Re: [PATCH] xen-blkback: defer freeing blkif to avoid blocking xenwatch

[Boris Ostrovsky <boris.ostrovsky@oracle.com>]

On 05/12/2014 01:04 PM, Valentin Priescu wrote:
> On Mon, May 12, 2014 at 6:40 PM, Boris Ostrovsky
> <boris.ostrovsky@oracle.com> wrote:
>> On 05/09/2014 07:34 PM, Valentin Priescu wrote:
>>> From: Valentin Priescu <priescuv@amazon.com>
>>>
>>> Currently xenwatch blocks in VBD disconnect, waiting for all pending I/O
>>> requests to finish. If the VBD is attached to a hot-swappable disk, then
>>> xenwatch can hang for a long period of time, stalling other watches.
>>
>> Note that we have a GSoC student (copied here) who is going to be looking at
>> xenwatch's single-threadness (?) over the summer.
>>
> AFAIK, in older kernel versions, you could pass a flag to a registered
> watch that told
> xenwatch to spawn a new kthread when it handles the events. That
> didn't work well
> and it was removed.


Right, it would have to be something coarser than a per-event thread. 
Maybe per-guest.

The goal is more scalability than correctness but as a side-effect it 
would prevent xenwatch from being stuck for everyone.


-boris

> Anyway, whatever solution he comes up with, I think he still needs to
> make similar
> changes in the disconnect path for blkback. Besides, blocking
> uninterruptible for a long
> period of time is not OK.