Re: [PATCH] pnfs: Kick a pnfs_layoutcommit_inode on recall

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Tue, Aug 26, 2014 at 2:19 PM, Boaz Harrosh <boaz@xxxxxxxxxxxxx> wrote:
> On 08/26/2014 08:54 PM, Trond Myklebust wrote:
>> On Tue, Aug 26, 2014 at 1:06 PM, Boaz Harrosh <boaz@xxxxxxxxxxxxx> wrote:
>
>>
>> The deadlock occurs _if_ the above layout commit  is unable to get a
>> slot. You can't guarantee that it will, because the slot table is a
>> finite resource and it can be exhausted
>
> Yes all I ever seen is 1 slot in any of the clients/servers I've
> seen so I assume 1 slot ever
>
>> if you allow fore channel
>> calls to trigger synchronous recalls on the back channel
>
> Beep! but this is exactly what I'm trying to say. The STD specifically
> forbids that. The server is not allowed to wait here, it must return
> imitatively, with an error that frees the slot and then later issue the
> RECALL.
>
> This is what I said exactly three times in my mail, and what I have
> depicted in my flow:
>         Server async operation (mandated by the STD)
>         Client back-channel can be sync with for channel (Not mentioned by the STD)
>
>> that again trigger synchronous calls on the fore channel.
>
>
>> You're basically saying
>> that the client needs to guarantee that it can allocate 2 slots before
>> it is allowed to send a layoutget just in case the server needs to
>> recall a layout.
>>
>
> No I am not saying that, please count. Since the Server is not allowed
> sync operation then one slot is enough and the client can do sync lo_commit
> while in recall.
>
>> If, OTOH, the layoutcommit is asynchronous, then there is no
>> serialisation and the back channel thread can happily reply to the
>> layout recall even if there are no free slots in the fore channel.
>>
>
> Sure that will work as well, but not optimally, and for no good reason.
>
> Please go back to my flow with the 3 cases. See how the server never waits
> for anything and will always imitatively reply to the layout_get.
> Since the server is not allowed a sync operation and is mandated by the
> RFC text to not wait, then the client is allowed and can do sync operations
> because it is enough that only one do async.
>
> BTW: If what you are saying is true than there is a bug in the slot code
> because this patch does work, and everything flows past this situation.
> I have a reproducer test that fails 100% of the time without this patch
> and only fails much later at some other place, but not at this deadlock,
> with this patch applied.
>
> Cheers
> Boaz
>

Whether or not your particular server allows it or not is irrelevant.
We're not coding the client to a particular implementation. None of
the other callbacks do synchronous RPC calls, and that's very
intentional.

-- 
Trond Myklebust

Linux NFS client maintainer, PrimaryData

trond.myklebust@xxxxxxxxxxxxxxx
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html




[Index of Archives]     [Linux Filesystem Development]     [Linux USB Development]     [Linux Media Development]     [Video for Linux]     [Linux NILFS]     [Linux Audio Users]     [Yosemite Info]     [Linux SCSI]

  Powered by Linux