsequent operation will not fail. Since we'll be leaking data anyway 
> (the slave crash case), I suggest that we don't use CHECK_SOME here.
> 
> Neil Conway wrote:
>     True, although for MOUNT disks we might expose data from the previous 
> volume to a newly-created volume on the same disk. Overall I agree that using 
> `LOG(ERROR)` is probably best for now.

For MOUNT, I think we have the same problem if slave crashes. We might want to 
do a clean during creation as well.


- Jie


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/45121/#review125492
-----------------------------------------------------------


On March 27, 2016, 6:08 p.m., Neil Conway wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/45121/
> -----------------------------------------------------------
> 
> (Updated March 27, 2016, 6:08 p.m.)
> 
> 
> Review request for mesos, Jie Yu and Joris Van Remoortere.
> 
> 
> Bugs: MESOS-2408
>     https://issues.apache.org/jira/browse/MESOS-2408
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Prior to this commit, destroying a persistent volume would remove
> the Mesos-level metadata about the volume, but wouldn't destroy
> any of the volume's filesystem content. We now remove the volume
> from the slave's filesystem, essentially via "rm -r".
> 
> 
> Diffs
> -----
> 
>   src/slave/slave.cpp f383605a52f31d7b805ad6153adc409dcb40f83a 
> 
> Diff: https://reviews.apache.org/r/45121/diff/
> 
> 
> Testing
> -------
> 
> make check
> 
> 
> Thanks,
> 
> Neil Conway
> 
>

Reply via email to