[pnfs] pnfs-client-temporary-patch-do-not-replay

[Benny Halevy]

On Oct. 22, 2007, 17:23 +0200, "William A. (Andy) Adamson" <andros at citi.umich.edu> wrote:
> On 10/22/07, Benny Halevy <bhalevy at panasas.com> wrote:
>> Hi Andy,
>>
>> when doing some removes in parallel from 8 clients
>> I'm seeing the following message repeated endlessly:
> 
> 
> are the clients endlessly sending replays of the removes?
> 
> the server does not have a sessions replay cache implemetation, so whatever
> we do is a hack until we have an implementation. what is supposed to happen
> is that the server will note the replay, and return the cached RPC from the
> session cache.
> 
> commit 73953ace1a7472a5adadd56d6f564c5281a41b9c changes the server behaviour
> from one hack to another (!) -  instead of a replayed sequence number on a
> slot  returning NFS4ERR_BAD_SESSION on the OP_SEQUENCE it now returns
> NFS_OK. this in turn lets the server continue processing the compound and
> re-do the RPC instead of replay the cached rpc.
> 
> but the client is not supposed to send retry's of a request..
> perhaps what is happening is that the client is not waiting for a reply
> before using the slot for another request (the replay???)
> 
> do you have more information on the clients behaviour?

No, I reverted the change for now in my tree
but I can try to reproduce it sometimes later.
It seemed returning nfs_ok caused one of the clients to go out of sync with
the server as the seqid just kept increasing:
Oct 22 04:43:45 perf-x31 kernel: [  348.103062] check_slot_seqid error. seqid 28 slot->sl_seqid 28
Oct 22 04:43:58 perf-x31 kernel: [  361.409020] check_slot_seqid error. seqid 29 slot->sl_seqid 29
Oct 22 04:44:12 perf-x31 kernel: [  374.714980] check_slot_seqid error. seqid 30 slot->sl_seqid 30
...


> 
> 
> -->Andy
> 
> 
> check_slot_seqid error. seqid N slot->sl_seqid N
>> This message comes from commit 73953ace1a7472a5adadd56d6f564c5281a41b9c
>>
>> Please advise,
>>
>> Benny
>>
>>
>